Listen to the latest news, expert opinions and analyses on the ever-expanding world of artificial intelligence, data science and machine learning, broadcast on leading talk radio shows and premium podcasts.
The Machine-Human Interface with Alex Newman, CEO at Human Protocol
"We were mentioning this kind of to set a marketplace and the protocol. let's say. I think it's fair to say that's infrastructure is providing infrastructure for this and. I've seen that. You guys are also working on decentralized labor market you want to expand upon that a little bit technically h gap jesuit decentralized labor market. Right like these are all It's it's a little bit interesting. But but yeah. The human protocol can describe any interaction online between a human and a computer. We try to be very careful not to talk about future applications too much. We we talk about what we're excited about And so one of the things that i'm excited in that space a lot of people talking about labelling off machine learning either reinforcement or supervise learning. But they're just a lot of actions that human beings have to take in this ritualized form. It has to be documented online on that extended throughout our whole life. Everything from how banking records were. You know. I don't want to once again. I want to get too much in future applications. But we're pretty much everything in the there are so many times where humans are interacting with robots in this ritual listrik or specific way but capture is a very interesting one So when we first started this project it was actually pre the capture. We had organized. A all of the major work was online to solve some of the largest data. Labeling problems that the that our big customers needed and so we ended up making a poor man's capture just because existing labor online they They don't they they. They don't necessarily they're not perfectly trusted. Not all of them have the same training and so what we would do is we would send the same question out to multiple labor pools and really can sense. The answer across that net interest of consensus method is one of the key parts of the capture product. But we learned while using catches dataset. It has significantly less different. Ias's then trained users. So what does that mean so one of the examples that i bring up to people He discovered the hard way. Is what happens when you ask people whether or not something is a long sleeve shirt while it turns out what country you ask them from is going to determine whether or not that sure is short medium length or long line which i find to be incredibly fascinating and it turns out that most of these questions which we believe have surefire answers really only have probabilities and probabilities are social capture. Can give you a light onto those that you really can't get from a trained staff. In addition all the stuff around Any biases they might have due to their training or previous experiences. That all goes away. It's like asking a question across the whole world in addiction. Unlike some other capture products we really are worldwide. You know we work in china. We you know we we. We work in so many of us. Complete who play puzzles across the world
Have You Heard of The Alexa Ghostbot?
"Here's the deal. I'm just gonna read the headline here. That i found on the internet and the headline says lax. Lexi ghost bond could let relatives speak to their dead loved ones by using artificial intelligence to create realistic replies using their old letters social media posts and videos. So what are the first things that come to mind with that. That is fascinating. Basically the story goes on to say that microsoft has applied for a patent that would basically as they put it in the article digitally reincarnate deceased people and would allow living relatives to interact with loved ones from beyond the grave. And why this is relevant to lexi is that it says that the technology could work with smart speakers such as lexi or google nest. So what do you think of that. Previously interviewed james who's This co founder of hereafter. Who is helping to digitally Record stories told by a person who is alive so that those stories can later be called back wants. That person has deceased but this is different. This is talking about creating an avatar if you will of a deceased person based on stuff on the internet about them so will see we'll see what happens with
Ready to shop with a "smart cart?"
"You pandemic has brought about a ton of tech innovation when it comes to shopping between ordering everything online in curbside pickup. The next phase of this might be the smart cart according to my colleagues at the cincinnati enquirer supermarket chain. Kroger is testing smart cards where shoppers pick their items place them in the car and leave without ever having to go to a checkout or register i this is how they work. The carts themselves use artificial intelligence to recognize what you put in the cart. So you can take a box of cereal. Draw pasta sauce a gallon of milk. Whatever it is the cart will recognize what you place inside. It also has scales built so if you're buying produce it will wait for you in the cart and then once you're done shopping you back up your items and then you make a payment directly from the car itself. This idea of checkout free. Grocery store isn't new. We've seen amazon experiment with this already. Last year. they opened in grocery store in seattle where users can download an amazon go app. They scan a barcode to enter the store. Then they scan what they're buying and once they're done they just walk out and received by email and it's all done within the app. The fact that kroger is testing smart cards is a pretty big deal. The cincinnati-based grocer operates several regional supermarket chains in thirty five states stores. That you may have heard of like fred. meyer harris teeter ralph's marianas and host of others. The company has nearly twenty eight hundred stores so the fact that they are testing this out is pretty significant. Once they get to a final place it's going to be at a lot of different locations so it looks like your grocery experience is going to get a lot more technologically advanced in the months and years to come
"You'll see. Where could i find your face now other than well. That's a good question. Probably everywhere you'll see now is a security engineer of two decades and one of the founders of cyber reason definitely can find it on on facebook you can find it on the lincoln On several newspapers Online publications Physical publications probably to your see isn't a public figure. But he's been around the block enough that if you manage to spell his name right on google you'll find plenty of pictures of his face some of you out. There are in the same predicament. Some of you you to your career or because you're proficient with some of media will have many more pictures online then you'll see does but even for cybersecurity audience. I can't imagine many of you out there. Having no pictures out on the web these days it requires diligence and effort to be that private. Is it possible that i might find your image in places where you don't yet realise it. Is i be surprised. Is it possible that you listener have images unveiled in it. You're not aware of you might not realize what's out there in the database of the gym. You used to go to or the office building. You used to work at or on your old. My space account the day make nelson. Our senior producer ran his and my face through a face. Search engine to test it out. I found shoot around twenty something with this kind of poise haircut. An earring and then there was one shot which was labeled adult content. Which i can only assume was because you're old sideburns. Were so dang sexy. Yeah that was my asimov period but hey it's nice to know that i have an alternative career in the adult industry in case this podcast thing never gets the graham. I don't know ron. There's a reason they put you on radio instead of tv Not for later fire. Make the fact that there are pictures of me. I don't remember in places. I didn't expect doesn't surprise me much and that's important for us story today. This episode is about a problem that arises when we have too many faces in too many places because like any data. Your face isn't something to be carelessly tossed around. There's value to it a market for it when i tell people about family sounds the audio documentaries. We create for families who wished to preserve an important piece of family history. A lot of them. Ask me if those family histories really interesting to listen to after all we're talking about our parents and grandparents not some famous movie stars or scientists
2021 Voice AI Predictions Part 2 with Paquiot, Kibbe, Palmiter-Bajorek, and Kemp - Voicebot Podcast Ep 189
"Have to events coming up. And they're both free. I'll be speaking and so if you have a few minutes you might want to check those out later this month. I have a ten minute presentation. Coming up for the season. Kickoff of voice talks. that's run by modell is sponsored by google. I think a lot of you've watched at least one and some of you may have watched several of those episodes last year. I appear to think three or four times for sure so in that presentation. I'll actually go through my three trends to watch for twenty twenty one. Are they predictions. I'll let you decide. I just revealed two of the trends and voice insider. But i saved one for the big reveal of voice talks. It's really the what i'm most intrigued. And by. Because i think the timing so perfect for this is coming up It'll be next week. Actually so you could register at voice talks today. I i talks dot i even. If you can't make it there will be recording. You can watch afterwards. Sign up there and you'll get the link for viewing or to watch it later. Also i'll be joining my good friends from one. Six nine labs germany for fireside chat on the twenty eight in the morning east coast time west coast a very civilized afternoon slap for europe. So that will be fun. This'll be a live discussion with the folks that bring you the all about voice conference. So tim kala dominic meissner derogate me and maybe i'll ask a few questions of them as well. 'cause i'm interested always in what's going on in germany. What did we learn in twenty twenty. And what should we expect in. Two thousand twenty is really the focus of that twenty. Twenty one's already started off so strong. They should definitely be a good time. You can register for free at all about voice dot. Io all about voice dot io to scroll down and you'll see my smiling face with a large register button beneath it and that'll take you where you need to go okay onto this week's guests first up. We have like patio from k. Four connect we get into her predictions about smart displays health. Care and the need for ai. That serves everyone great conversation. There was really excited to have that a new guest first time guest and voiced by podcast. Then we have kid from samsung's vid labs. He's been a guest before two years ago so almost like a lot of you know roger done some of his own podcasting. But he's best known as evangelists over there at the labs around the bixby voice assistant. Here's some predictions round voice mobile apps. It's actually not fixing specific but his voice in mobile apps he even let slip. What the name of the real killer app for voice what that will be so. That was sort of fun that that got revealed. Following roger is joan. Pomodoro landed his head of research at an lx and his also known as the founder of women in voice. Our discussion starts out as multi-modal and then we build on her prediction abed to pick a word that might capture the guy's voice designed for twenty twenty one and finally we close with mr future ear himself. Dave camp of oaktree products discusses the audio media landscape. How that's evolving. But specifically his prediction related spotify. Move into voice four great yesterday. Wrap up our deep dive of eight industry leaders and their predictions for twenty twenty one could voice and just type in one hundred predictions in the search bar to read. All of the predictions were fifty people in the industry. There's a lot of great stuff in there. I'm really happy. I was able to at least bring you. Eight of those who part one and part two. But today we've got four to wrap this up next up paquito kibi partner dark and kept. Let's get started so like a pack. Yo welcome to the voice. Podcast thank you. I am so happy to be here breath. Why i'm happy to be here to. Why don't you tell the audience who you are what you do. So they have that context before we get into your prediction for two thousand twenty one. Absolutely so i am. Vp product at k. For connect that is a senior living technology startup based in raleigh north carolina. So i am responsible for the strategic thinking for all of the features that go into the products four k for connect got it. Okay do you actually had a big announcement not too long ago. We did so exciting and directly related actually to the topics here so we partnered with in kala that's the north carolina association of assisted living communities and We were able to through our partnership with amazon providing amazon echo shows in all of the assisted living communities in the great state of north carolina. That's awesome I'm i have a lot of to north carolina. My wife is from north carolina greensboro. I lived there for three years myself. i was made an honorary north carolina while i was a resident citizen which was sort of fun I was like why. Am i honorary. I've married someone live here. Okay secretary of state thought. It would be funny to do that for me. I have a nice plaque next but that yeah upgrade. It's one of my favorite one of my favorite Certificates plaques. I've got it all frame. It was pretty fun to get that from the secretary of state but but that's an aside. That's let's get into your prediction so you several different predictions One of which is directly related to this idea of multimodal smart displays. Why don't you start with that one. And then we'll see where that takes us. Yeah absolutely so. A lot of mike. Predictions were framed around the impact of covid on work right and and the impact that is having on the use of of voice a specifically so one of the things that we're seeing in our industry is the increase of multi-modal displaced. And that is. I would say a cousin to telehealth the fact that telehealth became this huge thing in a twenty twenty the though the rains were loosened and folks were on not only allowed to but encouraged to have visits. Right so Jason to that is this use of a multi-modal displays like the amazon echo. Show so one of the things that has plagued voice Is this problem of discovery. And is this problem of Knowing what a skill in the case on alexa or a voice app can do for instance instill what multi-modal displays on do by having both voice and in actual visual display. It helps folks overcome those challenges by blending the two together and so particularly in our industry we serve older adults folks living with disabilities Voice for people with a higher acuity levels on has proven very very useful They're not able to do all of the things that that we can do with an app So voices hugely helpful but that problem of discovery but that problem of knowing what the devices called. How'd you invoke the device remembering the name and so on multi-modal displays Definitely help overcome that. And so it's going to be huge for house as well One of the interesting features that folks have been talking about. Is this ability
Machine Learning Carbon Capture with Diego Saez-Gil
"Welcome to the show thank you. Jeffrey excited to hear your pajama. Pyjama is a company that helps with carbon offsetting. Can you explain. Carbon-offsetting is sure. This is an idea that was invented at the united nations in the kyoto agreement which was a predecessor to the famous paris agreement. And the idea. Is that as we move away from fossil fuels or as a way to incentivize the move away from fossil fuels we need to make looters companies in countries to compensate for the carbon emissions. They're putting in the atmosphere that are causing climate change right so there is a system by which then projects that either reduce emissions or removed governor from the fear can receive carbon credits equivalent to the amount of car on this project. Effectively reducing or removing. And then these companies can purchase those car on create certificates and use them to compensate their emissions and in doing so the benefit. You want side now. They have an economic incentive to do away from posted fields so that they don't have to be spending money on compensating for those emissions and secondly that money goes to fund very important projects such us renewable energy and lowercase forest restoration conservation as you know forest remove carbon from dan lewis fear as trees grow. And you know thanks to framework if you have a piece of land in which you can plant trees or conserve an existing forests that otherwise would be cut down. You can get paid current credits to continue doing that work. So that is a framework and these frankly has existed for many years but unfortunately until today there wasn't a lot of technology a the software powering the certification and exchange of the concrete industrial. We decided to focus. On what kind of technology could be useful in the workflow of managing carbon offsets. Yes so the first part is violating monitoring. How much carbon is being sequester by forest into that we use remote sensing data in machine learning algorithms that basically aim to predict. How much carbon is there on a forest in particular for example one of the models that we built train on a combination of satellite data coming from nasa lanza data with lied are they coming from companies that collect that data from airplanes. And it's basically at three dimensional cloud of points that gives you the shape of the structure of the forest and then we have ground truth. That is coming from four services from around the world that send people to the field to count trees and sure trace with dates and know exactly how much carbon each of the streets have you know so then we run deep learning algorithms convolution new neural networks to train on the day dan the nba to predict how much car when they're in a forest just using satellite. That is one of the parts of the stock that were building being able to replicate the tests that are needed to validate how much carbon is being sequestered by a forest.
2021 AI Market Predictions
"So if you've been listening to a today podcast for awhile. welcome back. We really appreciate all of our fantastic listeners. But if you're a new to the podcast. This is your first episode. We like you to know that. There's hundreds of episodes that we've been producing over the last four years on with the have everything from great interviews with a i thought leaders and insights into the market trends and adoption in public and private sectors. And actually will be doing one of those insights into the mayor market trends on this podcast episode but also conversations on key topics on what's happening with a today and in the future so over our past for years almost two hundred episodes we've interviewed some incredible influencers. So we encourage you to go back and listen to a lot of these episodes. We have episodes interviewing folks. Ben kurzweil of singularity net and the sofia robot colin angle from founder viral anthony griffin. Yano from dun and bradstreet eager. Perry switch from lincoln. Suzanne can't the former us federal cio. The hose arrietta ceo former cio of the us department of health and human services. Lord tim clement. Jones keep people at organizations large and small and lots more so Definitely subscribe to the today podcast so that you can basically here are insights on the technology markets and how different industries are applying emerging concepts machine learning. And just in general long story short if you want to understand how. Ai is being put into practice today. Which is why this is called a today and where it's heading. Make sure to subscribe day today. On your favorite podcast provider and listen to our hundreds of episodes. Yes so as ron mentioned today we wanted to spend some time talking about our twenty twenty. One a. i. Market predictions and forecasts at the beginning of every year. We always you know. Take a step back and look at what happens over the past year and where we things going moving forward so acog melinda in case this is your podcast or you're just starting to listen to us. We're an ai. Focused research education and advisory firm and we really focus on market intelligence. We cover all over twenty thousand vendors in the space so we have a great pulse of what's going on and we work with both public and private sector companies so we really have a holistic view of the space so we wanted to spend some time today reflecting back on what we're seeing in the market and then making some predictions and forecasts about where the market will go in twenty twenty one so one of the first predictions that we have. These are not in any sort of ranking order. They're just how he laid out this podcast. So we have that worldwide adoption of artificial intelligence and machine learning. We've seen it growing at a very high rate and were predicting that this is not going to stop anytime soon. I mean so. There's a lot of indications that show that we are moving towards much more use of what we call the seven patterns of ai and we will link to them in the show notes but one of the things about is that it is a fairly generic term general term which corresponds to making machines intelligent and doing the things that humans would otherwise. Do you ask people as to what they're specifically doing. It's usually gonna be one or more of these seven pattern so it might be a recognition system or it could be a conversational system or could be something doing predictive analytics or trying to find patterns or anomalies or it could be trying to develop the hyper personal profile. The hyper personalization profile of you. So that it can no to tailor things better for your needs or it could be an autonomous system systems that are meant to operate with little or no human interaction. Or perhaps we're doing something we're trying to have. Machines find the solution to something you goal driven systems and when you talk about it from that perspective it's like yeah chat bots are growing recognition. Systems are growing the use of machine learning for patterns and anomaly detection as well as predictive analytics. that's growing. You know maybe hyper personalization. Maybe that that's been a little bit slower to grow. We are definitely seeing a lot. More autonomous stuff whether or not. They're all entirely successful a whole other story. But we are and we're seeing of course a lot more use of even goal driven systems and part of the reason why we say this is that there is some fud in the market Other analyst firms in particular are saying that they're seeing some large number of data science projects that are failing. You know gardner. Says eighty seven percent of data. Science projects failed to deliver on their for their executive sponsors and seventy percent of machine. Learning models lose relevancy overtime. Well these are. There is some truth to that. Yes models do have what's called drift and then later what we're going to talk about in this. Podcast is the growth of technology area technology market with an ai called l. Ops that specifically addresses this area of models overtime lose their relevancy. But that's just like the thing let's like saying well. I built an app in one thousand nine hundred ninety six therefore i need to update it in the year. Two thousand three two thousand eighteen thousand thirteen two thousand eighteen. Yeah yes. that's what. Technology and technology doesn't standstill. Say all the fact that you have to update it means. It's not like the fact that you have to up it means you're actually using it and the needs for that. Continue to grow. If you didn't care you just throw it away so
2021 Voice AI Predictions with Thadani, Tingiris, Stapleton, and Fields
"Welcome back to the voice about podcast. Worse okay so we're gonna get your prediction is why don't you just tell who rain agency and what you do. Short so ran is an agency that sales voice assistant experiences for large companies. We work with companies like nike nestle. Flat rock in the win and we really help them design and build conversational experiences. That are going to drive business. Value so You know some our collaborations in work have been with google on. Abc mouse and we also recently built a experience for the win hotel. Las vegas where we implemented a voice enabled concierge in their hotel rooms Slept on the work that we do. That's awesome well. I shouldn't have to talk about some of these projects you've been doing so we'll figure out an appropriate time for that. Because i'm i'm interested in what's going on with the win. They were an early adherent to voice. I know they've continued to invest. So that's a good example but today's focus is on what we should expect to the voice. Ai industry over the next year and to summarize you're looking at the rise of owned some people call them cost them some people call them branded voice assistance. Why don't you describe what that is to you and why you think that that's going to come to pass. Yes so You know. I think we've all been seeing. Rain has been very bullish on this space of own voice assistance. So these are you. Can think of america's erica or walmer. Ask sam so rather than relying on the army. Assistant landforms forms like alexa. Google assistant like some of these brands have in the past. They're building their own ranted. Ai assistance That sit on their own channels. They more control over the user experience in the data that comes out of those experiences and the reason that we've been particularly bullish on this in our twenty twenty one. Prediction is just based on a lot of the antitrust regulation conversation. That's been happening over the past year. in in our view at rain there has been unprecedented. Scrutiny of tat as a whole. It's really the first time that you know. Government leaders across both sides of the aisle are taking a vested interest in antitrust topics. Actually trying to understand them and understand that they're asking. There's a lot of reason to believe that they tech companies may not look like they do right now in twelve to eighteen months in q. Add on top of that. What's been happening just past week with twitter and amazon apple after some recent events is just more ammunition that will fuel this conversation around breaking big tech which will certainly impact voice assistance. Kind of how. How wall yeah. I think that's interesting. I was just talking to someone earlier today about this. They brought something up. And i said you know. There's a lot of companies out. There have been focused on different things over the last decade. And i think it's it it might be that a company like amazon which is really just focused customer. I you know they have their strategy customer. I you know that's been a twenty five year run of that focus. They might actually start looking more legal and regulatory focused over the next decade. And i take a lot of rash or maybe justifiably so to say we have to do that. In order to keep serving our customer. That is the way we best serve our customer. But i think that's really interesting thing but like not to go down that rabbit hole too much. Because i'm sure there's plenty of opportunities to do that. You're happy to comment on that. But this idea of assistance has been something that we've been talking about On voice for awhile. I know you've been talking about it. I think there's a sense that Doing this is too hard or that. The investment is really to significant that. If you're only only amazon or google can actually get away with this I don't think you agree with that. Yeah i think there are a lot of different ways to do this right. There are ways to do it Round up a lot of investment. Probably very hard as we've talked it out in the past like thank america did with erica and i have talked about the fact that we consider that to be successful. A lot went into that success rate but there are other companies that are doing it in very different ways. In still leveraging. I think solutions are kind of being patched together because today there are a lot of really interesting tool providers out there raza pico to fi- you could very well used pieces of these different assistance to at least start to test and understand when owned voice. Assistant would look like before. Making larger investment in rain has been kind of in the gamut of those types of experiences. So everything from you know. How do you do this from the ground up to starting to piece together. Some of those providers to It'll be solutions.
Mercedes reveals giant 56-inch, AI-enabled 'Hyperscreen' for upcoming electric car
"Listeners. It's brett molina and welcome back to talking tech. My co host. Mike cider is off today. Recently mercedes-benz announced plans for a fifty six inch curved screen that nearly spans the width of a new cars interior. That just sounds incredible. Saying it out loud. The hyper screen will debut the spring. And it's new electric sedan the mercedes benz s. It's part of a growing trend among automakers to make your car's function more like computers but are all these screens and other tech safe joining us now is money reporter and i believe green bay packers. Fan nathan bomi. Thanks for joining us brett. How are you. I'm good so you ask the question in your story about this topic and whether bigger is actually better what did you find out well. Fifty six screen is john. I mean we've never seen a screen like this in a vehicle and you know it depends on who you ask. Whether it's safe or not the mercedes benz says it safe and And safety advocates are concerned. I mean you know. The first company that really started to pioneer big screens and vehicles was tesla model s a luxury sedan and then later with the model three and other vehicles and they basically put a tablet screen at the center console in place of a lot of the old fashioned analog buttons but mercedes is taking this to a new level by basically extending the screen across the vehicle screen even now encompasses the instrument cluster. So this is a really a step toward bigger and bigger. And i think we're going to continue to see. Other automakers follow suit so it's weird hearing all these announcements about these giant screens and cars considering how much we've been told about smartphones in the dangers of having those and using those waller driving. How do automakers plan to tackle all these additional screens and cars. Well the most important thing is to keep people's eyes on the road that's really the challenge and so if there's a screening vehicle but you keep your eyes on the road it should be safe for the most part the problem of course with smartphones is oftentimes people are fumbling for them with their hands and they take their eyes off the road and then he ended up getting into an accident driver. Distraction is really serious crisis for this country and so we obviously want to avoid that here. Now when i asked the ceo of daimler on a conference call a couple of weeks ago how they incorporated safety concerns into the design of the hyper screen. He said that they actually designed it to ensure that people don't have to take the eyes off the road for example if ape. The passenger is looking at the screen. Has some sort of video plane on the screen In in front of them essentially in the front seat and the driver takes their eyes off the road and looks at the screen. The vehicle be able to recognize that. I've movement and stopped the film from airing on the front passenger seat Screen which is wild and then if you put your is back on the road than it keeps playing. This is some really high tech stuff. And you know we'll see if it works now at my understanding is that particular technology will not be in use in the us because of safety concerns here but it will be in use and other countries. I wouldn't be surprised if the regulations eventually change here to allow it so with all these fisted systems coming in cars. How long before this starts to change how we learn how to drive. I think it will you know. I think we're talking about things like augmented reality windshields being the next step. Where essentially you'll basically have a directions projected onto the screen that we if you've seen a new vehicle driven a new vehicle anytime recently you may have experienced what's called heads up display which is where. There's a tiny in a projected box. Bright above your instrument cluster on the windshield. That essentially will show you for example what speed. You're traveling but in reality windshield would project like an arrow onto the road head of you to say turn this way or that way and so. That's the next
Data and AI in the state of North Dakota, Interview with Dorman Bazzell, CDO of North Dakota
"Today with us. Our guest is dorman basell. Who is the chief data officer for the state of north dakota so high doormen and thanks for joining us today. Kathleen ron thank you for the opportunity. Either to hang out with you guys for a little bit. Yeah we'd like to start by having you introduce yourself to our listeners. Tell them a little bit about your background. And your current role as the chief data officer for the state of north dakota. Sure sure well. Good morning everyone So my my background is You know went to college. Got a degree in computer science mathematics and then when often like everyone else When i lived in saint louis you. It was kind of a requirement. You had to work for mcdonnell. Douglas which is now boeing corporation. So did that for it. But but then after a while Got got involved in consulting and worked my way up through the Consulting ranked says developer and they as a project manager is the data architect the solution architect and then finally got into a position of driving business intelligence and analytics for a couple of large international consulting firms where ran their north america. Big data and the i practice And the great ride. A thoroughly enjoyed all of the things we did. I think we added a lot of value to Our customers which was private industry And had great teams Had a strong onshore team strong offshore kimes and delivered a lot of value. But i think two years ago Over two years ago. When i applied for this position as the chief data officer At first i really didn't want position Didn't like the idea of state government state government has has a bad connotation Of kind of a nine to five job And a people people who just weren't really motivated to To move the world change the world and my boss who i interviewed my off. Now the cio. Sean reilly Who i interviewed with his his final comment to me was well. I can't pay what you make today. But are you wanna paycheck or do want to change the world. And i had never thought about life that way. Never tried to change the world and So i decided to take on this opportunity This was the first chief data officer position for the state of north dakota so there were a lot of unknowns Certainly certainly my presence Was a bit chaotic for the organization. Because i came in with a completely different agenda and completely different way of looking at the world through the eyes of the pillars that are assigned a line to me which application development and automation. And the second pillar is data analytics data science artificial intelligence and had some very different opinions about those things. And how we might move those forward So as i became involved with this role i became an. I had made an assumption that every state had a chief data officer come to find out there are only twenty seven of us out of fifty states So it's it's an interesting It's an interesting mix of of individuals who are chief data officers and getting to know them is. It has been a really amazing opportunity because they have such a very backgrounds and they bring such such different perspectives to cheap date officer role I like to joke and tell people that the last thing i focus on data which is obviously not true but but my real focus is really around cultural change within the city physician and what that means in the context of not not necessarily data. Because i have to executives are on my team who Are just are just brilliant at running the operations and managing the two pillars within my organization.
India 2020 Voice AI Year in Review with Haptik, Slang Labs, Klove Chef, and Women in Voice
"Aggravates. Welcome to the voice spot. Podcast thanks for having me. I'm excited i'm excited to so the topic. Today is the year in review and we really focus on what's going on in india. You have a very interesting. Let's say a front row seat to what's going on with conversational a more. Broadly you deal with chat and with voice as you look back on twenty twenty and you look back in india and the conversational space. What was the biggest story of the year. What stands out. I think the biggest of the if you look at everything in addition i Chad boys todd bobby platforms use cases gov support if you just look at the broader space end for us like you mentioned right. Rear look on the internet of things over. The biggest breakout story has been whatsapp By that what What i mean is that if you think about the decio. Facebook has genuinely taken a number of steps. do enable what's up to become a platform for businesses to do a lot more with it right the genuinely thinking about this was the first yard where they would enough actions that facebook is accompanied to basically say that look deadened mission with what chapters to become like. Ob chat by web businesses and branch can come on and a conversation experiences to enable commerce engagement. Ben support So far for for us. And for me specifically as i think about the landscape i actually you know. I want to go to the extent and say as far as i think in domes of build adoption of conversation in the seven yards that i've been doing this which is like a lifetime in this business. this is probably the biggest route moment that that's happened because of the simple fact that You know if you if you provide in platform read a hundreds of millions of users can genuinely use. Nlp conversational interfaces dude to businesses to get things done for commerce far support it it probably will be can be a genuine battered. I'm shift that. You know people like you and i have been talking about for many rs. Okay so you work with a lot of big brands. Yeah so when you look at when you look at what's app or these brands interested in what's happ before the new features came along. Yes so there is always been a great Interest from start dating back to two thousand seventeen on more street because of the simple logic that you know most of these bands and more of the bland managers or the c. Suite dr day us. What's up like five hundred times a day for everything right. so then i'd choose expansion. What product on my band on this bag form so the interest is always been dead. It's been it's been the case for like i said the red bottle to your four yards. You know it was just a matter of the features being available the platforming available and more broadly in honesty the features started begging big. Started becoming available about a couple of ers by but really your until the earlier question is whether it was the first show. Facebook obliquely came out multiple times in. I talk about the fact that whatsapp is going to be a platform for engagement with businesses. Right unlimited a lot of things down features actions announcements which gave ams especially large brands a lot of confidence that look this is a good platform for businesses to stay and not just something that was been for like monetization got and so we've seen other moves along these lines as well with google also have business messaging feature. What is the general reaction about those. Are those going to be supported as tier two. Are they going to be supported as enthusiastically. Or is it going to be more Something that if if i have to do it i'm going to do it. But what's abbas where i want to invest. I mean i think you do have a but not like a must have sort of need. I mean it's the classic Vitamin was painkiller sort of analogy read. whatsapp is like is like dope. Incommoded solids of very big problem in terms of engagement. Outreach was Assured the good to have that a check box i need to be on those channels honestly similar a little bit with alexa and google has read what we've seen play out over the last two yards. Which is yeah great. I'd love to. I'd love to be on alexa of beyond will so that we can check it off that a band present debt. But it's not. It's not solving the problem of engagement.
Consecutive Votes in Paxos
"My name is. I'm currently a phd candidate at the university of michigan. My work has been mainly in distributed consensus and has been now shifting into applying formal verification two systems and building systems correctly specifically with a lens on distributed systems. I'm looking forward to talking both about paxos and formal verification. Because i think they're both interesting topics and i was unaware of much intersection. But maybe the kickoff. We've talked about packs on the show. Before so i would refer listeners. What deeper details to go back there. Would you mind giving people a high level on it. what is paxos. So paxos is a distributed consensus protocol. That's goal is to allow a group of computers or replicas to be able to reach consensus on a series of values or commands which will allow these computers or replicas consistent with one another and there's a lot of variants. Why do we have so many flavors of paxos. It's a good question. One reason is that it's an interesting problem that draws the attention of researchers and there's room for expansion as well as this problem that paxos aims to solve is by no means in its final state. And there's always more challenges and more problems to solve so many of the different flavors variants of taxes that we see being published year after year are aiming to improve upon existing solutions or solve new problems that come up with expanding technologies well. Let's talk a little bit about your specific. Contributions paxos is an old algorithm. What was left to be done. Yes so that's sort of a fund bit with this paper. The significance on consecutive ballots in paxos wall. It doesn't currently have that great of a practical implication for these families of paxos protocols the interesting aspect in my opinion is that even though this protocol has been around and studied immensely in the past twenty years there are still parts of it that weren't maybe discovered or published realized. And that's sort of what this paper is trying to highlight. Is that the original description of taxes in this sort of carries on too many of the different flavors as we mentioned just previously carries the assumption that the original and variants which are the promises or glue that hold the protocol together and provide. The guarantees of the protocol are the most simple or ease. Are the building blocks that are required to make paxos work while in fact we can point out that. That's not actually the case. And we can show that what those in various are are more conservative than they need to be and we can show that they are indeed can be weakened or expanded while still maintaining the same guarantees paxos originally set out to guarantee so that part was surprising to me. Usually when i see some improvement it's always a trade-off get something but i have to give something is it truly the case that we're not giving anything up with the shift in this specific case you're not giving anything up in a practical setting. Your may not be gaining much either. It's more of just a insight as to our understanding about how this protocol works. Paxos has throughout the years been noted to its subtlety which is why often it becomes. Hard to implement and developers may choose to rely on using other consensus protocols as opposed paxos because of the subtlety. So you're not necessarily giving anything up but you can certainly expand upon what you have without any consequences. Well let's get into the notion of the consensus khoram's that i think plays a key role in your work. What is this courtroom. And how is it useful so at its core. Paxos will always provide safety which means that. It will always provide consistency. While if it's not able to do that it won't be able to make progress at all and the way it's able to always ensure that it's able to give you a correct answer and remain safe relies on the fact that you always will have majority quorums of participants intersecting so every step taken paxos would typically require a majority of the participants computers or servers in this case to agree or send messages to continue with the protocol and this just ensures that if a single computer replica in our system has learned the value that at least one of those will participate in any future steps protocol takes because by having a majority quorum at least one of the past participants will be able to voice its current state in the current step of the protocol so by relying on the fact that paxos uses majority quorums. We're always able to guarantee safety. Because paxos says at its core to paraphrase says only a single value can be learned so it paxos if after executing the protocol all servers can agree upon one valued. So if you had for one case a value can only be learned assuming that a majority of our servers have agreed upon that value so it could be the case that some of our participants of the participating servers have not learned value. But if they try to take any future step in the protocol that would require a majority quorum of all of the existing server participants at least one of those in the majority quorum would be guaranteed to have learned or agreed upon that previous value. Which would bring the current server up to speed and it would be able to guarantee that any future steps maintains that single value.
Who's Adopting AI for Chat and Customer Support? - with Abinash Tripathy of Helpshift
"So nash. I want to catch up with you here on the topic of who is adopting when it comes to artificial intelligence for chat. That's your world. It's wild you've been in there for what nine years now. Two thousand eleven or something like that. So you've seen a lot evolve you know when you think about who is really picking up on this who's starting to layer on top of their their serve chat conversations in a fruitful way. Who who do you see those early adopters or anything. They have in common with her industry size. Whatever a very good question. So i if you think about automation the effort in trying to get automation going in an enterprise. The biggest auto identity comes from people that are having high volumes of interactions. So you say when you find that as high volume deductions that are specific. Industries were customer service has very high volume of interactions and it's difficult visible in two ways. One is just they just have high volume second. Is the number of agency employees in contact center right. So if you think of industries that basically have the highest number of agence dealing with customer service. Those happen to be beat. Ac- industries anything that has to do with you know retail banking utilities telecom gaming which is where we got our start right because think about gaming for the second we gotta start in gaming because they are like any of the largest social media networks in the world they have hundreds of millions of people blaming online every month right so the monthly active users in gaming environment is massive. It's as biggest telecom and the volume is proportional to play the game. But i really think about the game. People wonder why gaming such big customer service. 'cause great games are just like casinos or retail outlets people play a game but to the game is a store and the store is usually selling stuff right that people buy an anytime people buy stuff that are all kinds of issues that can happen. The game itself may have issues. That are game blaze. She was that a fraud issues. This bullying issues all kinds of seeing the gaming world so we are starting gaming simply because gaming is a very digitally focused digital first world for them. The phone is a foreign concept. They don't want to have a phone based on people calling them on. The phones added costs so much money to serve a phone call so they want everything to be digital first until we dominate the gaming industry. Whether it is the largest games in the world that i can rattle off my mouth all of them of our help shifts right so that's where we got her start and now what we're seeing is. Lauded the bbc industries like to be dealt hospitality travel telecom deck banking at all. You know they have the dynamics but some of don those tend to be very traditional banking and banking does not have a digital first approach. Though i would argue both covid. Everybody's evaluating take time. What what their strategy is. Because you know if you noticed in observed when kobe what really happened. I wanted to lead this to you. Dan is that most of these. Contact centers. Operate with agents in manila manila down. And these workers that work in contact centers don't have internet cut home. Didn't even have computers. They had to come through physical facilities in manila Phone calls they couldn't even be online right so phone. Contact centers were shown hard immediately following all the lockdowns around the world. Right so my bank. For example chase dick would be when there was no way they could serve any of my phone calls right Goes if you had to do banking good luck that idea to wait for many days. Get through the phone contacts so everybody. Every large to a vertical is starting to think about how do better leverage digital and specifically in digital self service capabilities and so if you now that down subsurface capabilities broken up in gulick automation technology in lots or informational self service. So that's what i think. It's what we're doing is so interesting. Got it so just to touch on where you started. This kind of explanation was that you began in gaming because it's so digitally native and it's so averse to the phone right if you're a seventeen year old or a twenty seven year old in you're in the middle of a game. Yeah you're you're not gonna call a company you know that's something your mom would do. You're just going to chat or you're gonna put in an email ticket or whatever the case be so. That's a digitally native space. You mentioned retail. Do you see sort of basically for for retailers in terms of adopting a and chat from your perception looking out of the market whereas their trash and where whereas they're real buyers for this stuff is just a function of how big the retailer is or is it a function of both how big they are and their relative amount of revenue from e commerce. Because i would imagine that that would be indicative of their digital savvy on some level what are the factors that make a retailer more hot to trot more likely to actually be doing the kind of work that you guys do with other vendors or whatever i think do dynamics blame the bill segment. One is look at the gross margins of redel even e commerce businesses like amazon right. The gross margins are single digit vegetables single digit percent. Maybe even one percent right and with those kinds of gross margins yearly want the most effective customer service model right not that you want the cheapest most effective customer service model. And so it's it's there's a lotta margin pressure in the retail business. That's driving the shift digital but more importantly post covid. What has happened is if you look at the large format retailers like walmart for example for the first time what they've seen are completely changed sort of forced behavioral change in in the shopping process. Right so walmart now if you look at what's going on it's all like curbside delivery and cook's at pickup right so you can you can order ahead to a digital service like apple. A website like walmart dot com or target dot com and you can then scheduled to go pick up the products in the store during the walk into the store and walking the aisles. Thanks to the holiday shopping models changed. I would say that whole delivery. Big up curbside pickup. Even the food industry were observing this right. So all the restaurants like whether it's mcdonald's or at all like doubling down on the on the apps for ordering because it's all about pickup and got means you have to be able to order you don't want people calling the call center order. You would rather have a digital app where they can do that and so a lot of things can go wrong in that you know that whole process you order your orders delay and you need to talk to somebody deal time. messaging Becomes the perfect way to solve that and having people call a call center. Wait on the phone for twenty minutes back agent to those. The dynamics are mostly margin pressure and this changed sort of behavior of you know orbiting big our pickups or
Artificial Intelligence and Digital Transformation at Rolls-Royce
"Welcome to the cyber-security weekly podcast. I'm jay leno podcasting from singapore today and today we are very privileged to have dr becky bengal who is the president of south east asia pacific and south korea at rose to join us in the podcast. He repeats sharing with us. The work in a digital transformation and the recent ai breakthrough in ethics and trustworthiness at ross writes thank. You thought the bengals for joining us in the podcast today to be here for many of our listeners. dr bandou rose. Rice name has a long lasting romance and history going back to the first car built more than one hundred years ago but the motos business was separated out some time ago in nine thousand nine hundred seventy three. I believe and rose rises. Now in the business of pioneering the power that matters so tell us more about the journey business. That rose is in today and your role as the president of south east asia pacific and south korea at rose rice. Well has been rooted in engineering since we established in eighteen. Eighty four and this expertise has evolved the business to become one of the world's leading industrial technology company that we today and as the president for the region covering southeast asia pacific and south korea. I'm responsible for the regional strategy our external relations and governance of all our operations across the three businesses that we have civil aerospace is one of them manufactures of ever engines for large commercial aircraft on our regional jets and business aviation and we have decades of engineering expertise to take us through life through life. Service and support solutions for customers in the defense were market-leading aero-engines for military transferring control labor's including combat helicopter applications. I'm needles and power systems. Where leading provider of high speed reciprocating engines providing complete propulsion systems distributed energy solutions. So you can say that we. We have a diverse but volume that includes civil defense and power system and it is because w that our activities have tremendous impact on the world today and tomorrow we have always pursued clean safe and competitive solutions and we believe our technology will be fundamental in helping society transition to the low carbon future. And we're not going to do this on our own. We're going to do this. In partnerships and global partnerships to collaborate and co create solutions and with the regional hub. That we have here in singapore. We've developed collaborations with government agencies untucked -demia like a star and anti eu and us to pursue advanced research and technology in daytime smart manufacturing electrical systems. You touch on engine. That paolo many other products across the road strikes businesses. And i believe including aircraft of course and i imagine that you have been collecting analyzing the performance data of your engines for that case and in fact i see from one of your rolls royce presentation that you have been collecting data for some seventy trillion data points across twenty-sixth dimensions on your engines. So i think our listeners will be interested to know how you been. Harnessing that power data to make sense of this of information and into insight and action. And i believe in many ways is supplying the data to a machine learning throughout the life cycle of the engine from the initial stage of designed to manufacturing to maintenance repair. Overhaul that's right so we we've been applying data analytics for more than thirty years and using ai. With our real time engine. Health monitoring system but service w. lunch to back in nineteen ninety nine and our ai. Capabilities are deeply embedded into products and services so they aren't visible And not widely. Now we're able to monitor six thousand to eight thousand flights every day which is equivalent to monitoring three thousand engines in the sky at any one time so we have multiple sensors on board that continuously relay inflammation with were able to analyze five million data promises from our engines every day and we used to provide insights to our engineers for future development and services that we provide for our customers. But it's not just about the asian and the behavioral for engines. Current work includes applying a with a dedicated team that we have inside rolls royce school the day to labs to improve the risk management in supply chains predict market demand improved the efficiency of our operations and more recently nepal systems. Father of the business. We've been applying a on microbes making our industrial powered technology more reliable and sustainable and in the future we see a. I will continue to evolve. Play a bigger role especially as we saw increasing use of cloud based services which will be governed by data ethics framework and this becomes really essential and today more than two hundred projects that are starting to apply more and more of Framework so
AI draws dog-walking baby radish in a tutu
"So. What was the prompt on the radish walking the dog an illustration of baby dicon radish in a two two walking a dog. I mean that's pretty good and then from literal lines. What's what's remarkable about. This is these are three completely unrelated. Concepts tycoon radish to walking a dog and it was able to synthesize those into an image. That's impressive. I guess
New GPT-3 Models Can Generate Images From Text
"Open a has introduced two new. gpt three models. The first is called clip. Which can classify images and categories from arbitrary text. The second is more interesting. It's doll e which can generate images entirely from snippets of text coding. The it technology review for all gpt threes flair. Its output can feel untethered from reality as if it doesn't know what it's talking about that's because it doesn't by grounding text and images researchers at open a and elsewhere are trying to give language models a better grasp of the everyday concepts that humans used to make sense of things dolly and clip. Come at this problem from different directions. At first glance clip contrasted language image. Pre-training is yet another image recognition system. Except that it has learned to recognize images not from labeled examples in curated data sets most existing models do but from images and their captions from the internet. It learns what's in an image from a description rather than a one word labels such as cat or banana clip is trained by getting it to predict which caption from a random selection of thirty. Two thousand captions is the correct one for a given image to work this out. Clip learns to link a wide variety of objects with their names and the words that describe them this then let's it identify objects in images outside. It's training set. Most image recognition systems are trained to identify certain types of objects such as faces in surveillance videos or buildings in satellite images. Like gpt three clip can generalize across tasks without additional training. It is also less likely than other state of the art image recognition models to be led astray by adversarial examples which have been subtly altered in ways that typically confused algorithms even though humans might not notice a difference instead of recognizing images dolly which i'm guessing is a wally slash dolly pun draws them. This model is a smaller version of. Gpt three that has also been trained on text image pairs taken from the internet. Given a short natural language captions such as a painting of a capybaras sitting in a field at sunrise or a cross section view of a walnut dali generates lots of images that match it dozens of capybaras of all shapes and sizes in front of orange and yellow backgrounds. Row after row of walnuts though. Not all of them in cross section and quote. The results are apparently striking. Maybe not as striking as when gt. Three i got everyone's attention a few months ago in the show notes. There's a link to the open. Ai blog where they show you examples of what this ai can achieve. You can also use the tool. Apparently to generate your own images. Sam altman had it draw an illustration of baby shark in a wizard hat wielding a blue light. Saber and it did it though. Apparently the tool was neutered. Just a bit. So people couldn't produce porn with it still as daniel rack tweeted quote. People think is coming for truck drivers first boy. Do i have news for you and quote and as a laser. You'd hausky tweeted quote. Consider this your notice. If you're a manga artist you have and years left before you're out of a job. I wish that i had any grasp whatsoever of how to relate to announcements like these my initial sense is an equals to wisely adjusted upwards to actually after the end of the world and quote
Machine Learning Lifecycle 2021 Conference Preview
"In today's episode though we wanted to highlight our upcoming machine learning life cycle event. That's taking place. January twenty sixth through twenty eight. Twenty twenty one. I know that some of our podcast listeners signed up for our data for a conference that took place in september of twenty twenty and that was a really awesome event. we had hundreds of sessions. And you know lots of live sessions. Many more on demand with some incredible speakers at that event so just like with our data for a icon machine. Learning life cycle event is going to have that as well. We're going to have incredible. Keynotes incredible presenters and sessions as well really focused on that whole learning life cycle including m l. operations building models model management. So we'll have key five. He is for our sessions and our content for the upcoming machine learning life cycle event. If you like to check out that event we encourage you to go to life cycle. Comp dot com so m. l. life cycle co nf dot com. You can sign up as always. Our events are free to attend but so we wanted to spend you know a little bit of time going into the key. Main areas of focus for the sessions. And then some highlight. Some of the presenters sessions as well so the first area of focus is going to be on machine learning model development sessions in this will include and be focused on building and developing models so for supervised learning unsupervised learning reinforcement learning and then also across all different kinds of algorithms. So we talk about the three different types of learning and a lot of our podcast but just to go over that supervised learning through example unsupervised is learning through discovery and then reinforcement learning is learning through trial and error so that that area is going to focus really on model development for for that That all of that stuff. The next area of focus is going to be the machine. Learning model management so sessions in this category will focus on how companies should go about managing their models once. They're actually in production version eating life cycle management and then also aspects of that human side of management as. Well you know we don't want to forget at the end of the day. There are humans that are overseeing and managing this. So let's not forget about that aspect as well related to this idea of emo model management is this emerging space called ops mo operations which does overlap a bit and he talks about dealing with model life cycle in terms of The version being an iteration of models. And that's our of stuff but mls also keep an eye on the model as it's in production so things like model drift and data drift so these sessions on mo ops. I talk about what. Emma lops is how it relates to a related area which is taking a development code in the operational aspects of managing the development life cycle. There's a lot of things in common with that. There's a lot of different things though. Dealing with data drift a model drift and aspects of modernisation and models security and all sorts of things. So people definitely have sessions on. Emily ops quite a few and some emerging technology vendors that are expanding in the space. You're gonna wanna hear from them. Also we're to be looking at model governance so machine learning model governance. This is really focused on aspects of the the management of models in terms of access and control and decision making and who gets access to the various models and maybe differences in version in terms of. Maybe wanna keep multiple versions of a model around for different purposes. Maybe models trained in different data sets for different things you know. How do you handle. That can become very complicated. And you have to. Of course deal security and privacy and all these sorts of aspects. We're gonna be talking about that in the nfl model governance aspects. And of course we have lots of other sessions focused on adoption and development and management of machine learning models across the life cycle from people across the board who have been implemented on implementing machine learning. So just like our events we've have in the past We have three tracks. Are we have a track. That's focused on the technology side. So you can learn about how to do things and get insights on the from a from a technology perspective. We also have an industry track which is really focused on the industry applications. How are different industries applying these technologies. What are their use cases. What are the some of their lessons. Learned there's also some panels we might have from people. We do have actually throw people in different industries who are share around a similar topic and then of course we have large government and public sector contingency because governments around. The world are making use of machine learning for a lot of things. We're dealing with on our day to day basis of course dealing with things like pandemic and public health but still many other aspects of our life because the government's you know facing the same remote world that we are an ai machine learning Forms apart so we have incredible speakers and sessions and we'll we'll highlight a few of them right here in this podcast right so we have some great keynotes that we wanted to talk about. We have harrison smith. Who's the director of enterprise digitisation at the internal revenue. Service the irs. He's going to be talking about from digitisation to ai. How technology is transforming the government and in particular the irs. I know that many government agencies especially in twenty twenty digitisation and digitalization has been a big focus for that because as we shift from in person to remote work they've needed to really ramp up quickly a lot of their processes to make sure that people are able to stay functioning in their roles. So that's going to be a really incredible keynote. We also have maria wrote. Who's the deputy federal chief information officer from the office of management and budget ownby. She's going to be delivering a keynote as well about Ai in the government. We're also going to have tim persons whose chief scientists in managing director of the science technology assessment analytics team at the. Us government accountability office gao. He's going to be delivering a keynote about planning an agile government with artificial intelligence
Interview With Dr. Abhishek Pani of Bright Machines
"So i know that assembly and inspection in manufacturing is the name of the game at at bright machines. The bulk of your work there. I wanted to put in context what that process looks like. Now what what is assembly when we think about something like electron ix or medical devices and wise such an important part of of manufacturing today. So if you look at the assembly process there actually is a fair bit of automation in front part of the process but you hear about these millions of people involved in manufacturing so what part of the process of the human beings actually solving and that's typically the back end of the process which is putting things together doing inspection wooding some screws in a memory chip in those kind of activities is aware human beings are involved and so there is a lot of human beings doing these back activities and over the last two to three decades. The strategy of manufacturing efficiency has been around moving be backing of the operations to countries which have lower costs and as we know these days. Various reasons changing demographics trade policies head other uncertainties. You want to manufacturer closer to the consumer and preferences are changing. So what needs to happen is if you want to have the ability to manufacture products in a shorter duration. You need to have automation and you need to have systems that can work across diplo and the ability to move your production facilities across the globe and dot smith automation And that's actually right machines. We are trying to solve those losses. Props got into. We'll just we'll just kind of frame up the problem in the process as it is now what i really love to do. In these use cases episodes those of you who are longtime listeners will be aware of this on is we will like to talk about. What's the process. Look and feel like today. and then what's the process look and feel like in terms of where a fit into the the edges in starts to to add value. I'm imagining in my head. Abbot not being fulltime in the manufacturing space. You know we've got a press that will make the actual plastic bits. We've got you know some other shipment of all the little metal. Bits that we're going to need and maybe some of the initial conjoining of the bigger pieces can kind of happen upstream in some big automated robotic processor is flow of materials but then it comes down to putting together little pieces making sure the thing actually works you inspecting to make sure it's not danger damaged in any major way that really is you know i imagine kind of a conveyor belt again this a novice imagination but some kind of a conveyor belt with with the different pieces that i'm working with in my part of the workflow as as a worker and my job is to connect these two things in the items slides down. You know i'm doing my little piece. And then it moves down to the next person and they're doing their piece for example is this is this often what it looks like. Electric's made to get medical devices. Made other things. That need to be assembled now. I think you have described very well. That is pretty much the base process. Obviously they'll be rations on that. Depending on how many units you want to produce how fast you want to produce and there will be fetishistic systems to deal with inherent variability in your competence in the processes and so forth. But what you described is actually the in process and that current process human beings are on one part of the process which is closer to final assembly then on the initial part of the process. Which as you described as much more automated or you have components already available in shape are homeless needs to be assembled at a later stage. Got it okay. This is what it looks like now and we can talk a little bit about where. Ai fits into the mix. I imagine that there have been automation efforts to try to do some of this because it would seem to me abstract the more that we can get done really reliably really quickly and without meeting human hands human attention maybe even human margins of error the better so i imagine companies are working on. How can we have less pieces of the things fit together and we can build a more quickly. And then there's probably another phase where we have some automation of. Can we get a machine to maybe handle. Some of the initial screw is the initial bolts the initial connection conjoining pieces or whatnot. But of course we're you folks are trying to take. It is is a very granular level of almost kind of putting finishing touches on things. Can you walk us through what it looks. Like to apply machine learning and computer vision to this very dextrous tasks but this is just the kind of thing that i think many of us who are aware of the limitations of robotics no to be very very hard sure I think or commission obviously exists has existed for quite some time before we get into how they're solving the problem. It's important to understand the limitations of automation even if you was to go down that up. So the automation that typically exists in manufacturing. It's very custom so at some point. The manufacturer chooses to go with the automation solution. There will be a custom designed for that specific product that needs to be assemble and The process of doing the design takes time. It is not flexible to handle variants of the product or new product classes. It is sequence in the sense you. I needed to design. Then you need to order the hardware then you program. The robots Takes months sometimes even within a year to get the automation solution in place. Now what is the issue there if you do that or one specific product for one specific in the woods your ability to replicate that or move. That is going to be extremely difficult up. Plus you cannot deal with variations in supply chain and your product plus so what is actually good that if it take us back if you have. Zero variance usually go with deterministic approaches. But when you have very the you need solutions and that's the approach taking rare. Can we actually embiid lot of the logic. Insa fair and when i say software it is leveraging machine learning computer vision. It is leveraging the robotics and associated drivers in the cloud. Can we do that so that we are able to actually account for the ability and provide the lippi that automation requires and that also helps drive a higher. Roi through automations. I don't need to think about depreciating asset automation acid in three to five years. And then selling it for scrap i can actually reuse and leverage lot of the automation i did for up info be and that's our
AI Enables Predictability and Better Business
"Call myself a data geek. I before anything else. So i started in the world of data analytics just coming out of college some a longtime ibm Grew up in the world of data and analytics comes naturally. I think to me. I probably think in models I spent a bunch of time building out. A variety of businesses like data privacy. It'd immigration at ibm. I did crossover end. Spend time building out. Api management in the cloud the initial journey that clients took to the cloud and the middleware required behind it and at every step of these areas both on the data during the end the cloud journey there was this constant of security underlying it. And it's always been kind of their But just a few years ago. I jumped head on into a full security role with my current role of leading products for the ibm security brand. I think it really helps having that background and cloud and data coming into this role given what's happening in the security world right now but if you if you asked me to define myself into words ever i'm just gonna call myself a data geek fair enough fair enough but what sort of Transitions and evolutions. Have you seen within. Ibm itself in the time. That you've been there to be honest as a with by data hat on. Which is where i tend to Go to for questions like this and its impact right now on security. There's been on journey of going from reactive to proactive. In every sense of the word rate we in the world of data over the years You did analytics. And you found out behaviors and you found out patterns after something had happened and it's nearly been an evolution over the last couple of decades. Where nearly everything. We do. His gotten to a proactive. Predictive kind of behavioral pattern now. It's not always great but it's greed in a lot of different areas where that predictive ability allows us to do better business it allows us to better client service that allows us to do of a whole host of interactions better. So when i look it clients moving to the cloud and ibm helping them on that journey. So much of it is being better prepared better plan better more structured and more ready for that transformation and a lot of thought. Actually very often comes down to going from being reactive to being proactive and predictive in nature. And i love the fact that it's coming to security. I mean we will always have to do a bunch of defense that's just bought for the course with security but the most predictive we get the more we can understand the environment. We can pinpoint that defense. That is required while being very predictive. So for me. Personally i just as a data. Geek won the world to be more predictive than reactive. And it's a great journey. That i think has gone through. The market's gone through as well and it's not security specific It might have even happened in other areas of the business before security. But i'm really glad securities on that precipice of doing all of that now. And what role does artificial intelligence played in that ability to be. Predictive helps us get there. I know a lot of people will tie to being predictive. I think what does is b. Is its ability to analyze large scale of information in short durations of time and looking for patterns which would be very hard for human beings to do and those patterns allow us to then start building more advanced models any i and beyond that allow us to go down that predictive journey. So the speed and accuracy of finding these batman's needle in the haystack. Kind of things in some cases allow us to be able to get predictive nature. It would be very hard to do without the power of to be on that journey.
Interview With Joe Petro CTO and EVP of R&D at Nuance
"Joe. Petro welcome to the voice podcast. Thanks to be here. Glad to have you really cited for me to get someone from nuance particularly. Who's been there as long as you have. a dominant name and the voice of the i industry over decades and things are a little different. Now because there's also other big names in this space i think there was a long time when nuance only big name. And the voice. Today i space but we we have some other household names that creep into the conversation from time to time. So i've been looking forward to this for a long time. Why don't you tell people just to get started a little bit about who you are what you do on a regular basis day-to-day basis in nuance. Sure so on the chief technology officer of nuance and i've been here for about twelve twelve years or so. When i came in joined the organization. I was coming criminal Medical record bender company. And and i joined a nuance to basically run research and development for healthcare and i think that division at that point in time with something like i want to say two hundred two hundred twenty million or something like that in revenue you know after the last twelve years on that now be close to a billion dollars in five hundred hundred dollar enterprise organization as as well which show. We'll talk about but see. Te'o i'm responsible for all of the products and all the technology and all the research you know that. That nuance does both amana prized healthcare side. And really over the course of the last two years i transitioned from healthcare. Are the svp to the cto a. We pulled it everything our company together and when we When we did that a lot of it's kind magic started to happen. You know we made a lot of progress. You know both Both in the market as well as you from an innovation point of view. So it's been a super exciting a couple years you know as a cto. I basically lead the entire organization. So we worry about you. know how. Innovate and wood products. Bring to market in you. Know how talking to the market our creating messages around the product. How the products connect with you know the value propositions we spend our money as well at the at the same time so this kind of operational responsibility as well. It's been a good ride less twelve years. Yeah i think about it as as looking at your background you are you as an iraqi i think originally and there's not a lot of people i mean in your role. Who have mechanical background. Usually it's doubly computer science something related to that occasionally linguists So so that was low surprising. So how did you get from a mechanical engineer earlier. In your career. What you're doing back into that like full software into eclipses. Emr like a graduated from chemical engineering in the early nineties and And i was graduating in boston. It's kinda distress market at that moment. A moment in time in i was really fascinated with computer aided engineering so the application of computers to really hard engineering problems time. A company called electron data systems was was hiring. It was ross perot company. They had they had a program called c. Four technologies and basically what they did is they. They lived with general motors a michigan in all over the world and they did all of their it but they also did all of the computer aided engineering finding element. Analysis computers factor. It was a it was very much like an inflection. Point because compute was just getting to the point where it's becoming super powerful. So cad system solid modeling So i immediately kind of went out of school in directly directly into that and i was using computers to solve really hard mechanical engineering problems. In really what happened was i got a. I finally got involved in software companies involved with the application engineering. Some leaders in those companies realized. I could talk to clients so i spent a lot of time to doing that. And then you know it's just it feels like it's been a twenty five year journey journey. is kind of really really quickly. I just kind of progressed and kinda migrated up through you know. Increasing levels of responsibility you know had the lucky app stance of running some really really big organizations which eventually position me for you know for roles like this kind of interested in your your time and eclipses to media bars evolved for certain since since you were there obviously was in it was a really dynamic time when you were there to fifteen years ago You know what are your thoughts about how that space has evolved electronic medical records for those who are listening or ernest space. How that's evolved over time because we've got a couple of big players spent some really big concentration of some players but then there's all these satellite systems of engagement and specialty assistance. Which what are your thoughts. On that general. I think in some ways things have come a long way in in some ways. They're very kind of the same I got to be honest with you that that role that i took there when they called me i was actually in a in a distress. Kinda startup company that we we were turning around. And you know when i got the call was an odd call because i didn't know anything about healthcare at the time and they convinced me co executives. They are in the board of directors. Convince me oh you don't need to know much about healthcare. We need some the deliver good product. And i didn't know this at the time but it was. The a lot of people were kind of recycling their way through like healthcare's small community and yes it basically convinced me like we've interviewed everybody. We know who's out there. We need some of outside the industry. We'll teach you healthcare.
Joe Petro CTO and EVP of R&D at Nuance talks about his role at the company
"Why don't you tell people just to get started a little bit about who you are what you do on a regular basis day-to-day basis in nuance. Sure so on the chief technology officer of nuance and i've been here for about twelve twelve years or so. When i came in joined the organization. I was coming criminal Medical record bender company. And and i joined a nuance to basically run research and development for healthcare and i think that division at that point in time with something like i want to say two hundred two hundred twenty million or something like that in revenue you know after the last twelve years on that now be close to a billion dollars in five hundred hundred dollar enterprise organization as as well which show. We'll talk about but see. Te'o i'm responsible for all of the products and all the technology and all the research you know that. That nuance does both amana prized healthcare side. And really over the course of the last two years i transitioned from healthcare. Are the svp to the cto a. We pulled it everything our company together and when we When we did that a lot of it's kind magic started to happen. You know we made a lot of progress. You know both Both in the market as well as you from an innovation point of view. So it's been a super exciting a couple years you know as a cto. I basically lead the entire organization. So we worry about you. know how. Innovate and wood products. Bring to market in you. Know how talking to the market our creating messages around the product. How the products connect with you know the value propositions we spend our money as well at the at the same time so this kind of operational responsibility as well. It's been a good ride less twelve years. Yeah i think about it as as looking at your background you are you as an iraqi i think originally and there's not a lot of people i mean in your role. Who have mechanical background. Usually it's doubly computer science something related to that occasionally linguists So so that was low surprising. So how did you get from a mechanical engineer earlier. In your career. What you're doing back into that like full software into eclipses. Emr like a graduated from chemical engineering in the early nineties and And i was graduating in boston. It's kinda distress market at that moment. A moment in time in i was really fascinated with computer aided engineering so the application of computers to really hard engineering problems time. A company called electron data systems was was hiring. It was ross perot company. They had they had a program called c. Four technologies and basically what they did is they. They lived with general motors a michigan in all over the world and they did all of their it but they also did all of the computer aided engineering finding element. Analysis computers factor. It was a it was very much like an inflection. Point because compute was just getting to the point where it's becoming super powerful. So cad system solid modeling So i immediately kind of went out of school in directly directly into that and i was using computers to solve really hard mechanical engineering problems. In really what happened was i got a. I finally got involved in software companies involved with the application engineering. Some leaders in those companies realized. I could talk to clients so i spent a lot of time to doing that. And then you know it's just it feels like it's been a twenty five year journey journey. is kind of really really quickly. I just kind of progressed and kinda migrated up through you know. Increasing levels of responsibility you know had the lucky app stance of running some really really big organizations which eventually position me for you know for roles like this
Diversity and Inclusion Opportunities in Conversational AI With Roger Kibbe, Senior Developer and Evangelist for Viv Lab
"And kind of building off of that We kinda started speaking on it. But what would be your perspective on the state of inclusiveness and diversity in conversational ai. So so i like to think of things. Globally sometimes big picture and draw down to the to the more local small picture but there is something aram near eight hundred million people in the world who are illiterate and although the internet. There's something like i believe that. Internet access about sixty percent people have internet access. That means four point. Two billion people don't have internet access. So i think about. There's this huge group of people who don't have internet access and nan bear is a non trivial portion largely probably of them who are illiterate And so one unlock waste can be a technology certainly to those eight hundred million people who are literate right. it'd be a way for them to interact with technology in a way they were previously completely locked out but i also think of home bringing a voice assistant may be a way to bring technology to and the internet and access to people who otherwise don't have it because it's relatively cheap and inexpensive compared to some other technologies so i get all excited about kind of the world opening up with voice but i think even if you think about let's just take it down to an i'll just talk about the us. So certainly. the diversity of our population is not yet reflected invoiced designers and developers. And you know it's not only Being a good human to be diverse. But i also. I always say it makes great business sense when we're not diverse we you really constrain. The number of uses of our product look. I'm a white middle-aged american male. That gets when i build experiences. I build experiences that likely. Resonate with white millage american males. That's a very limited subset of the people. Though the more diverse that people are who build your product the more diverse your customers are traits so it behooves the industry to really have a diverse a As possible people building products. Because they're going to bill things that there's a whole audience that then will love that experience become engaged with it and it just makes you know you get another virtual site virtuous cycle where You build more diverse things you get a greater more diverse audience. You get more users. It's good. I think when i look at voice i think we're doing okay and i say that relative to the tech industry and i say we're doing so on a on a gender basis. We're probably doing better than some other areas of technology which is heartening to me. But there's certainly some work that needs to be on there on a racial basis. Now we have to do. I just don't see us as nearly as diverse is like. I said that the country as the world is You know we need to. We need to go out and find those people with diverse backgrounds and diverse diversed genders and diverse racial to go become part of the industry one of the beauties. This is a new technology. So i would love us not to make some mistakes that we've seen in the past new technologies where the same old same old belt built the new stuff right. So how can we go and get people who never touch technology before. Bring them in and say. Hey here's this. Cool new tack a be part of it be part of that as a consumer but i also would love to be part of it is go build design. It right are building. 'cause you're gonna build super cool interesting experiences. That wouldn't have been thought of in less someone with your background. was brought into the industry. So i could go on. And on. But i think we're we're doing okay but darn there's a lot of opportunity up there we need to keep on talking about it and besides hockey we can't keep acting upon it. It's it's kind of imperative.
AI for Translating Enterprise Content - with Spence Green of Lilt
"So i know. We're gonna be talking about ai for translation and clearly. That's a a workflow that you folks are impacting. Get an understanding of what it looks like. Now you get a big brand like an intel or some some large company. That's got localize all their web content. How do they do that with people. Sure so dan let me let me. I tell you a story so you can understand what the motivation for doing. This translation is so most people today are familiar with google. Translate and use it when you browse the web and you could translate chrome or you use it on your phone and you'll see this in a lot of consumer devices and that's really useful for these consumer use cases where you on translate a sign you see on the road or you see a restaurant menu or you just want to get a gist of a web page of language that you don't speak okay so that not as widely used. Google translate has had enormous impact. And everybody's familiar with that. The second case. I think it more. Broadly is in business and so our motivation for building. This company was experience. I had about fifteen years ago. I was living overseas in a country. A non english speaking country Were but were. Linguists was spoken as lingua franca. And i had a friend who didn't speak english but spoke the native language of this country and he told me that he was paid less and could not get a job because he didn't speak english. And this occurred to me as a sort of inequality that we don't talk about very much are you can learn a language but you can't control your native language. If you don't speak english it's a barrier to opportunity undertake percent okay so taking that as the premise. Now take got to business. Businesses sell their products and services. And they wanna make those available to people in different languages so that they can do their job so that they can do research for products so that they can learn and presently the way that that is done is by hiring people to type translation. There's assistance used but for the most part it's just hiring human translators to translate much the same way that transition has been done since i don't know before agricultural so what we wanted to do and no machine translation was used so this was about twenty ten when we were learning about this problem. So what we are initial idea was to emma and i met working on google. Translate to take that technology and art meant what translators do the translation. Problem is not a solved problem so we can't fully remove the human translator but we can certainly amplify what they do and if we can do that then we can make it much more affordable inefficient for companies to make all of their information available to anybody who wants it and so for comparison right now companies pay about eighty dollars for an eight and a half by eleven one and a half spaced. Page of twelve point type. So it's very very expensive. Let's ten yeah. Yeah and so we wanna use technology to drive that price down so that for a fixed budget. Companies can localize more of their information. That's good for them. They can grow their business. And it's good for the world because more people have access to the products and services that you and i take for granted. Yeah so. And i guess there's like the equality element here that you're sort of bringing in on the side but clearly it's just like if intel wants people in japan to read their stuff. There's that benefit to right. We were in english primarily we can. We can move our product. Yeah that's right. There's there's four hundred no there's about four hundred million native speakers of english there's about another seven hundred million l two speakers and then there are a couple of hundred million speaking for language. You're talking one and a half million people in the world that speak english that leaves what five and a half billion humanity. Yeah regarding of humanity right so it's it's not just for positive social impact. We're capitalists and this is an opportunity for businesses to to grow their son of customers at eighty bucks a page. I mean that's a business opportunity right the folks that we've seen working on translation of all kinds of pick their lane. Your lane seems to be sort of digital asset. So maybe we'll walk through what that looks like now. So let's say a big company knows a general electric you know they're they're got a new jet engine in they're going to sell it around the world as do normally they're going to create their when it's web pages it sales collateral you probably know the list and then and then they would hire their team of people hopefully have a relationship with who are going to a really nice fine comb job at eighty bucks a page of four for all those different assets that more or less what it looks like now. Yeah that's exactly right. So typically a business will decide we're going to do business in these markets and that's a business decision and then the choices which of their of their english experience because usually companies translating out of english four. You're going to make available in that market and so for example. One of our customers is intel. Everybody knows intel's products and intel sales. They look lies into about sixteen languages right now. And so that's everything that's until dot com product manuals technical specs marketing materials. Everything having to do with intel's products and services and it turns out that despite the fact that translation so expensive it's typically less expensive than offering new content from scratch every day work in the second sort of operational constraint is companies generally want a consistent brand across their products and services. If you're offering things in different locales you've got a lot more people involved. It's very very difficult to control the brand so it's much easier to author in one language. Control the brand messaging there and then localized into a bunch of different languages. So that's kind of the business. Motivation for
Model-Based Offline Reinforcement Learning with Aravind Rajeswaran
"All right. everyone. I am on the line with arvin. Roger swaran arvind is a phd student in machine learning and robotics at the university of washington. Arvin welcome to the tuomo podcasts. Thanks famine federal pleasure. I'm really looking forward to our conversation. Motto based offline reinforcement learning is the topic of research papers called morrell motto based off. Find me enforcement learning. That topic has come up quite a bit over the past year or so It's getting quite popular. And i'm really looking forward to digging into your take but before we do. Tell us about your journey and how you came to work in our l. and robotics chip pretty interesting question my undergraduate background. Actually in something completely different. I was mostly doing big statistic and lucrative like chemical engineering as my formal degrees and i took a machine learning class by professor of in back in india and that kind of transformed my perspective on things that essentially had matt very similar to what is used in statistical it so i was able to pick up on it pretty quickly by the obligations. Seemed like really really cool. So i wanted to maybe fever debate and focus more on machine learning. And so that's how i moved into the bronx field of ai and when it started out i had more of a theoretical inclination ben. I started working with my adviser. Professor shawn kakabadze. Who's like an expert in machine learning theory on be discussing like what might be interesting projects topics to work on or we felt was start matz. If the research decide in machine learning deep learning was largely explanative nature so deep learning was already working really well and the gresh any hard. We understand explain. Why deep learning working enforcement learning. What was interesting is that we actually didn't have very good algorithms things. Were actually not working that ball. And there was a very interesting scope to have like an interplay between teary analogous from both develop new algorithms to break very well and also tried to explain why it is working than gain a more fundamental understanding and that sort of been might be at the joining us electric to show board they. Competitively goodra sauce ended the same time having a theoretical bench to my work. Nice nice and as the large focus of your work been on model base are l. in particular or have you explored a wide variety of topics within the space. I see interesting in model based in fort learning relatively recently the i'm i think about my research at least in the last couple of years has been the central question of. How do we make a. How do we create agents that can solve a diverse set of tasks with a modest amount of expedience but each individual dos and this is of course a very broad question that the number of different fields including multi task learning micro. Learning offline learning. And so on that constitutes space of problems that have been thinking about what you think is really cool approach algorithm to make progress on these domains model based startled so i view model based on. That's providing the mic toolkit to make progress on questions related to multitask turning meta learning offline. Learning got got so maybe the best way to go through this is to start from the beginning And have you kind of explain the. I'm curious the way you explain. Model based rl. And if it's you know the extent to which it'll be different from other explanations we've heard here on the show so let's start there. Yes so. I guess maybe i could try to merge some of from model based on offline learning on. What got me working on this project. I'm hopefully doing that. There may be an explanation for what is model based donal on. If you think about firmament a moment blake questions and computer vision are be the questions. There tend to be much more ambitious and interesting than in traditional reinforcement. Learning for example. The questions to be still asking learning are how can we solve a particular tastic pickup. A particular object with the robot solve a particular game with. Ask me more samples as possible. This is very different. From how people phrase digressions and computer vision for example offs kennedy chanukah detector with ten samples can attain got victor with twenty points. That's just not an interesting question. The questions are the how can i identify. What is the object. An image of thousand categories much more broader in scope and much more ambitious daily. Use a lot of will get to make that happen. Mine goal was to try to emulate that in reinforcement joining us. Now if he start thinking about how can an agent which flows in a very complex A kitchen for example. There are so many things that it can do the cabinets they clean bans the setup dishes and so on the dishwasher on the speed of things that the robot can do so much more diverse than what that means is we need to be able to use the data to extract as much information about the world as possible and i believe models are the we accomplish doctor just given a particular state of the wall some either league and explicit stately things that particular joined configurations of the robot throughout the different objects in the armour richard descriptions such as images and video slater scans of assets points to any potential action that the robot can take how would the world evolving change on if you are able to learn such a model it modern sculptures many of the details that want about the world on on the basis of what we learn. We can then downstream from finding and reasoning in order to accomplish necas of interest. So in my mind to go back to your question what is a model. I believe it is. How would the world respond to any changes be made potentially making the former factions
"Also win elections by redrawing districts in a process called gerrymandering in which voting districts are redrawn. Not with the intent of collecting geographic communities together but rather to strategically positioned boundaries of support this data skeptic consensus in the twenty second installment in our series about how multi agent systems achieve collective decision-making today on the show. I speak with brian. Brubeck for a deep dive on these topics and discussion around his recent paper. Meddling metrics the effects of measuring and constraining partisan gerrymandering on voter incentives. I'm