Digital Production Buzz - Oct. 25, 2018

Automatic TRANSCRIPT

On the buzz. We are looking at the impact of artificial intelligence on media as we talk with industry thought leaders about the current status of AI where it's used and how it impacts the work that we do we start with Philip jets. The CEO of lumberjack system is a technologist who is thriving the discussion regard to media tonight. We talk with him about what is how it's used and what we can expect in the future. Next candy Steinbach this assist and founder of paradigm shift talks with us about the new science of AI machine learning versus deep learning and how AI is being used today. Next Sambo Koch, the CEO of axel explains. The practical uses of AI in media. He describes how it can be used for speech to text transcripts. Object. Facial end local recognition. And tagging footage to allow us to find exactly the section. We're looking for all of this. Plus James Rubio with our weekly Donal news update. The buzz starts now. Since the digital film-making. One show serves a worldwide network of media professional. Uniting industry experts Russian filmmaker Russian and content created around the plant distribution from the media capital of the world, Los Angeles, California digital production goes live now. Welcome to the digital production. Buzz, the world's longest running podcast for the creative content industry covering media production post production and marketing around the world tie. My name is Larry Jordan almost every day some emerging facet of artificial intelligence. Makes the news headlines like will robots take your job 'cause all of us concern in two thousand thirteen Oxford University estimated that forty seven percent of all jobs today were at risk of being automated reports since then have lowered the number. But while no one questions that Amish in will have an impact there is a lot of debate on how much tonight we take a closer look at artificial intelligence in post production. We talk with three different technologists to help us get a better idea of where is being used today. And where it's. Likely to take us in the near future, by the way, if you enjoy the buzz, please give us a positive rating and review in the I tunes store. We appreciate your support to help us grow our audience. And now it's time for our weekly Donal news update with James to Rubio. Hello, james. And welcome back. I'm back ready to rock. So what's our lead story this week? This is like a really sudden breaking news kind of story lex are has decided to drop their plans for executing media card production and will instead focus on compact flash express. They cited difficulties with Sony and other parties in getting the approval status. So therefore, the resurrected media card maker will instead focus on the future with the more eight K friendly, compact flash express card standard, and that's not a bad thing either. Because cameras like canon Nikon's f. Series and panasonic's new S one camera support compact flash express with a simple firmware update. The cards work in the exact same executing media card slot. So what does this mean in real life? Well, sonny's always been rather proprietary in their thinking and with insecurity being one of the few exceptions. So now, I'm wondering if after lex are closed its doors the first time, then Nikon Panasonic avenue. Mierlo cameras that support the executeDi format Sony may have decided just to start dragging their feet in order to just completely control the Cutie media market on the other hand since compact flash can work in the same cameras with that simple firm where gate and is up to eight times faster than execute for file transfer. Perhaps this is just a case of lex are deciding you know, let's just move forward and not backward because we're looking at an AK feature. All right. That's story. Number one. What story number two? Well, canon has finally patented and ibis image stabilization system that will have the DSL our body working concert with the image stabilization of the canon lenses to create a more five access kind of image stabilization system, but it could only be further DSL ours. The description doesn't engine either DSL are or merely camera in the patent application, but the patent drawings feature a DSL our camera and not the use are merely design. So where do you see this fitting in well knowing the snail's pace, conservative nature of canon? It wouldn't surprise me. If this new ibis technology would make its appearance in lower end use camera like maybe a seventy or sixty or maybe even a five d Mark five before it goes into the EOC are merely system. That's just the way they test. Out feel test. Their new emerging technologies they did it with dual Seamus oughta as well. But Canada's also said it's focusing its attention on the Mierlo platform. So perhaps a Mark to Yasser Murless camera could get the either way if good news or Karen fans. What does the ibis system do that lend stabiliser doesn't the Ida system works in combination? You'll have basically you'll have to image stabilization systems. You'll have the system inside the camera body itself, and you will have the image stabilization inside the lands both working in concert to damp out any camera movement. It's just a more efficient and affective image stabilization system. Thank you, James. What are the stories you follow in this week other stories we're following this week? What media cards are best for the EEOC are merely camera and the black magic pocket cinema camera. Four K, our friend, heath McKnight returns with a review of film convert color. Grading soft. And Panasonic puts out a new jets g h five update only a week after its last one. That's what we're working on what's tonight show about tonight. We're looking at the impact of artificial intelligence in media almost every day. I'm seeing new articles on the growing impact of a on our lives and industry. So it's time to take a closer look. Well, as long as his name isn't how I think we'll be okay. And James where can we go on the web to learn more about these and other stories you and your team recovering all these stories and more can be found at Donald news dot com and James to Rufo is the editor in chief of Donal news and joins us every week. I'll talk to you next Thursday. Have good weekend. Here's another website. I want to introduce you to Donald news dot com dot on news gives you a portal into the broadcast video and film industries. It's a leading online resource, presenting news, reviews and products for the film and video industry. Donald news also offers resource guide and crew management platforms specifically to sign for production. These digital call sheets along with their app directory and premium listings provide depth organizational tools for busy production professionals dot com news as part of the failure arts community, a world wide community of artists, filmmakers and storytellers. From the Tara feet of filmmaking performing arts define arts and everything in between is filled with resources you need to succeed. Whether you want the latest industry news need to network with other creative professional's require state of the art online tools to manage your next project. There's only one place to go, Donald news dot com. Philippon is recognized as a leading technologist as well as the CEO of lumberjack system. Even better. He's a regular here on the buzz where he specializes in explaining new technology. Hello, philip. Welcome back. Philip tonight, we're looking at artificial intelligence, but before we start talking about AI, we should probably define a and some other terms. I've heard used with it like machine learning and deep learning. How would you define AI? Umbrella him, but everyone is locked onto. And they needed to go to that. Oh tournaments intelligent machine. Unfortunately, no, we need that out of people. Somewhat tournaments Shane's that can make decisions narrative, my that would be out of intelligence in at proper. Machine learning where I particularly Shane can be trained to make decisions based on very specific remedies recognize the skin cancel something like that. And then we have the date learning stuff wear training, the machine come sort of challenging, hey now go for that's kind of devote maybe couple of examples of h wanted clarify different way into that into the landscape term. Intelligence is would be something like a self driving car. Fabled Thomas make you, but it's within a limited to mind, I would not expect mile tournaments. Casa my my loan. For that. Machine learning is what we we've really come to get a lot of benefit from Shane when he behind all of these speech takes transcription software. We've we've benefit from now we behind immigrated in recognition, it's behind DIVY sense. As Kalama matching the latest snow for me pro and automatic docking in in rerun. And condition for that matter. I hadn't caught up on that. These may she thank you. And. And things like predicting which mile ball up particularly Email not going to in in f- amount. Simple application machine. When colorizing was lot of a lot of ways machine Lenny has been used and the loonies way give a computer shown, so. Have of humanoid characters by humility character that they gave it environment included, gravity, basically move forward. The upright and interested open over Philo thousands and thousands of million times until it work at how it will could move forward. Under those conditions. Mrs closest figuring a computer, but it it's. The challenge and machine composition music is getting Shawn's to get better and better with time. And sometimes they say machines against each other. I actually waking pieces over is challenged by it. Thanks generating self like face is not think they find. Take a breath for second. If I'm hearing you correctly. I just want to summarize this if I'm hearing this correctly, a is an umbrella term and machine learning is where we are teaching the machine we meeting some subject matter expert and deep learning where the machine is iterative learning on its own if I summarized that accurately. Lovely. Yes, exactly precise. As other categories of artificial intelligence between aside from how the machine learns. Or another words, what are the categories? Are. We going to hear buzzwords about. Sure, I understand. That question. Sorry. That's all right. I'm just curious if I've used to terms that are ways that are out official intelligence is taught machine machine learning and deep learning, but I don't know enough about a I know those the only two categories if there's other categories as well. Well, that is the autonomous artificial intelligence. But but medication you, right? There's machine leading their leading there. They were very Asians on learning. But that's not I think. All right. Well, I'm just thinking that seems to burst into our consciousness over the last couple of years, though, I'm sure it's been a development for a long time. But what's happened in technology that suddenly made a is so accessible to everybody. It was. The history of intelligence as the umbrella tone. There have been many to get a little bit of progress, and then Ryan's it wasn't. So we Newell networks which are able to have feedback mechanism. So they can train machine money thought of it becomes impossible and out of a different strains of research today. Quitting had come as well. The generally thinking they have research project, the don't touch most people's lives, we most benefit from the end of a model being trains and and run on on Aaron Ave systems. You used to term that I also needed to find you talked about a neural network. What's that? You know, it the metric blackbox. Series of indignation liars as the things we and that his has the trading feedback loop, which is what needs some sort of trying to be trying, but the neural network just basically gets his about what if it's close and twitchy to gets more in that direction. It's close to the right treatments. Open the office direction. They did possible to trek through believe very school for insignif- magicians and look at exactly how a machine making the decisions that gets trying to might. But basically, it's a black FOX that trying just run is the neural network. The hardware is that the software that the hardware is running or is the data that computers can fed. I really want to say, yes. Because it has all all as I see. I mean, it is obviously has a hobby because all computers have layers some point. But it also as a way of of tiny theater in soft with and without training data. It's probably not get actually happen. So that that tiny day to provide the feedback loop, which is why? So essential has noted the data that you will also the ranking that you want so even if it was theoretically possible to generate a machine models that would recommend an ADA way, we find thousands of millions of examples raided Edison, if agree on the grade that much more suited to recognize. He's in cash has very precise. Accuracy to mention some of these things that we can make you and successfully guide in the Shane Cologne. Okay. So far we've been talking globally about what is but let's drill down into media. I mean, we've got Syrian Alexa to consumer level. And extremely sophisticated industrial machine control at the enterprise level. But for the purposes of of the buzz, how is being used in media today likely what I group under the heading coordinates of these things that are shame little already trained, and as I mentioned earlier things like they should take tricky nation in him as an emotion recognition entity recognition, loco recognition these things that can be trained object detection, these things are very common in exit. Well, no, common SMS major assistance. But it certainly want asset management system that office as great Mesa office of excellent. Hey, I is none that uses machine learning these services to help find it. So it media, and I certainly see that we should have this sort of indie and speech building. Editing. So in the future taking Biro is nowhere near as fun. As it seems like when you think about it. Use the word training multiple times. And I've got a picture of you at a whiteboard. Talking to computer was training actually mean. Well, you have to teach the machine what you wanted to do. I know that very day. But. For example. I think it's such an easy. It's amazing example machine has been trained on numidian of radio allergy and images of skin cancers and been ranked by radiologists as this one has no scene came. So this one is being cancer prisons this when I was in this location, and so on so we have been lodged at data that he's scribes a team each, and we get the we feed each image Tuesay hero network that is the machine again to try, and and it says, well, I'm gonna guess that is a cancer in this corner. And it says no sorry try again. And it does in a deter tries again, and again, and again until it gets ninety five ninety six ninety seven accuracy, and we could use you can live with. Italy you can use a apple provided have to get back into the my Koro mill framework that can do any recognition, but he might want to try on the meticulous set of images that you want your act to to recognize that they've done the heavy lifting and tried it on the last mile that we still need this training data and have to train the middle. We need another day to actually pissed with kind the modal. There's nothing spiking out cheating on the test. So to speak. Because we can deployed into the real world. So in order for a machine to be trained. We have to have an expert. That knows what the answers are to say, this is the image. And this is the answer in the machine learns from that over the course of repeating multiple times one by the that if that if that machine that data said is in a machine readable full will that would make it much easier. So no whiteboard. No. Like this. Your software lumberjack system integrates into it software. How are you using it for your stuff? Well, I wouldn't based on reviews, I wouldn't call it. I would close the shade money. Certainly we've checked completed the integration of speaking to take in the southwest. So that you can once you import your event final prochaine, you can choose which clicks, you send away for transcription ends up back in staggering eight times real time, and our so if for example, if you had a forty minutes that would all be back in about five or six minutes. What go ahead? And then the next step is to do extraction and t would extraction from those transmit that date or this out that down except. What's the advantage to editors to having this machine language applied to our clips? Well, it makes everything easy to find out. Typically, they were too hot to doing the role of editor one is finding the material wanted sequencing material in an acidic leasing the emotionally compelling way. And if you can't find anything in the project, then you're not going to be out with get discover ability when we first transcriptions into final pray for those AK incensio commensurate seven they found that they could they located footage locating satellites that they would not otherwise fine because they could search the entire transcript by would and behind by content rather than having had something tagged to be on the line. And. What do you see is the future for machine learning for the medium-term not future future because that's impossible predict? But like next year or two. I think we go to find more and more smart assistance, Ivan bilking, assault where or edge up to out arid software, and I work for processes. I mean smarter search of find ability of exit. That's going be pod smart column, attaching smarter audio schools a lot of the things that we we do is part of that process can be assisted. I like to to talk about state Yeltsin, his metaphor, all you know, I manage not cross the fastest animal on the planet. But a man on a bike is close to the faucets and blue on the planet and thinking with the smartest systems we're going to be closer to the the man on motive. Because you know, a lot of a lot of what gets in the way of people being created the open is eight. I if rate creative ideas what it comes down to actually old with hero and finding story in the they they run into this wall and become homeless possible them to progress. And I see these the tools that hope through what we've got and pull out storylines and fees that we can build on apply creativity to I think that that's the next couple of years and Philip for people wanna keep track of jor thinking and work in this process. Where can I go on the web, Philip Huggins dot com? When it's my put the playground and lumberjack is builder can be found. Both of those are single word. Philip Hodge it's dot com. Lumberjack system dot com. Filipacci is the CEO of lumberjack system. And thanks for joining us today at my take your. Andy Steinbach is a physicist with a PHD from the university of Colorado boulder. He's led to GMs developing some of the world's most revolutionary technology, including the world's highest resolution microscopes novels, so my conductor processes and high speed optoelectronic devices at companies like Invidia Kalay ten cores ice and J D S Uniphase most recently, and he has launched the consultant company paradigm shift, which is dedicated to helping other prizes develop a and deep learning applications. Hello, Andy, welcome. Hi there. I don't get to talk to physicists, very often. You're at the leading edge of making this stuff happen as opposed to the rest of us who are trying to figure out how to use it after it's been invented. Which brings me to my first question. What interested in AI? I thought it was starting to solve problems that were so complicated. You can never. I thought that they could be solved. Can I ask myself wine is businesses? You know, what I realized is it? We're calling one of the most difficult problems that we face in physics imagine that you have a system where you have a complicated computer network, and you're trying to detect cyber security attacks. It's just an example. There might be a thousand different variables that allow you Entschlie repet- when a pack it is an attack or not an attack. But the problem is that when you have a thousand variables get something called the combinatorial explosion. Those variables take all the values. They can have you very quickly get sort of an infinitive combination. What I realize is that as neural networks started to finally come into their own and really started working around about two thousand twelve that they were actually solving this problem. Did you ever wonder why people's brains? You might have heard that the synapse. Is in the brain fire at five milliseconds that's two hundred times of second. Well, that's really slow. If you think of the computer as a big powerful brain question is why is it powerful? Because we know that these wonderful chips, we have nowadays operated gigahertz, and even gigahertz the what is the brain works. So well when those connections fire thousands or millions of times slower, and it's really two reasons. One is that it takes all the input in parallel. Like envision these narrow networks take all the information in parallel. But the second big reason is they process the information through a series of steps and the way that these algorithms that neural network essentially, execute it's designed as sort of decision tree like a flow chart where at every step you eliminate a whole set of decisions in this big combinatorial space. And then you never go back and look at that. Part of the space again. And so you can imagine a big sort of fear, and you make the sphere of where your answer could be smaller and smaller and then eventually gets to the right answer. And it happens exponentially quickly when you step through a series of steps, and so long long story short these narrow networks finally started working, and they solved this combinatorial explosion this cursor dimensionality. And so we can finally solve problems that were just simply off the table and the performance that companies like Google and Facebook getting and researchers demonstrates that they're succeeding, and so it's really a new scientific revolution. And when I saw that I I wanted to get involved. Well, before we get to deeply into the subject. Let's take a step back. How would you define artificial intelligence, and do you consider those two words to be synonymous with machine learning simpler end of the spectrum artificial intelligence, you can think of that very simply as a framing a problem as a question and answer pair with machine learning you need a domain expert to do what's called handcrafting features. So it's not very scalable where people earning will automatically discover or detect these features. And so you do not need the main specific knowledge in order to make people earning work. Really? Well, so it works better higher, performance and it works without the knowledge to that. It's scalable, so that's a winning combination for her heap learning if I were to summarize could I say that artificial intelligence of the category on a subset of artificial intelligence machine. Learning and another subset would be deep learning. Yeah. Absolutely. I think the way to think about it is artificial intelligence is very general term people that technically studied the topic would say that machine learning is is a large classes message and deep learning a specialty class inside of that. You're absolutely right. You recently gave the keynote the storage visions conference. What was the topic of your talk? Artificial intelligence is now science this coming to town with some of these technical innovations that I explained earlier it really is a new science where rather than again going back to the sector with most sciences means people that spend essentially their whole life studying a specific domain, and they've become world experts in it with planning and de Niro networks, you can now take large amounts of data, and you can train models his to give you answers. So it goes back to this sort of question answering capability without having these. Domain experts just feed it all the data in the beginning. And it sort of figures out the domain knowledge on its own, and that's really a revolution or paradigm shifts in in the way that science is able to be done data science is really going to be a new science that sort of on the part of other sciences like physics chemistry and biology some examples a good example and in -application area that I'm really excited about is. So called IOT IOT censor the internet of things any machine or instrument nowadays, it could be your toaster. It could be an automobile or could be some industrial piece of machinery is in manufacturing. Plant has a bunch of a bunch of sensors that are typically wide computer network and the sensors bid out data, and they're monitoring all the inner workings of the machine whatever it is. And turns out that you can take that time. Data and you can you can apply. Deep learning algorithms, and you can analyze all kinds of things about the state of the machine. I'm particularly interested in the case of what's called generally industrial IOT, big pieces of machinery that you would have an industrial company or factory with these industrial IOT deep learning algorithms, you can do predictive maintenance. So in a maintenance cycle, you can say when the tool needs to be maintenance. And of course, normally you do it by some preset time period every two months, but sometimes it needs to be service faster and later, and it can affect the performance of the outcome of the product that it's producing. And so wouldn't it be great to be able to understand where you are in that cycle? So that's called a predictive maintenance, you can also do what's called health monitoring, so many of these complicated. Industrial tools naturally have failure modes that occur from time to time in some. Cases frequently. And so wouldn't it be great to get an early? Warning prediction that something in the operation is looking just a little bit different than normal. And it hasn't affected the performance of the the thing or the parts let's say that that that tool is producing, but that it will. And if you could predict that you could really improve yield in factories. That's a I used a monitoring obligation are there other examples of how I can enable an enterprise another example that I'm excited about is something called reinforcement learning. So imagine that you're doing this industrial IOT application, and you're monitoring some tool, and you can predict that something's going wrong with the tool, and so that you should perhaps take that one off line and services. So that's what's called predictive intelligence, but even when operating normally wouldn't it be great not to just be able to predict what's going to happen to actually optimize the performance of that tool. So that you can increase your. Yielded the performance of your product it's producing and so there's a way to do that also on with learning and that optimization is called reinforcement learning or reinforcement learning that's a very exciting area. And some of your listeners may have heard that for example, Google plied that to optimize the power usage of the giant computer data centers. And so they were able to save something like thirty percent on the cooling tower required. So that's a really nice example of the real world application, you've talked a lot about a in tech. But can help us in medicine one application medicine that I'm particularly cited about is something called precision medicine alien rhythms. For example, can take a look at patient data records and so one application, and I'm working on is g electrons Graham, which is electrodes on the brain. And you can take a patient, and you can. Diagnose some for example. Are they are they with for depression? Are they risk for having an epileptic seizure? And that kind of thing anything classify them as normal or having a problem, which you detect, but when it's great if you could use those learning algorithms on that data, not only the to tell them, this is what problem you have to actually help treat that problem that basically look in more depth at the information, those medical records contains and what you planning is really doing is comparing that particular patient with finding similarities between that patient and other patients that has an database or training data, and it's comparing them to successful treatments, and what those treatments worth from other patients and by doing that you cannot only predict that you're at risk of epilepsy. Seizure. Let's say, but it can actually discern what treatment would be more effective than advise the doctor on a how they might best treat that patient and says that the whole world, and it's definitely coming. These are very very exciting powerful applications as learning in medicine, and in all three examples that you've given you've used the phrase one thing I'm really excited about is. Why is being excited about a subject important to you? I guess I'm excited for numbers reason. And that's a great question. But I think for me the reason that I get excited about an application is there's a technical side says the Cest on really thrilled every day to get up and be working on these amazing algorithms like I said to solve problems that we never thought were solvable. But what I love is nesting that -nology to the end up location and vertical. So I've spent significant parts of. Krier on teams building these industrial machine factories, and I know what it would mean to those customers to make that machine better and not only to those companies that make the machines, but to their customers, and so it's really enabling technology that can improve how we live in society. And so medicine again, it's other sample. I love the algorithms with the idea that you can now apply. Those and make health care better is is in mazing that it passes through. So clearly to such a clear line to the end application and the impact on people society is the province only of the largest companies or does AI filtered on we all with Alexa with Siri. We take the benefit of AI. But in terms of smaller companies, can they harnessed the power of AI for themselves. That's a great question. And it is challenging and that's actually what led me to start my consulting company paradigm shift. The I because. It is the new technology. It's wonderful for companies like Google that have thousands of researchers doing it, and they have twenty years experience, and they can get the dust people. But it's a challenging if you're smaller company and you're asking yourself. How could I get started in two problems that they say if you generalize it, which is one piercing veil all the buzzwords trying to understand how all these moving parts together. What's machine learning learning and things like, you know, what what sort of applications could. I do once I decided not what should I use? What what would be biting off more than I can choose versus what is sort of feasible to get started. And so all those questions. Some of them are strategy questions and some of them are simply technical and at the end you end up with an application that you want to try to do. And then it's very hard to get because there's really a supply demand in balance if people that that, you know, how to do this work. And so the reason I started my company is to provide a resource both for that early sort of strategy formation say's. And then also for the ability for companies to outsource this finding and hiring technical talent. And to get started in and maybe they want to eventually build their own internal team or maybe they're happy to have customized engineering done to create applications that we help them decide on we can decide together. And so that's an interesting challenge. And we're trying to address that, but is very cool for people to want more information about what you and your company can do for them. Where can they go on the web our website isn't quite up yet? It'll be up in about a month when it is. They can. Go to WWW dot paradigm, shifts dot that website is all one word paradigm shift dot A I P A R A D. I G M S H I F T paradigm shift dot, hey, I an Andy steam Bach is a physicist who's been studying and Andy thanks for joining us today from my. You're welcome. I want to introduce you to a new website. They loaded dot com. Falem is an artist community and networking site. For creative people to connect the inspired and showcase their creativity. They low dot com features content from around the world with a global perspective on all things creative. They load is the place for creative folks to learn collaborate market and sell their works. They Lewis apart of they low arcs of worldwide community of artists filmmakers and storytellers from photography to filmmaking performing arts fine arts and everything in between. They low is filled with the resources you need to succeed. Visit they load dot com. And discover how their community can help you connect learn and succeed that stay low dot com. Sambo Koch is the CEO of axel, which has developed a new approachable system for asset management also called axel AI, which prominently features not surprisingly, they I as part of its technology. Hello Sam combat. Thanks much. Great to be with you Sam before we start talking about artificial intelligence itself. Tell us about axl what is your product. Do it. Basically scans the contents of folks storage, usually containing lots of video because that's our intended customer base, and we have over five hundred customers today or from a number of different walks of life. And basically it scans their storage makes lower as versions of all the media that it finds and then dispatches those to be analyzed by one or more. I engines some are on board built into the software and others are outboard as in the case of say Microsoft's the index product. We're starting to work with a growing list of vendors because the space is getting very diverse and and very competitive in terms of who can analyze things the best. Well, this is an interesting concept that I hadn't considered before they I is sort of a description of machine learning and deep learning, but you're plying that there's different products that do a I meaning there's different functions within it. How does axel use? An what do you mean? By different engines to give a short list of the possibilities. There are already a engines on the market that will do things like transcription rights speech to text where you can just take them soundtrack of something. And and spit out text of what the person or people said, there's they shall recognition where you can actually identify who is in the video, especially if they're celebrities or public figures there's object recognition where you can pick up. That's a chair that's door car cetera. And then there's things like. Brands. So you could pick up brands when they appear on screen, you can do OCR character recognition where for instance, if if it's a player, and you can these wearing number forty five number forty five will show up in the meta data, and there's even things like emotional analysis. You can tell if the scene involves people getting very excited or happy or sad. That is actually just a short list of what's available right now because the options are proliferating and video is getting a whole lot easier to analyze as a result. How does it work not in terms of hot as the recognize the characters because I think that that's too deep, but in terms of my operations hot do I get to determine who the people are on in a video or the speech to text. What do I need to do with our software? We're trying to simplify it. So that you basically need to just let the software loose on your storage. It scans all the videos makes low rez versions, and then you can pick from a menu say I want all the videos. This folder to get sent off and analyzed for transcription. Or I want all the videos in this other folder to get sent off to identify the bases there in the video. And so we give you literally pull down menu. You send these things off to the right engine's in some cases like with object recognition where able to do that on board in our software without any external fees, in other cases, there are plowed vendors, for instance, that will let you do a few hundred hours a month for free, and then above that, they charge you a certain amount per hour typically anywhere in the range from four to ten dollars an hour, but occasionally more than that. Why did you decide to implement this because AXA was doing fine on its own? Why the shift into a? Fundamentally media management is is a chore. It's something that's done way to sell them because nobody wants to do it. And in the past required, a whole bunch of interns or a bunch of time that people didn't have. And this is even in big famous broadcasters will go nameless. But I I've met with some of these folks, and you would think they would have their act together and everything would be catalog, but they literally don't have the time. So it Kerr does that with the advent of AI, you really can change that and you can make media management and search ability the core of your work because it can be done automatically and just a compound that. Because of the way video is being watched today, which is in all these bite-sized flavors, whether it's on mobile devices or or on laptop screens or on on ipads across that spectrum. The video needs to be sliced up into smaller chunks that it's actually critical to. Have all the descriptive meta data back in the day. When you would just watch the prime time TV shows, you just needed to know the name of your show, and you'd sit down at eight thirty or whatever and watch it for half hour, and you'd be happy. But now everyone wants to I wanna find that, you know, the part of Walking Dead where he tells that story about the, you know, and and they they wanna go directly to the ten minute clip that they wanna see that's only possible. If you have all the descriptive meta data in there with the clips now, what does descriptive metadata main to you? It basically means some combination of stuff I was talking about the transcript who is in the frame what's in the frame. And then there's also stuff that you don't need for like the name of of the episode the name of the series. What aspect ratio was shot in the duration of the clamp all of these things. So all of that. If you add all that up. That's basically the meta data. So do you see this as an enhancement to the deters role a replacement for some of the work the used to do as it stands? Right now, the role of the craft editor is not going away. Right people need shows to look good. And fortunately at the moment, the only good way to do that is with editors I will say that a lot of the routine editing like splicing together back to back highlights. Let's say or putting together simple. Sizzle reels. I mean, something where like someone just wants to see a bunch of stuff back toback. There's evidence that hey, I can start to do those things. But anything that requires judgment is is going to continue to be a person is going to be required in the loop. Thinking about the pow the problem, we've got what we barely have enough room to store the stuff we've already as we start to shift into a I do these additional files take tons of storage space. They actually take relatively little space compared to the video itself because videos already so huge, especially nowadays when it's being shot in four k or eight k typically you might have hundreds of gigabytes video, but only. Hundreds of megabytes of meta data to go with it. So it's it's usually quite a bit smaller than the video itself. And because it makes that video so much more searchable invaluable it's well worth carrying around. Do we need to change the way that we operate to be able to leverage our meta data more meta data, by the way of simply labels for a portion or all of our media is act. Do we need to retrain ourselves to use meta data? I would say so I would argue that the good editors were always doing stuff like this. They might have been putting it as comments in the time line or doing some logging when the stuff was first ingested, nobody likes to rummage, right? If you're working out of feature length though, and there was a ton of footage shot to make up the movie, nobody wants to be the guy who's just rummaging throw that footage to to find the right part. So all this is really an extension of best practices that were already followed. I would say by the most efficient folks in the industry and the other thing that is changing though is that the motive. Storage is shifting because you generally can't do this. If you're driving are just sitting around loose on the shelf right to catalog all material it probably needs to be organized in a central folder structure. Our recommendations you buy some kind of shared storage, but the cost of that is coming way down. And then once you have on the material on a storage pool of some kind then you can start clicking on different. I engines to analyze it and enrich it. So in other words, we gotta move the media that we would normally have undirected attached rates and put it up to a server so that the server can then be accessed by oxo and dealt with in the background while we're busy doing other stuff in the foreground. I think that's right. And that's obviously true, not only of axel. But really of any kind of logical system that's gonna do this stuff. There are ways you could catalog individual drives and put them back on the shelf. But then the minute you need something you'll be going back to the shop to get the drive and that could get pretty tiring. So, you know, back when shared storage was really expensive in required, fiber and a sen- and costs hundreds of thousands of dollars. There were lots of good reasons why a small teams wouldn't wanna do it. But nowadays, you can get a very powerful, but inexpensive Nasr put, you know, a few eight or twelve terabyte hard drives in there and have a very powerful system for not a lot of money. One of the new products you released a product called connector what's the advantage of connector, if we look at it in terms of AI, the big thing is that once people have more meta data, and they have a better global view of of their media. It's only natural that they'd wanna do. Stuff with it and connector makes it possible to visually. Cook up connected workflows where you might say publish something video send somebody in Email FTP, something, you know, these different steps trance coatings another example in the past. These were very arduous to do because you would have to just rinse and repeat manually and connector makes it very easy visually to build automatic versions of those workflows sound for people to learn more about the work that axel AI can do for them. We're gonna go on the web, WWW dot XL, X L, E dot A. I'm Sam Bogota's the CEO of actual thought. Hey, I and sound thanks for joining us this week. Thank you. Thank you. You know, I was just thinking in nineteen sixty six the national commission on technology automation and economic progress wrote in its final report, quote, the basic fact is that technology eliminates jobs, not work close, quote, the technological shifts caused by our unstoppable and have been equated to the changes created by the industrial revolution more than a century ago. In January of this year, the Harvard Business Review wrote, quote, humans have always shifted away from work suitable for machines into other jobs. This is true in the nineteen thirties. When the shift was away from agriculture through the nineteen nineties early two thousands when the shift was largely out of manufacturing. The challenges of automation, extend far beyond media into a discussion of education public policy and society returning to that nineteen sixty six final report of the national commission on technology. They wrote quote, constant job displacement is the price of a dynamic economy history suggests that it is a price worth paying, but the accompanying burdens and benefits should be distributed fairly, and this has not always been the case as the Harvard Business Review continued quote, policymakers should focus on cushioning the necessary transitions following job loss by strengthening the social safety net such as unemployment insurance, Medicaid and a wage insurance program for all displaced workers to help and courage. People to remain attached to the labor force. The challenges automation presents media, creators are as with most new technology, a two edged sword the benefits are that we can work faster or better or accomplish tasks that we could never do before the trade off is that in many cases, there will be fewer of us doing it with a resulting decrease in budgets. We can't stop this change. But we can start preparing ourselves and our companies for this change by continuing our education, strengthening our client relationships and helping clients understand how to take advantage of all this new technology. Hiding our heads in the sand won't make the problem. Go away, the digital production buzz concentrates on media. But that doesn't mean we're ignoring the changes going on in the greater society. Just something I'm thinking about. I want to thank our guests this week Filipacci with lumberjack system Andy Steinbach with paradigm shift. Hey, I Sam Bo gosh with axl. Hey, I and James to Rupo with doddle, news dot com. There's a lot of history in our industry, and it's all posted to our website at digital production buzz dot com here, you'll find thousands of interviews all online and all available to you today. And remember to sign up for our free weekly show newsletter that comes out every Saturday talked with us on Twitter with DP buzz and Facebook at digital production buzz, stop on our theme music is composed by Nathan Dookie Turner with additional music provided by smart sound dot com. Our producer is Debbie price. My name is Larry Jordan and thanks for listening to the digital production buzz. Digital production. Buzz is copyright two thousand eighteen by fellow LLC.

Coming up next