AAI, Utah, Sumed discussed on CBC Radio - Spark
Utah already brought up the aspect of machine learning that we're going to talk about today data. It's what fuels machine learning. The traditionally computing relied on humans, writing rules that computer would then have to follow in machine learning the computer looks at data and creates patterns to help it understand the world. It predicts how things will act in the future and then makes decisions based on those predictions. And the data is the information that the machines us to learn and to to grow their knowledge. But in raw form, the big data sets used to train AAI are bound to contain information that is relevant or that's confusing. So it needs to be refined to be cleaned. One of the things we want to do when we create data sets for machine learning engines is we want to increase the signal and decrease the noise meaning that we want to emphasize the dominant patterns or the the patterns that are going to use to allow the the machines to make useful decisions. But in the pros. Of doing that. We clean out what we call the outliers or the edged data because that makes it more difficult to draw conclusions. The problem is that outlier data isn't actually a relevant. It can represent the experience of real world people. How does that process of cleaning the data and up excluding those who are outside the norm, like people with disabilities? Well, people with disabilities are different and they're usually different from the average. And so they tend to lie at that jagged edge of outlying data. If you take any population and you plot their needs or their characteristics on a three dimensional scatter plot, it looks like an exploding star where you have a bunch of dots in the middle. And then as you get further from the middle, they spread out a lot. So what happens when a big data system or data cleaning engine or a norm engine cleans the day. Data or norms the data. Those individuals are actually taken out of the data set. So we eliminate the noise we limit the outliers. So somebody's experience who's hazardous ability or is otherwise different from the norm is treated as noise. Exactly, right. So what would be some real world affects only this cleaned data? Oh, there's quite a few effects. I don't think we've seem to realize how many places we Hughes machine decisions in say, there's a large firm and they get a thousand applicants for a desirable job. What they'll do is they'll use a machine decision system to filter out those CV's or the resumes that are there to determine what are say, the ten or eleven people they're going to interview. Well, if you are unusual, if you have an unusual resume, which most people with disabilities or people who are in the minority are likely to have, then you're never going to make that list of ten. Are there any examples based on personal experiences of people that you know who have disabilities who encountered issues? Yes, actually encounter them quite a bit, but one example that quite scared me. I had the opportunity to work with a number of automated vehicle projects, so I could play with some automated vehicle simulations that simulate what a car will do give an particular decision. And I brought to this scenario, an example, which is a friend of mine who actually propels his wheelchair by going backwards he has through palsy is his legs. A quite strong is not very controlled, but he can move very, very quickly backwards. And so I ran that scenario and low and behold the car chose not to stop because it is Sumed he was going forward. And so he ran the car simulated car over him. The interesting thing for me has been that a machine learning. And making these phenomena manifest in machines has started to get us to think about how we do research in general and how we address people who are outliers are different. And one of the things that we do is we assign particular characteristics or assumptions to particular group of individuals. And in fact, the greater confidence is about our knowledge of that category. The more difficult it is if you don't, in fact fit the category or if you don't follow those those assumptions or presumptions that people have about you, which you know the promise of this approach to machine learning and the promise of big data.