WEBVTT - The Think 2018 Science Slam Part Two 0:00:04.160 --> 0:00:07.200 Get in tech with technology with tech Stuff from half 0:00:07.240 --> 0:00:14.440 stuff works dot com. Hey, they're welcome to tech Stuff. 0:00:14.480 --> 0:00:17.400 I'm Jonathan Strickland. I'm an executive producer with how Stuff 0:00:17.440 --> 0:00:20.560 Works and I love all things tech and I'm continuing 0:00:20.600 --> 0:00:24.880 the special series coming to you from Las Vegas, Nevada, 0:00:25.120 --> 0:00:31.160 at the IBM Think Conference and UH. In our last episode, 0:00:31.240 --> 0:00:35.680 I talked about attending the IBM Research Science Slam, in 0:00:35.680 --> 0:00:39.920 which several people who have been doing some incredible research 0:00:39.920 --> 0:00:44.240 into different fields and utilizing technology and interesting ways took 0:00:44.280 --> 0:00:47.240 the stage to talk about their work. And it was 0:00:47.280 --> 0:00:50.680 a fantastic night. I really enjoyed myself. I'm very thankful 0:00:50.720 --> 0:00:52.560 that I got to attend. I also got a really 0:00:52.600 --> 0:00:57.320 comfortable seat right up front because I happened to figure 0:00:57.320 --> 0:01:01.440 out where the main doors were before they opened. That's 0:01:01.480 --> 0:01:06.520 that's something I'm really proud of, even though it was 0:01:06.680 --> 0:01:08.760 really anyone could have done it, but I managed to 0:01:08.760 --> 0:01:12.080 get a nice, big, comfy seat, and I chatted a 0:01:12.120 --> 0:01:15.600 little bit about the first presenters. But we've got a 0:01:15.640 --> 0:01:18.560 couple more to talk about today, so we're going to 0:01:18.640 --> 0:01:23.679 transition into that episode and what else I saw while 0:01:23.720 --> 0:01:27.319 I was at the IBM Research science Slam. Hope you enjoy. 0:01:27.440 --> 0:01:33.600 The next presenter was Tom Zimmerman, who uh MS Garcia 0:01:33.840 --> 0:01:36.839 referred to as mcgever, saying that he could take any 0:01:36.880 --> 0:01:39.640 two objects and turn it into kind of a microprocessor. 0:01:40.560 --> 0:01:44.200 He was credited as a human slash machine devices and 0:01:44.280 --> 0:01:48.920 paradigm scientists, which honestly did not know was a thing 0:01:49.240 --> 0:01:52.560 before last night, so this was a treat for me. 0:01:52.880 --> 0:01:58.080 He was very expressive, very funny, probably the most humorous 0:01:58.120 --> 0:02:00.400 of all the presenters who came up there. And he 0:02:00.440 --> 0:02:03.880 talked about how he made sort of a private project 0:02:03.920 --> 0:02:06.360 for himself and how that turned into something that could 0:02:06.440 --> 0:02:09.760 be much greater. And he did this by taking an 0:02:09.800 --> 0:02:12.520 image sensor, essentially the same sort of sensor that you 0:02:12.520 --> 0:02:15.760 could find in a smartphone for the camera. He took 0:02:15.800 --> 0:02:17.480 the image sensor and a couple of l e ed 0:02:17.680 --> 0:02:20.440 s and he put it into sort of a waterproof 0:02:20.480 --> 0:02:26.919 container and created a basic three D microscope. Um. He 0:02:27.000 --> 0:02:31.440 used Python, the programming language, to create a method to 0:02:31.600 --> 0:02:35.000 plot out where things were within a three D space, 0:02:35.120 --> 0:02:37.359 so not just the x y coordinates, but also the 0:02:37.480 --> 0:02:39.520 z coordinates if you think of the x y z 0:02:39.760 --> 0:02:42.320 ax s. He was able to plot all that out, 0:02:42.400 --> 0:02:45.760 so not just how high up in a picture something 0:02:45.840 --> 0:02:47.880 is or how low down in the picture something is, 0:02:47.880 --> 0:02:50.560 but how close or far away that thing is. And 0:02:50.600 --> 0:02:53.600 he used it to look at the life forms within 0:02:53.639 --> 0:02:56.440 a single drop of water. And he found this to 0:02:56.520 --> 0:03:00.280 be really fascinating watching all these tiny, little microscope opic 0:03:00.320 --> 0:03:04.360 life forms moving around and being able to plot where 0:03:04.360 --> 0:03:07.760 they were and track their progress. And this got him 0:03:07.760 --> 0:03:12.160 interested in the subject of plankton. Now, as Mr Zimmerman 0:03:12.240 --> 0:03:18.320 pointed out, plankton are incredibly important to our our world. 0:03:19.040 --> 0:03:24.679 Without plankton, we would find it very, very difficult to exist. 0:03:25.000 --> 0:03:29.040 Plankton produced two thirds of the oxygen that we breathe. 0:03:29.480 --> 0:03:32.800 So you know, plants take carbon dioxide and then they 0:03:32.840 --> 0:03:36.000 convert that into oxygen, and then we breathe that oxygen, 0:03:36.040 --> 0:03:39.320 we exhale carbon dioxide. We're all part of that cycle. Well, 0:03:39.320 --> 0:03:43.680 plankton are responsible for two thirds of that oxygen. So 0:03:44.080 --> 0:03:45.960 while you might think of all the big forests out 0:03:46.000 --> 0:03:48.160 there is being really important carbon sinks, and they are, 0:03:48.400 --> 0:03:51.840 don't get me wrong, we don't want to cut those down. Uh, 0:03:51.920 --> 0:03:56.080 Plankton are even more important. They are huge carbon sinks. 0:03:56.080 --> 0:04:00.440 They sequester carbon from the ecosystem. Which means that they 0:04:00.520 --> 0:04:04.440 also can counteract that that effect. Obviously, if we we 0:04:04.560 --> 0:04:08.960 keep dumping carbon into the ecosystem that contributes to climate change, 0:04:09.680 --> 0:04:14.320 it contributes to the greenhouse effect, which some people would 0:04:14.720 --> 0:04:17.719 say is all about global warming. As it turns out, 0:04:17.880 --> 0:04:21.719 the climate on Earth is more complicated than warming or cooling. 0:04:22.080 --> 0:04:24.360 That's why a lot of people now call it climate 0:04:24.440 --> 0:04:29.520 change rather than global warming. But controlling the amount of 0:04:29.520 --> 0:04:33.400 carbon that we introduced into the ecosystem is incredibly important, 0:04:33.880 --> 0:04:36.440 and plankton are really good at soaking up carbon. But 0:04:36.520 --> 0:04:38.880 here's the problem is that we're actually dumping more carbon 0:04:39.000 --> 0:04:43.000 into the environment than the plankton can easily absorb, and 0:04:43.040 --> 0:04:46.640 plankton are dying as a result, So we're killing off 0:04:47.440 --> 0:04:50.800 the life form that is responsible for producing most of 0:04:50.839 --> 0:04:54.040 the oxygen we breathe. Not only that, but plankton are 0:04:54.080 --> 0:04:58.359 also the bottom of the food chain over in or 0:04:58.760 --> 0:05:01.200 very close to the very bottom the food chain over 0:05:01.240 --> 0:05:03.880 in the oceans, so they serve as a food source 0:05:03.920 --> 0:05:07.360 for just about every species of baby fish out there. 0:05:07.880 --> 0:05:12.000 So if the plankton die off, then the food supply 0:05:12.160 --> 0:05:14.800 for these fish die off, then the fish die off, 0:05:14.800 --> 0:05:17.520 then the predators that eat those fish die off and 0:05:17.560 --> 0:05:21.240 you start to see the food chain collapse in on itself. Obviously, 0:05:21.279 --> 0:05:25.960 this is a really bad thing, but it's also uh 0:05:26.120 --> 0:05:29.720 a tricky thing to study plankton because typically the way 0:05:29.720 --> 0:05:32.960 scientists would study plankton is they would go out into 0:05:33.000 --> 0:05:34.920 the field and by the field, I mean the ocean, 0:05:35.080 --> 0:05:38.320 and they take a big net and they would trawl 0:05:38.680 --> 0:05:41.520 the ocean and they would pull up some plankton. They 0:05:41.560 --> 0:05:44.240 would put this in jars with preservatives which would kill 0:05:44.279 --> 0:05:47.400 the plankton, and then they would come back to the 0:05:47.520 --> 0:05:50.640 lab and they would study the plankton under a microscope. 0:05:51.200 --> 0:05:55.520 Zimmerman equated this to someone who is it's their job 0:05:55.560 --> 0:06:00.480 to UH to evaluate to analyze sports, and they are 0:06:00.480 --> 0:06:04.479 following a football team, and the way they figure out 0:06:04.520 --> 0:06:08.480 how well the football team performs is they are allowed 0:06:08.520 --> 0:06:12.440 to go on the football team's bus after a game 0:06:12.880 --> 0:06:16.120 and take pictures of the football players as they're asleep, 0:06:16.720 --> 0:06:18.760 and then try to analyze how well they play the 0:06:18.760 --> 0:06:21.840 game based upon those pictures of sleeping people. He said, 0:06:22.560 --> 0:06:24.440 that's kind of the equivalent of what scientists are having 0:06:24.440 --> 0:06:26.680 to do with plankton. That if you're only able to 0:06:26.720 --> 0:06:29.240 study them after they've been preserved and therefore they're no 0:06:29.320 --> 0:06:33.120 longer alive. You can only gather so much information about them, 0:06:33.120 --> 0:06:35.360 and it's not terribly useful. It would be better if 0:06:35.360 --> 0:06:38.440 you could study plankton within their own ecosystem and not 0:06:38.520 --> 0:06:42.200 brought back to this this microscope he had been playing with, 0:06:42.240 --> 0:06:46.200 this idea he had created, and he said again that 0:06:46.360 --> 0:06:48.719 the basic parts were all pretty easy to use. You 0:06:48.720 --> 0:06:51.760 could have an image sensor from a smartphone, a couple 0:06:51.800 --> 0:06:56.400 of LEDs, and a waterproof container. You could program some software, 0:06:56.600 --> 0:07:00.159 some artificial intelligence software, use chips that were develop for 0:07:00.200 --> 0:07:04.320 smart cameras that were meant to do things like image recognition, 0:07:04.400 --> 0:07:08.039 face recognition, you know, the sort of basic artificial intelligence 0:07:08.080 --> 0:07:10.760 that I talked about in my preview episode. But then 0:07:10.920 --> 0:07:15.520 reprogram it, retrain the neural network to recognize plankton and 0:07:15.560 --> 0:07:18.880 to track their behaviors. By looking at those behaviors, you 0:07:18.880 --> 0:07:23.120 can learn more about that plankton, like how plankton eat 0:07:23.520 --> 0:07:26.280 other things. And then he gave an example of plankton 0:07:26.560 --> 0:07:31.080 that like a particular type of algae and occasionally this 0:07:31.200 --> 0:07:34.080 plankton will eat a different type of algae that is 0:07:34.120 --> 0:07:38.520 capable of creating a toxin, and that toxin UH more 0:07:38.600 --> 0:07:41.920 or less makes the plankton drunk. As Zimmerman explained, the 0:07:41.960 --> 0:07:45.480 plankton on its own would dart all over the place 0:07:45.520 --> 0:07:48.440 and be able to elude predators because it's able to 0:07:48.440 --> 0:07:51.440 to move around quite a bit. But when it gets drunk, 0:07:51.440 --> 0:07:54.320 when it eats this particular type of algae, it just 0:07:54.400 --> 0:07:57.160 will swib in a straight line. It's kind of how 0:07:57.920 --> 0:08:02.200 UH and intoxicated plankton would move around in its environment. 0:08:02.240 --> 0:08:03.800 But if it moves in the straight line, it makes 0:08:03.840 --> 0:08:07.560 it very easy for predators to eat that plankton. Well. 0:08:07.600 --> 0:08:12.320 As predators eat the plankton, then that plankton is less 0:08:12.320 --> 0:08:16.120 capable of eating algae, you know, obviously, because the population 0:08:16.160 --> 0:08:19.320 of the plankton starts to drop, algae has fewer predators 0:08:19.320 --> 0:08:21.840 of its own, and then its population begins to grow, 0:08:21.880 --> 0:08:23.920 and then you get algae blooms. That could be a 0:08:23.960 --> 0:08:27.880 big problem. So it's better if you're able to monitor 0:08:27.920 --> 0:08:31.080 the plankton and monitor what's happening in the ecosystem and 0:08:31.120 --> 0:08:35.520 be able to perhaps intervene if things are not going well. 0:08:36.480 --> 0:08:38.720 But you can only intervene if you have all the information. 0:08:38.760 --> 0:08:42.839 You can only make a meaningful and helpful act. If 0:08:42.880 --> 0:08:45.880 you know what's going on, without that information, you may 0:08:45.920 --> 0:08:50.679 do more harm than good. So the Zimmerman's point was 0:08:50.720 --> 0:08:53.480 that we now have the capability of making these tools 0:08:53.480 --> 0:08:57.200 to gather the information we need to make more responsible choices. 0:08:57.960 --> 0:09:04.319 And it was a really fascinating way of putting technology 0:09:04.360 --> 0:09:07.360 in the role of an effective tool for a really 0:09:07.400 --> 0:09:10.960 difficult problem. I have more to say about the science 0:09:11.080 --> 0:09:14.840 slam over at the THINK two thousand eighteen conference, but 0:09:14.880 --> 0:09:17.480 before I jump into the next little segment, I'd like 0:09:17.520 --> 0:09:26.760 to take a quick break to thank our sponsor. The 0:09:26.840 --> 0:09:31.920 next person to take the stage was Francesco Rossi. Francesca 0:09:32.000 --> 0:09:35.760 Rossi's area of expertise was something that I thought was 0:09:36.040 --> 0:09:41.560 truly interesting, artificial intelligence ethics. She talked about AI in 0:09:41.600 --> 0:09:43.760 a way that a lot of people at IBM like 0:09:43.880 --> 0:09:47.319 to talk about AI. They don't necessarily talk about artificial intelligence. 0:09:47.320 --> 0:09:51.560 They talk about augmented intelligence. In other words, these are 0:09:51.720 --> 0:09:54.880 the devices and the programs, the software of the firmware 0:09:55.240 --> 0:09:59.600 that help us make decisions. They don't necessarily make all 0:09:59.600 --> 0:10:02.640 the di visions for us. They aren't thinking, they aren't 0:10:02.679 --> 0:10:06.880 having communications with us. They are guiding us as we 0:10:06.960 --> 0:10:09.920 try to make decisions, and then we use that information 0:10:09.960 --> 0:10:13.800 as a tool to help us in our tasks. So, 0:10:14.280 --> 0:10:16.920 how do we build machines to help people make smarter 0:10:17.080 --> 0:10:22.120 and more grounded positions and decisions? Uh, Artificial intelligence can 0:10:22.280 --> 0:10:27.000 help solve some of the world's most difficult problems. And 0:10:27.400 --> 0:10:29.959 Rossie talked about how she's been working in the AI 0:10:30.080 --> 0:10:33.840 field for decades and that the conversation has gradually shifted 0:10:34.040 --> 0:10:38.320 during her time and studies of AI. She said that 0:10:38.559 --> 0:10:42.360 early when she was studying AI, the conversations were all 0:10:42.400 --> 0:10:45.319 about how can we make it smarter? How can we 0:10:45.800 --> 0:10:51.640 make these artificially intelligent programs faster, more capable, Uh, make 0:10:51.720 --> 0:10:56.480 decisions more reliably? How do we do that? And so 0:10:56.559 --> 0:10:59.400 the focus was just on performance. It had nothing to 0:10:59.440 --> 0:11:01.920 do with the quality of those decisions, or maybe the 0:11:01.960 --> 0:11:07.439 impact those decisions might have on other people, but rather, uh, 0:11:07.640 --> 0:11:11.280 just can we make a machine that's able to to 0:11:11.800 --> 0:11:14.840 do this task better than the ones we have right now? 0:11:15.160 --> 0:11:17.600 These days, she said, you know, and back then it 0:11:17.679 --> 0:11:20.520 was just computer scientists who are having this conversation. These days, 0:11:20.520 --> 0:11:23.640 she said, there's a huge number of disciplines that all 0:11:23.640 --> 0:11:26.440 get together to talk about these sort of things that 0:11:26.760 --> 0:11:33.440 not only include computer scientists, but also philosophers, lawyers, economists, policymakers, 0:11:33.760 --> 0:11:36.960 people who have recognized that machines not only have the 0:11:36.960 --> 0:11:40.719 capability of making decisions quickly, but that those decisions can 0:11:40.760 --> 0:11:45.520 have a real effect, positive or negative, on actual human 0:11:45.559 --> 0:11:47.920 beings in the real world, and that there has to 0:11:47.960 --> 0:11:51.079 be some sort of ethical approach to the development of 0:11:51.120 --> 0:11:54.760 AI if we want AI to actually benefit humanity. One 0:11:54.800 --> 0:11:57.800 of the big problems, or several of them, actually, she mentioned, 0:11:57.920 --> 0:12:00.720 transparency is a huge issue. How do you know how 0:12:00.840 --> 0:12:04.640 the AI arrived at its decision? You want a transparent 0:12:05.240 --> 0:12:09.440 way of communicating that. Without that, then you just have 0:12:09.520 --> 0:12:12.640 a black box. You have something that has taken data 0:12:12.920 --> 0:12:15.960 and produced a result and you have no idea how 0:12:16.000 --> 0:12:19.720 it went from A to b uh, And without knowing, 0:12:19.800 --> 0:12:22.360 you don't know if the decision is a good one. Right, So, 0:12:22.720 --> 0:12:26.400 as AI gets more complex and starts to make more 0:12:26.520 --> 0:12:30.600 complicated decisions, if you don't have transparency, it's it's like 0:12:30.640 --> 0:12:34.839 you're consulting a mysterious oracle and you don't really know 0:12:35.240 --> 0:12:38.839 if the oracle has his or her act together or 0:12:38.920 --> 0:12:43.080 is just making stuff up. So transparency is very important. 0:12:43.120 --> 0:12:47.040 Explainability also very important. Can you explain how the machine 0:12:47.480 --> 0:12:50.360 came to its conclusions, not just the pathway it took, 0:12:50.400 --> 0:12:53.679 but how it decided one set of factors was more 0:12:53.720 --> 0:12:56.440 important than another set of factors. And she also talked 0:12:56.440 --> 0:12:59.160 about bias, and in fact, most of her conversation was 0:12:59.200 --> 0:13:04.160 about bias. Bias is prejudice. It could be positive or 0:13:04.280 --> 0:13:07.520 negative in regards to any particular set of data points. 0:13:07.679 --> 0:13:10.920 So bias is something that we humans have. Now, it's 0:13:10.960 --> 0:13:13.959 something that it's a quality we possess. We do get 0:13:14.000 --> 0:13:17.480 a bias four different things. Uh, we could have a 0:13:17.520 --> 0:13:21.120 positive experience with a particular thing. Let's let's take Let's 0:13:21.120 --> 0:13:24.600 take roller coasters for example. Let's say that you ride 0:13:24.600 --> 0:13:26.679 your very first roller coaster when you're a little kid, 0:13:26.760 --> 0:13:29.280 and it's a wonderful trip, it's a wonderful ride, you 0:13:29.320 --> 0:13:31.200 love it, and then you have sort of a bias 0:13:31.240 --> 0:13:33.720 toward roller coasters because you love that feeling you had. 0:13:33.880 --> 0:13:36.480 Or let's say the opposite happened. You ride your first 0:13:36.600 --> 0:13:39.160 roller coaster and it it rattles you around a lot, 0:13:39.679 --> 0:13:42.199 makes you feel sick, and you get off that ride 0:13:42.280 --> 0:13:46.200 and your decision making process tells you, hey, this is 0:13:46.200 --> 0:13:48.719 not for me. Roller coasters are bad. They are not 0:13:48.920 --> 0:13:51.880 well designed rides. They hurt, they make me feel sick. 0:13:52.200 --> 0:13:55.240 I don't like them. They scare me. I'm never writing 0:13:55.280 --> 0:13:57.760 a roller coaster again. You've created a bias based on 0:13:57.800 --> 0:14:00.520 that experience, and may very well be the your bias 0:14:00.600 --> 0:14:03.880 plays out properly. Uh. It could be that that experience 0:14:04.000 --> 0:14:05.719 just tells you that this is how you're going to 0:14:05.800 --> 0:14:08.120 react every single time you can move forward with this 0:14:08.200 --> 0:14:11.319 particular thing. In some cases, that's not a bad thing, 0:14:11.360 --> 0:14:14.720 and she actually Rosie talks about that about how bias 0:14:14.800 --> 0:14:17.960 is not inherently bad. But when it comes to things 0:14:18.000 --> 0:14:22.840 like judging people, then obviously that's much more problematic. If 0:14:22.960 --> 0:14:26.600 you go to a different culture and you encounter something 0:14:26.680 --> 0:14:30.840 that upsets you, you might end up developing a bias 0:14:30.920 --> 0:14:34.120 against anyone who comes from that culture. And that's not 0:14:34.240 --> 0:14:37.880 necessarily representative, it's not fair. It can mean that you 0:14:38.160 --> 0:14:41.640 then treat an entire group of people unfairly based upon 0:14:41.720 --> 0:14:46.120 this bias. And that's where the scary part comes in 0:14:46.200 --> 0:14:49.360 with AI, because while AI is going to follow very 0:14:49.400 --> 0:14:52.760 specific rules that are set out based upon the AIS programming, 0:14:53.320 --> 0:14:55.880 and AI can still be biased. Now, that doesn't mean 0:14:55.880 --> 0:14:59.840 the AI is developing opinions of its own about people. 0:15:00.240 --> 0:15:03.440 It means that the AI is referencing the data sets 0:15:03.440 --> 0:15:06.200 that were fed to it, and data sets are created 0:15:06.240 --> 0:15:09.120 by human beings. If the human beings who create the 0:15:09.240 --> 0:15:14.800 data sets failed to include enough diversity, enough representation in 0:15:14.880 --> 0:15:18.440 that data set, then the people who are not represented 0:15:18.680 --> 0:15:23.040 can be affected negatively. And there are great examples of 0:15:23.080 --> 0:15:24.960 this out in the world that you can actually see 0:15:25.080 --> 0:15:29.240 things that have shown that there are problematic implementations of 0:15:29.320 --> 0:15:34.640 artificial intelligence that do in fact indicate a bias is present. Uh. 0:15:34.680 --> 0:15:40.640 There were stories about facial recognition technology that worked fine 0:15:40.800 --> 0:15:43.320 if you happen to be a white person, but if 0:15:43.320 --> 0:15:47.960 you were of any other race, particularly if you were uh, 0:15:48.000 --> 0:15:51.480 if you were black, then it wasn't working properly. It 0:15:51.600 --> 0:15:55.880 wasn't detecting people properly. Well. That indicates that perhaps the 0:15:56.000 --> 0:15:59.520 data set that was used to train that artificial intelligence 0:15:59.640 --> 0:16:03.560 at a lack of representation of people of different races 0:16:03.560 --> 0:16:06.760 than than just white people. And it's not necessarily that 0:16:06.840 --> 0:16:11.200 it was uh planned that way, or that the people 0:16:11.200 --> 0:16:14.880 who were designing it were specifically excluding an entire group. 0:16:15.160 --> 0:16:18.120 There might have been no malicious intent whatsoever. But that 0:16:18.160 --> 0:16:21.000 doesn't really matter if there was malicious intent from the 0:16:21.040 --> 0:16:25.600 beginning or not. The effect is the same whether it 0:16:25.680 --> 0:16:29.600 was intended to exclude a group or just accidentally excluded 0:16:29.640 --> 0:16:32.320 a group because the person who is designing the system 0:16:32.480 --> 0:16:36.920 didn't belong to that group. That lack of diversity creates 0:16:36.920 --> 0:16:40.640 a bias, and that bias has the potential to negatively 0:16:40.720 --> 0:16:44.840 impact an entire population of people as a result, this 0:16:44.920 --> 0:16:47.120 is not a good thing. You want to have AI 0:16:47.400 --> 0:16:50.520 that is as unbiased as you can possibly be. Now, 0:16:50.640 --> 0:16:54.280 Rosie argues that in the long run, over years and 0:16:54.360 --> 0:16:58.120 years and years, we will have an explosion in AI 0:16:58.360 --> 0:17:04.760 over multiple disciplines, multiple industries, and I completely agree that 0:17:04.920 --> 0:17:08.000 is exactly what we're already seeing it. We're seeing AI 0:17:08.200 --> 0:17:11.040 being developed in all sorts of different ways. And she 0:17:11.119 --> 0:17:13.879 also argues that the ones that will stick around, the 0:17:13.920 --> 0:17:17.160 ones we will rely upon, will ultimately be the ones 0:17:17.200 --> 0:17:19.960 that do not have bias. We will realize that those 0:17:20.040 --> 0:17:24.119 are the ones that are valuable, and we will abandon 0:17:24.280 --> 0:17:27.920 all the AI constructs that contain by us. But that's 0:17:27.960 --> 0:17:31.040 the long run. In the mid term, we're going to 0:17:31.119 --> 0:17:35.680 have problems. We're going to have a I that because 0:17:35.720 --> 0:17:38.280 of a lack of diversity in their data sets, are 0:17:38.320 --> 0:17:40.639 not going to be able to handle real world situations 0:17:40.640 --> 0:17:42.879 that are gonna have real world impact on people. So 0:17:42.920 --> 0:17:46.960 she said, it's absolutely imperative that we have these ethical 0:17:46.960 --> 0:17:52.800 discussions now and start consciously developing AI with an attempt 0:17:52.880 --> 0:17:56.719 to avoid introducing bias. In order to do that, you 0:17:56.760 --> 0:18:04.560 have to create multidisciplinary, multi under multi stakeholder, multicultural teams 0:18:04.600 --> 0:18:07.600 to develop that artificial intelligence. You have to have this 0:18:07.720 --> 0:18:12.200 representation and this diversity from the ground level and then 0:18:12.280 --> 0:18:15.399 build up as you are creating this AI, and only 0:18:15.440 --> 0:18:19.200 then can you be reasonably certain that you have the 0:18:19.280 --> 0:18:24.000 representation you need to avoid bias. At that point, you 0:18:24.040 --> 0:18:27.239 would have an AI that, no matter what it was 0:18:27.359 --> 0:18:30.960 it was intended to do, will be considered much more 0:18:31.000 --> 0:18:36.680 trustworthy and beneficial, not just smart, not just efficient. And 0:18:36.800 --> 0:18:40.080 so I found this to be really a fascinating UH 0:18:40.400 --> 0:18:44.840 presentation again to to think about how our way of 0:18:44.880 --> 0:18:48.640 thinking about AI has changed so dramatically over the last 0:18:48.680 --> 0:18:52.040 couple of decades, and that we've shifted from how can 0:18:52.040 --> 0:18:54.640 we make this machine think too? How can we make 0:18:54.640 --> 0:18:58.080 sure that this machine is performing in a way that 0:18:58.240 --> 0:19:01.639 is not inherently unfair to any particular group of people. 0:19:02.560 --> 0:19:06.639 Um and obviously in today's environment, as we get more 0:19:06.680 --> 0:19:08.239 and more sensitive to this sort of thing, I mean, 0:19:08.280 --> 0:19:11.080 we have whole sections of the world where people are 0:19:11.080 --> 0:19:16.960 becoming more xenophobic and they're becoming more isolationist. They they 0:19:17.000 --> 0:19:20.399 are are banding together with people they identify, with the 0:19:20.440 --> 0:19:24.080 people that they feel represent who they are, and they 0:19:24.080 --> 0:19:28.600 are more readily excluding people who don't fit that group. 0:19:29.480 --> 0:19:33.240 That's a dangerous way of thinking, as a dangerous approach. 0:19:33.240 --> 0:19:36.320 In some cases, it's necessary if you are part of 0:19:36.400 --> 0:19:40.760 a very small population, if you are a minority within 0:19:41.320 --> 0:19:44.920 a population that is the far outnumbers you, then you 0:19:45.000 --> 0:19:48.719 might be banding together with other people of your identity, 0:19:49.080 --> 0:19:52.240 you know that that share these cultural or or ethnic 0:19:52.320 --> 0:19:55.200 identities that you have in a way of protecting yourself, 0:19:55.440 --> 0:19:59.639 which is completely understandable if you are vastly outnumbered by 0:19:59.720 --> 0:20:03.399 other is that's a self preservation technique. But then you 0:20:03.440 --> 0:20:04.919 also have the flip side of it, where you have 0:20:04.960 --> 0:20:07.639 the majority. If they're doing it, then they are more 0:20:07.680 --> 0:20:14.159 likely to create situations that are disadvantageous or oppressive to 0:20:14.280 --> 0:20:18.160 those minorities. And so it's it's really important moving forward, 0:20:18.560 --> 0:20:21.560 that we try to break through that that we try 0:20:21.600 --> 0:20:27.880 to embrace this sort of multicultural, diverse, representative approach so 0:20:27.920 --> 0:20:34.200 that we don't create inherently unfair divisive scenarios, whether it's 0:20:34.240 --> 0:20:39.680 with technology or anything else. Honestly, so, uh, I really 0:20:39.680 --> 0:20:42.760 like this presentation. I have a feeling that not everybody 0:20:42.800 --> 0:20:46.400 would because some people feel very strongly about this sort 0:20:46.440 --> 0:20:50.479 of xenophobic kind of philosophy. They probably don't even think 0:20:50.520 --> 0:20:52.960 of themselves as xenophobic. Um. In fact, I would be 0:20:52.960 --> 0:20:56.000 shocked if they did. But yeah, I thought that this 0:20:56.080 --> 0:20:59.320 was a really valuable talk. We're in the home stretch, 0:20:59.400 --> 0:21:02.720 we're all molost done, but there's still some more to 0:21:02.760 --> 0:21:05.439 talk about that I saw over at this Science Slam, 0:21:05.440 --> 0:21:09.000 this incredible evening of science and technology and just geeking 0:21:09.000 --> 0:21:12.520 out like crazy. Before we conclude, let's take another quick 0:21:12.520 --> 0:21:22.520 break to thank our sponsor. The last person to get 0:21:22.600 --> 0:21:25.960 up and speak was Talia Gershon. She got up to 0:21:25.960 --> 0:21:28.880 talk about quantum computing and AI challenges. So again there's 0:21:28.920 --> 0:21:31.720 some overlap here with some of the previous discussions. She 0:21:31.840 --> 0:21:35.680 was talking specifically about that example I gave earlier with 0:21:36.320 --> 0:21:40.480 miss Garcia, the the polymer chemist and talked about how 0:21:41.119 --> 0:21:43.960 using a computer to accurately simulate the bonding of large 0:21:43.960 --> 0:21:48.040 molecules does grow exponentially as you grow the size of 0:21:48.080 --> 0:21:51.560 the molecule itself. So if you add more atoms to 0:21:51.600 --> 0:21:55.119 a molecule, then the amount of computing power it takes 0:21:55.160 --> 0:22:00.320 to simulate and model that molecule grows dramatically to point 0:22:00.359 --> 0:22:04.159 where a even the most powerful supercomputer would find the 0:22:04.240 --> 0:22:07.920 problem so difficult that it would take ages to create 0:22:07.960 --> 0:22:10.840 a simulation. So even if you were creating a simulation 0:22:10.880 --> 0:22:14.639 that was supposed to represent a micro second of time, 0:22:15.040 --> 0:22:20.000 it might take days or longer weeks to create that simulation. 0:22:20.760 --> 0:22:23.760 So you're taking weeks of real time to simulate a 0:22:23.800 --> 0:22:28.040 micro second of simulated time. Obviously, this is not an 0:22:28.040 --> 0:22:33.639 efficient way to go about things. So quantum computing, she argued, 0:22:33.680 --> 0:22:38.240 could help solve this. And she asked the audience, who 0:22:38.320 --> 0:22:41.640 here has heard buzz about quantum computing? And you know, 0:22:41.680 --> 0:22:44.159 about two thirds of the hands went up in the audience, 0:22:44.160 --> 0:22:45.760 and then she said, who here feels like they have 0:22:45.800 --> 0:22:48.760 a really strong grip on what quantum computing is, And 0:22:48.800 --> 0:22:52.480 then there were maybe a dozen hands still up in 0:22:52.520 --> 0:22:56.000 the audience. I wrote, how quantum computing works, and I 0:22:56.040 --> 0:23:00.040 did not raise my hand because while I did a 0:23:00.119 --> 0:23:04.159 lot of research into quantum computing, I felt like my 0:23:04.320 --> 0:23:07.480 understanding of quantum computing is still in the very, very 0:23:07.560 --> 0:23:12.600 basic level, largely because there comes a point with quantum 0:23:12.600 --> 0:23:17.720 mechanics and quantum computers where my understanding hits a wall, 0:23:18.320 --> 0:23:22.040 and rather than feeling like I really have a grip 0:23:22.200 --> 0:23:26.240 on what is happening, I'm just communicating what smarter people 0:23:26.320 --> 0:23:29.760 are telling me quantum computing is all about. But I don't. 0:23:29.760 --> 0:23:32.600 I feel like I don't have a real grasp of it. However, 0:23:32.640 --> 0:23:35.560 I say that I also remember distinctly when I first 0:23:35.600 --> 0:23:38.960 started studying quantum computing, I was also looking into string theory, 0:23:39.440 --> 0:23:43.000 and I watched a documentary in which a leading physicist, 0:23:43.000 --> 0:23:47.880 a leading expert on string theory, said, I I sometimes 0:23:47.920 --> 0:23:49.760 get asked at the end of the day when it 0:23:49.800 --> 0:23:55.240 all boils down, do I really really understand the science 0:23:55.280 --> 0:23:59.040 I'm talking about? And my answer has to be not really. 0:23:59.080 --> 0:24:03.280 There gets to a point where mathematically I can see 0:24:03.320 --> 0:24:07.040 what's supposed to be happening, but there's a barrier between 0:24:07.080 --> 0:24:12.919 the mathematics and my actual human understanding. And I've found 0:24:13.000 --> 0:24:15.800 some relief in that. Talia Gershawn got up and talked 0:24:15.800 --> 0:24:18.120 about this sort of thing. She's talked about how quantum 0:24:18.160 --> 0:24:22.880 computers encode information into complex quantum states and then they 0:24:23.040 --> 0:24:28.840 run uh quantumized processes on these quantum states. They use 0:24:29.080 --> 0:24:34.080 a method to measure the final state that results as 0:24:34.119 --> 0:24:38.119 a part of these quantized calculations, and then they record 0:24:38.119 --> 0:24:40.920 a result which doesn't necessarily clear it up very much 0:24:40.960 --> 0:24:42.600 for for us. And in fact, she was doing this 0:24:42.680 --> 0:24:45.200 to comedic effects, saying that's if you wanted to say 0:24:45.200 --> 0:24:47.840 it in the basic level, and that's still really complicated. 0:24:48.240 --> 0:24:51.920 She argues that quantum computing is an interdisciplinary problem, that 0:24:52.000 --> 0:24:55.440 it requires lots of people working in lots of fields 0:24:55.440 --> 0:24:58.920