Yes, You Have a Critical Role in Raising Artificial Intelligence.

Jana Eggers Author

02/27/2019 Added

175 Plays

Description

Jana Eggers is a math and computer nerd who took the business path for a career. She’s CEO of Nara Logics and active in customer-inspired innovation, the artificial intelligence (AI) industry, and Autonomy/Mastery/Purpose-style leadership. Her passions are working with teams to define and deliver products customers love, algorithms, and their intelligence, and inspiring teams to do more than they thought possible. In her talk, she will address the ways that AI is already present in our lives, helping us understand what artificial intelligence is, where it’s heading, and why we should embrace it.

Searchable Transcript

Search:

[00:00:00.133](upbeat music)
[00:00:03.370]EMCEE: Welcome to the 30th anniversary year
[00:00:05.071]of the E.N. Thompson Forum on World Issues.
[00:00:09.609]VOICEOVER: Today, you are part of an important conversation
[00:00:12.045]about our shared future.
[00:00:13.780]The E.N. Thompson Forum on World Issues
[00:00:16.783]explores a diversity of viewpoints,
[00:00:18.651]on international and public policy issues,
[00:00:21.488]to promote understanding,
[00:00:23.022]and encourage debate across the university,
[00:00:25.425]and the State of Nebraska.
[00:00:27.427]Since it's inception in 1988,
[00:00:30.263]hundreds of distinguished speakers
[00:00:32.631]have challenged and inspired us,
[00:00:34.634]making this forum one of the preeminent speaker series
[00:00:39.105]in higher education.
[00:00:42.175]It all started when E.N. "Jack" Thompson
[00:00:44.978]imagined a forum on global issues
[00:00:47.680]that would increase Nebraskan's understanding
[00:00:50.183]of cultures and events from around the world.
[00:00:53.420]Jack's perspective was influenced by his travels,
[00:00:56.656]his role in helping to found the United Nations,
[00:00:59.592]and his work
[00:01:00.460]at the Carnegie Endowment for International Peace.
[00:01:04.263]As President of the Cooper Foundation in Lincoln,
[00:01:07.400]Jack pledged substantial funding to the forum,
[00:01:10.603]and the University of Nebraska
[00:01:12.705]and Lied Center for Performing Arts
[00:01:14.541]agreed to co-sponsor.
[00:01:17.177]Later, Jack and his wife, Katie,
[00:01:19.345]created the Thompson Family Fund
[00:01:21.714]to support the forum and all their programs.
[00:01:25.018]Today, major support is provided by the Cooper Foundation,
[00:01:30.623]Lied Center for Performing Arts,
[00:01:32.459]and University of Nebraska Lincoln.
[00:01:35.395]We hope this talk sparks an exciting conversation among you.
[00:01:42.202]And now, on with the show.
[00:01:47.207]MIKE ZELENY: I'm Mike Zeleny with the university,
[00:01:49.242]and I'm pleased to welcome you to this E.N. Thompson Forum
[00:01:51.678]on World Issues.
[00:01:52.946]As you know, for 30 years,
[00:01:54.614]the university, now 150 years young,
[00:01:56.950]and Cooper Foundation,
[00:01:57.917]have partnered with the Lied Center for Performing Arts,
[00:01:59.752]to make this forum possible.
[00:02:01.621]Tonight's lecture features Jana Eggers.
[00:02:04.123]Jana is the CEO of Nara Logics,
[00:02:06.626]a neuroscience-based artificial intelligence company,
[00:02:10.497]headquartered in Cambridge, Massachusetts.
[00:02:12.632]She received her Bachelor's degree
[00:02:14.467]in mathematics and computer science at Hendrix College,
[00:02:18.338]followed by Graduate degrees
[00:02:19.939]at Rensselaer Polytechnic Institute,
[00:02:22.041]and then performed super-computing research
[00:02:24.043]at Los Alamos National Lab.
[00:02:26.546]Jana has held technology and leadership positions
[00:02:28.982]as Intuit, Blackbaud, Lycos, American Airlines,
[00:02:32.886]Saber, Spreadshirt, and other startups.
[00:02:35.822]She has more than 25 years experience in areas
[00:02:38.324]including artificial intelligence,
[00:02:40.226]mass customization, security consultation,
[00:02:43.563]and organizational leadership.
[00:02:45.365]Jana is a frequent speaker, writer, and mentor
[00:02:48.368]on AI and startups.
[00:02:50.537]She is also a marathon runner and an Iron Man.
[00:02:53.106]Tonight she will address what artificial intelligence is,
[00:02:56.109]where it is present in our daily lives,
[00:02:58.144]and why we should embrace it.
[00:02:59.979]Following her remarks,
[00:03:01.247]you'll have an opportunity
[00:03:02.715]to ask Jana questions via Twitter,
[00:03:04.384]using the hashtag ENThompsonForum.
[00:03:07.854]Also, ushers will be in the aisles
[00:03:09.289]to collect your written questions
[00:03:10.723]and bring them to the stage.
[00:03:12.625]The title of tonight's presentation is,
[00:03:14.761]yes, you have a critical role
[00:03:16.362]in raising artificial intelligence.
[00:03:18.464]Now, please join me in a warm, Nebraska welcome,
[00:03:21.067]for Jana Eggers. (applauding)
[00:03:22.969](audience applauding)
[00:03:26.005]Welcome. JANA EGGERS: Thanks a lot.
[00:03:27.273]Thank you.
[00:03:27.540](audience applauding)
[00:03:33.846]Hi.
[00:03:34.714]I wanna thank you all for coming out.
[00:03:36.115]It's a little bit cold outside.
[00:03:37.584](audience laughing)
[00:03:38.651]And there's a bit of snow on the roads.
[00:03:40.954]When I first moved to the Boston area,
[00:03:43.523]I was told, "Oh, you're from Arkansas,
[00:03:46.559]"you have no idea how to drive in the snow!"
[00:03:49.362]And then the first snowfall happened,
[00:03:51.264]and I was like, "You people have no idea
[00:03:52.799]"how to drive in the snow,
[00:03:54.133]"'cause your roads are clear.
[00:03:55.602]"Mine look like that!"
[00:03:56.970]So thank you for welcoming me.
[00:04:00.139]And I feel like I'm at home in Arkansas because of that,
[00:04:04.377]and it didn't bother me a bit,
[00:04:05.678]thought it was lovely.
[00:04:07.113]So I appreciate being here in Lincoln,
[00:04:08.848]and excited to talk to you about helping raise AI.
[00:04:13.720]As Mike mentioned, how do we raise AI?
[00:04:19.692]And a lot of you are probably thinking,
[00:04:21.427]well, what part am I gonna play in that?
[00:04:24.130]And so hopefully this presentation
[00:04:26.165]will give you some ideas of how you can approach that,
[00:04:29.535]even if you're not a mathematician or a computer scientist.
[00:04:33.239]How you can think about AI,
[00:04:35.475]how you can think about the news when you hear it,
[00:04:37.610]the hype that you hear,
[00:04:38.845]and how you can think about it critically.
[00:04:40.680]And, also, how you can talk to people and help us,
[00:04:45.385]because as this says, raising AI is gonna take a community.
[00:04:52.792]When you think about AI,
[00:04:54.193]this is what most people think about, right?
[00:04:57.263]It's what most of my friends that aren't in the field
[00:05:00.566]are saying, "Oh, so you're doing the Terminator?
[00:05:03.670]"What about Skynet?" (audience chuckling)
[00:05:05.838]Right?
[00:05:06.639]That's what I get most of the time.
[00:05:08.775]But in reality,
[00:05:10.877]what I really do is this (chuckling).
[00:05:13.913]It's cobbling pieces together.
[00:05:17.583]We're not quite sure what they are
[00:05:19.952]or how they're gonna look.
[00:05:21.554]It's figuring out how do these pieces work.
[00:05:24.957]And the bigger point is, it's in its infancy.
[00:05:28.695]So it's very young, AI is very young,
[00:05:32.398]and we haven't quite figured out
[00:05:34.067]how all these pieces are gonna work together,
[00:05:36.602]but we're working on it.
[00:05:38.971]What's really important for you to know despite that,
[00:05:43.042]is kids grow up really fast.
[00:05:45.511]You know this.
[00:05:47.113]You're gone for a month,
[00:05:48.948]and then you come back and see your child,
[00:05:50.450]and you can't believe how much they've progressed.
[00:05:53.186]This is what happens all the time with us with AI.
[00:05:57.390]This is from Wait But Why,
[00:05:58.891]which is an awesome site,
[00:06:00.093]if you wanna know what's going on with something.
[00:06:02.662]And I think they're really to show this appropriately,
[00:06:07.800]like, haha, AI, it's not that smart, I can fool it.
[00:06:11.938]You've all played tricks with Siri and Alexa,
[00:06:15.174]and get them to not understand you,
[00:06:17.577]and it doesn't take very long;
[00:06:19.145]heck, it doesn't even take you very long to confuse Google,
[00:06:21.981]who's one of the powerhouses in AI with the search.
[00:06:27.754]But it is progressing rapidly,
[00:06:30.656]and many pieces are coming together,
[00:06:33.659]and all of a sudden we're gonna be like,
[00:06:35.762]"Whoa, wait a second!
[00:06:38.731]"This has gotten out of hand."
[00:06:41.234]And it's only when we get to that point in the curve
[00:06:44.103]that we start getting scared.
[00:06:48.174]And what I'd like to do
[00:06:49.542]is make sure that you guys are ready for that curve,
[00:06:53.079]that you're thinking about it,
[00:06:54.247]that you're engaging early,
[00:06:55.782]in this early stage when we can still make fun of it.
[00:06:58.885]Before we get to that point,
[00:07:01.320]that it's going so fast that we can't get out ahead of it
[00:07:05.658]and start working with it more.
[00:07:08.961]People ask all the time when this is going to happen.
[00:07:15.935]If you look,
[00:07:16.736]I try and kinda keep up with it,
[00:07:17.904]and I look at all the experts
[00:07:19.839]and where they are on this.
[00:07:22.542]And what usually happens,
[00:07:25.978]I'd say that it varies, and it's about,
[00:07:28.381]right now it's running about 2060
[00:07:31.851]for when that we may hit super-intelligence.
[00:07:35.988]And you have plenty of people
[00:07:37.323]that'll say it's gonna be in 10 years,
[00:07:38.691]and you have other people
[00:07:40.193]that'll say it's never gonna happen.
[00:07:42.395]I'm a little bit,
[00:07:44.630]I don't think it's gonna be 2060.
[00:07:47.967]I don't think that it'll be our grandkids,
[00:07:49.936]or their grandkids, honestly.
[00:07:52.338]There's a lot more complexity.
[00:07:54.640]We don't understand much about how our own brain works,
[00:07:57.777]so when we say super-intelligence,
[00:07:59.445]I think we have to understand our own intelligence more
[00:08:02.481]before we can actually say that's where it is.
[00:08:04.517]That doesn't mean that there won't be smart things around,
[00:08:07.687]particularly in narrow fields.
[00:08:10.289]So I think those are some important things to understand
[00:08:13.826]as we think about what might happen with super-intelligence.
[00:08:18.898]But there is something important, which is,
[00:08:22.134]kids do a lot of scary things
[00:08:24.103]before they're very intelligent, right?
[00:08:27.607]Has your kid ever scared you (chuckling)?
[00:08:29.775](audience chuckling)
[00:08:31.110]They do!
[00:08:32.378]So some of the scary things that people see today,
[00:08:34.881]you've probably seen things about deep fakes,
[00:08:38.217]where they can take something that Obama said,
[00:08:41.419]and actually have Putin saying it.
[00:08:44.991]And it's very scary that what we believe is real,
[00:08:50.029]what we can see and heard spoken actually isn't real.
[00:08:55.501]And so that's a scary thing that AI can do now,
[00:08:59.238]and this is pretty easy,
[00:09:00.306]this is pretty widely available.
[00:09:01.941]So it doesn't take much advanced technology.
[00:09:05.011]Many of you as students here could do this,
[00:09:08.147]probably in a night
[00:09:11.183]with the technology that's available now.
[00:09:14.754]You've probably heard things like this about the bias.
[00:09:18.891]This was a big one that came out at the end of last year,
[00:09:24.330]where Amazon had been using a recruiting tool
[00:09:27.166]for a few years,
[00:09:28.634]that was clearly biased against women.
[00:09:31.604]No one trained it that way,
[00:09:33.339]no one said, "Hey, ditch the women,
[00:09:35.708]"'cause the men are more awesome at this."
[00:09:37.777]But it learned from how they hired before,
[00:09:40.212]so it learned from its teachers,
[00:09:42.281]and its teachers were biased.
[00:09:46.218]So these are some of the things that happen.
[00:09:50.823]And this is probably one of the biggest things,
[00:09:53.759]and it I thought this, John Battelle from NewCo,
[00:09:57.396]wrote how does Amazon lose?
[00:09:59.765]And his actual point was,
[00:10:01.567]they actually have a big, big risk
[00:10:04.003]in all of their algorithmic merchandising.
[00:10:07.640]So has anyone here ever been frustrated
[00:10:10.843]by the recommendations that Amazon's given them?
[00:10:13.179](audience laughing)
[00:10:15.748]I hear some laughter, so I think a few people have.
[00:10:18.451]You get the stories of,
[00:10:23.556]why does it continue recommending things to me
[00:10:25.658]that I just bought.
[00:10:27.059]Why doesn't it know that?
[00:10:28.761]One person on Twitter wrote,
[00:10:33.032]hey Amazon, I bought a toilet seat earlier this year,
[00:10:38.537]and I just want you to know I'm not a toilet seat addict.
[00:10:40.840](audience laughing heartily)
[00:10:43.776]I bought one toilet seat,
[00:10:45.378]I'll probably need another one in 10 years,
[00:10:47.947]but you're constantly showing them to me.
[00:10:50.016]And we all laugh at that.
[00:10:52.451]The first reply after that was,
[00:10:54.854]the same thing happened with my mother's funeral urn.
[00:10:57.423](audience gasping)
[00:10:58.891]Yeah.
[00:10:59.792]It's funny, until you get something like that,
[00:11:02.428]and we all go, "Oh."
[00:11:04.530]Amazon has been somewhat public,
[00:11:09.568]they're not completely public about this,
[00:11:11.704]and this is not against Amazon,
[00:11:13.272]it's just understanding that even the people
[00:11:16.142]that are the best at this, have some major limitations.
[00:11:20.946]And Amazon has said publicly
[00:11:24.583]that they have more like tens of algorithms
[00:11:27.720]that are doing this.
[00:11:29.221]I think that's one of the things
[00:11:30.289]that we're gonna have to learn as a community,
[00:11:31.891]that we're probably gonna need hundreds of them.
[00:11:34.026]Ones that can take into account
[00:11:36.028]a lot more contextualization of,
[00:11:38.864]there's probably certain types of products
[00:11:43.069]that we don't want to recommend more.
[00:11:45.805]There's different things we need to take into account,
[00:11:48.874]and one algorithm, one model, isn't going to work.
[00:11:53.212]But then it comes down to how many you manage,
[00:11:55.181]and that's one of the things he said,
[00:11:56.649]they're like, "Wow, we're managing tens of algorithms!"
[00:11:58.851]And I'm like,
[00:12:00.186]"Yeah, we're probably gonna need to be managing
[00:12:02.488]"thousands of them.
[00:12:03.856]"That's something that we need to develop as a technology."
[00:12:06.325]And again, that comes from you.
[00:12:08.928]And I'll give you more tips on what you can do.
[00:12:12.164]But speaking up about this stuff.
[00:12:13.933]I'm glad that people are tweeting about and saying,
[00:12:18.404]"Hey look this is a problem."
[00:12:19.772]Because that awareness,
[00:12:21.273]and the fact that it impacts someone,
[00:12:22.775]like I said, everyone laughs about toilet seats,
[00:12:24.643]but as soon as you say the funeral urn,
[00:12:27.279]everybody feels it.
[00:12:30.015]And so those examples really make a big difference.
[00:12:38.257]Someone tweeted, actually, when we were talking about this,
[00:12:41.427]my coming to this forum,
[00:12:44.163]and they tweeted back to me and said,
[00:12:45.631]I hope you're gonna cover the jobs issue.
[00:12:47.867]So thank you, if you're here, thank you for doing that,
[00:12:49.869]if you're watching online as well.
[00:12:52.371]Kids are taking our jobs, these crazy kids,
[00:12:57.643]'cause they're cheap labor, right?
[00:13:00.713]What I really like, these are two very recent headlines
[00:13:04.984]that I just pulled out for this,
[00:13:07.253]is, you know, are they competing for your job?
[00:13:10.055]And it says, probably, but don't count yourself out.
[00:13:13.125]And what I love is that we've evolved to that,
[00:13:15.528]for, I'd say the last five years,
[00:13:17.830]it's been much more focused on,
[00:13:19.398]oh my God they are going to take our jobs!
[00:13:22.535]And what are we gonna do?
[00:13:23.502]And, oh, the truck drivers!
[00:13:25.237]And I used to work in logistics,
[00:13:26.972]so I know a lot of the truck drivers.
[00:13:28.574]But what I really like is, you know,
[00:13:31.744]here's how we can protect workers,
[00:13:33.379]don't count yourself out.
[00:13:35.147]That's more the message that's starting to get there,
[00:13:37.750]and I think that's a more helpful one for all of us.
[00:13:40.886]So rather than driving the fear,
[00:13:42.254]because actually, as humans, we don't react well to fear,
[00:13:45.558]it makes us shut down,
[00:13:47.326]we go into our lizard brain,
[00:13:49.061]and all of our cognitive processes
[00:13:52.198]aren't working in that case.
[00:13:53.799]So I'm glad that the news
[00:13:56.602]is turning more to this kind of messaging,
[00:13:59.238]and then that's gonna help all of you engage
[00:14:02.341]and think about,
[00:14:03.142]okay, well how can we make sure probably not?
[00:14:06.078]One thing I wanna point out about that is,
[00:14:08.614]and this is a futurist from salesforce.com
[00:14:13.219]that was presenting it at P&G,
[00:14:15.454]and I wanted to give him credit on that,
[00:14:17.289]is he pointed out that in the history of unemployment,
[00:14:22.328]actually major technological shifts that we worried about,
[00:14:25.264]like automated teller machines taking all the teller jobs,
[00:14:27.666]actually didn't,
[00:14:29.368]and we have dramatically increased,
[00:14:32.071]you know we have three times more people to employ now,
[00:14:35.207]and we haven't really,
[00:14:39.044]we've actually increased employment,
[00:14:41.847]even with that many more people.
[00:14:43.649]So I'm not saying I know for sure,
[00:14:45.217]well gosh, look at the history,
[00:14:46.485]it's obvious that that's what's gonna happen;
[00:14:49.121]that's not my point.
[00:14:50.623]My point is more, we don't know what's going to happen,
[00:14:54.360]so don't assume that we know,
[00:14:57.896]and everything's going to go bad.
[00:14:59.932]Let's look at it and say,
[00:15:01.300]"Hey, in the past it hasn't,"
[00:15:02.635]so we have something to work with and some frameworks.
[00:15:05.337]Well, what did happen to the tellers?
[00:15:07.773]We can look at things like that and say,
[00:15:10.009]"How do we do a better job?"
[00:15:11.844]Well, actually, bank employment went up for the most point.
[00:15:15.714]My dad was a banker.
[00:15:16.949]And they did more customer service work.
[00:15:19.985]So there's a lot of things that we can do
[00:15:22.521]to protect ourselves from massive job loss,
[00:15:26.525]but we still have to take action.
[00:15:29.528]The other thing that I wanna say about that is,
[00:15:33.065]we are really not good at predicting the future.
[00:15:37.036]So I don't know, does anyone here use a telephone?
[00:15:39.805](audience laughing)
[00:15:41.807]I mean, because Western Union clearly thought
[00:15:44.576]that they weren't gonna be a good means of communication.
[00:15:47.746]This is, you know, Bill Gates said the spam problem
[00:15:51.817]would be solved in a few years, right?
[00:15:55.954]Watson said there'd be no use for a personal computer.
[00:15:59.425]We are really, really bad,
[00:16:03.562]particularly with tech, at predicting the future.
[00:16:06.965]So remember these examples when you go into it,
[00:16:11.303]when people say things like,
[00:16:13.105]"The whole human race is gonna die."
[00:16:14.873]It's like, oh yeah, well, we've heard these things before.
[00:16:18.310]But how do you deal with that?
[00:16:21.246]'Cause my message is not no, it won't happen.
[00:16:24.416]And if any of you read Good to Great,
[00:16:29.421]you probably know about the story,
[00:16:32.191]which is Admiral Stockdale,
[00:16:33.325]who was a Vice Presidential candidate at one point,
[00:16:35.861]he was a prisoner of war.
[00:16:40.833]So let's assume worst case scenario, right?
[00:16:43.335]The robots take over, and we're prisoners of war;
[00:16:46.338]that's the worst thing that could happen.
[00:16:48.874]How did he survive being a prisoner of war?
[00:16:52.211]He said the pessimists died first,
[00:16:56.014]because they just worked themselves into,
[00:16:59.318]this is all gonna end, this is terrible and it's awful,
[00:17:01.887]and they died first.
[00:17:03.922]The optimists actually died second,
[00:17:06.492]because eventually the optimism broke,
[00:17:09.094]and they died of a broken heart.
[00:17:12.897]It was actually the people that were realists,
[00:17:16.468]real enough to deal with the reality
[00:17:18.804]of what's happening today,
[00:17:22.007]while still believing that they would be rescued.
[00:17:24.943]Now again, I don't wanna say
[00:17:26.444]we're gonna be prisoners of war (chuckling),
[00:17:28.213]that is not my message at all!
[00:17:30.282]But what's interesting about this is,
[00:17:32.384]that's also a key to happiness.
[00:17:34.453]And if you look into some of the work that's been done
[00:17:37.656]in positive psychology,
[00:17:40.325]the people that are actually happiest,
[00:17:42.961]that are the most productive,
[00:17:45.197]are actually people that deal with the reality
[00:17:48.333]of where we are today,
[00:17:50.436]and are optimistic about the future.
[00:17:53.972]So I hope I'm gonna give you some tools
[00:17:56.442]to deal with the reality of AI today,
[00:17:59.478]which is a lot of uncertainty,
[00:18:01.280]and we don't know where it's gonna go,
[00:18:02.915]but be optimistic about where we can take it,
[00:18:05.584]and optimistic that you can have a role in it.
[00:18:09.288]So let's talk about that current status.
[00:18:12.791]So (clapping).
[00:18:14.960]I have an analogy for you.
[00:18:16.962]So people often say,
[00:18:20.232]"Oh well, robots are gonna replace us,"
[00:18:22.835]when they think about artificial intelligence.
[00:18:25.938]And so it took me a long time,
[00:18:27.973]but I thought a long time about,
[00:18:30.476]well what's a good analogy?
[00:18:31.810]And what I'd say is,
[00:18:32.945]rather than thinking about artificial intelligence
[00:18:34.813]replacing a human intelligence,
[00:18:37.249]think about artificial intelligence
[00:18:38.917]just like artificial light.
[00:18:41.086]I think you would all agree
[00:18:42.354]that artificial light hasn't replaced the sun.
[00:18:44.823]Right?
[00:18:46.959]The sun is still very useful
[00:18:48.360]in this traffic circle during the day,
[00:18:50.963]and the artificial light doesn't help out very much.
[00:18:54.366]So what has artificial light done?
[00:18:56.368]Well, for one, it's allowed us to stand here
[00:18:58.403]on a stage like this, right?
[00:19:00.606]It's safer.
[00:19:01.473]They used to use candles, that wasn't safe.
[00:19:04.409]It's allowed us to explore areas
[00:19:07.546]that we couldn't explore before.
[00:19:10.649]But, it's also caused problems.
[00:19:13.519]People that work under artificial light
[00:19:16.488]have higher incidents of cancer.
[00:19:18.991]Think about migration;
[00:19:21.493]I used to live in South Carolina
[00:19:23.095]and we worried about the sea turtle migration.
[00:19:26.331]So it's impacted migratory habits.
[00:19:29.101]So there's a lot of bad impacts
[00:19:30.869]that artificial light has.
[00:19:32.371]So I'd like you to think about,
[00:19:33.605]when you think about artificial intelligence,
[00:19:35.407]stop saying, "It's like us,"
[00:19:38.243]and think more about,
[00:19:40.245]"Wow, this is a new tool that may help and augment us,
[00:19:44.149]"but also can open up new avenues,"
[00:19:47.252]and think about it a little more that way.
[00:19:49.721]Now, I'm gonna take this light analogy
[00:19:51.623]a little bit further, (hands clapping)
[00:19:53.158]because it's the second question that I get asked the most,
[00:19:56.161]which is why is this happening now?
[00:19:59.164]And there is an old New Yorker cartoon
[00:20:02.534]about a drunk underneath a streetlight,
[00:20:05.437]and he's fumbling around, and a policeman comes along.
[00:20:09.541]And the policeman says, "What are you doing?"
[00:20:13.011]And he says, "I'm looking for my keys."
[00:20:15.414]And the policeman says, "Oh, did you lose them here?"
[00:20:18.450]And he said, "No, I lost 'em about two blocks away,
[00:20:20.352]"but the light's better here."
[00:20:21.920](audience laughing)
[00:20:25.123]So, the mathematician, physicist, and computer scientists
[00:20:30.829]like me, are drunk on data and computer power;
[00:20:35.601]this is what's giving us the ability to do the AI
[00:20:38.103]that we always dreamed of.
[00:20:40.405]The stuff that I was working on at Los Alamos
[00:20:42.507]in the early '90's,
[00:20:44.843]is now available to everyone.
[00:20:51.049]It's not locked up in the way that it was before.
[00:20:54.453]So we're excited, and we're over-promising,
[00:20:57.623]and we're under-delivering,
[00:20:59.625]because we're still getting our sea legs here;
[00:21:02.961]we're drinking all that computing power and all that data
[00:21:06.398]and we're excited that, wow,
[00:21:08.433]this is actually working the way we thought!
[00:21:10.802]But we're forgetting to say it's working
[00:21:12.170]in a very, very narrow way.
[00:21:15.107]So what we need is for more people to get involved.
[00:21:18.644]This is why I joined Nara,
[00:21:21.513]is because they weren't doing the same thing
[00:21:23.548]that I was doing out at Los Alamos almost 30 years ago.
[00:21:27.753]They were bringing neuroscience into it.
[00:21:30.155]Like, not the neuroscience we thought in the '40's and '50's
[00:21:33.492]which gave the coining the term of artificial intelligence
[00:21:36.561]and neural net.
[00:21:37.863]But what actually we know, which is a lot more now,
[00:21:39.865]about neuroscience.
[00:21:43.235]It's also bringing in people, product managers,
[00:21:46.004]it's becoming less about the engineering
[00:21:47.773]and the actual algorithms,
[00:21:50.575]and more about what are the results;
[00:21:53.178]and I'll talk a bit more about that.
[00:21:55.947]But we also need ethicists,
[00:21:58.216]we need even lawyers, I love the lawyers that are out there,
[00:22:01.720]but most people don't say you need a lawyer,
[00:22:04.523]or you want to need them.
[00:22:05.857]We need entrepreneurs.
[00:22:06.992]We also need suits.
[00:22:08.593]We need kids, and we need grandmas.
[00:22:10.796]You're getting the picture here, we need you.
[00:22:13.131]We need all of you involved in AI and thinking about AI,
[00:22:16.468]and challenging thinking on AI,
[00:22:18.603]so that we're not the drunks under the streetlights,
[00:22:20.706]looking for our keys that are two blocks away.
[00:22:26.545]So, let me quickly define AI for you.
[00:22:28.680]I'm gonna give you,
[00:22:30.449]you know the magicians' code
[00:22:31.883]is that you never reveal your secrets,
[00:22:33.885]I'm gonna reveal the secrets to you,
[00:22:35.987]which is, the definition of machine learning.
[00:22:40.859]So AI and ML
[00:22:41.927]are used pretty much interchangeably these days.
[00:22:44.763]We've kinda gotten over, like, people would say,
[00:22:46.965]"Oh, that's not AI, that's machine learning,"
[00:22:50.602]or, "That's not machine learning, that's AI."
[00:22:54.172]And people use them interchangeably a bit now.
[00:22:57.109]You'll have some people that would argue with me on that.
[00:23:01.046]But the bigger point is, both of them are really about,
[00:23:05.717]it's algorithms learning from data.
[00:23:08.720]So, it's that simple.
[00:23:11.957]How do they do it?
[00:23:13.291]Oh, some common words, sorry, are, you know,
[00:23:15.727]you'll hear people say supervised and unsupervised.
[00:23:18.463]Supervised is, you're basically tagging the data
[00:23:20.966]so the machine can learn from it,
[00:23:22.567]and then it's just doing a lot of calculations
[00:23:23.902]about what's important.
[00:23:25.403]Unsupervised is you're not telling it anything,
[00:23:27.739]and it's trying to figure it out itself.
[00:23:30.275]And much more of the AI that's done right now,
[00:23:34.546]or the machine learning/AI is done,
[00:23:36.815]is supervised these days.
[00:23:39.785]The other things are different types of algorithm,
[00:23:42.020]that again, I'll talk a little bit more about
[00:23:44.623]a little bit later on.
[00:23:48.193]And then, you know, it's really the next generation
[00:23:50.829]of things that we've been doing for a while;
[00:23:52.798]really, statistics.
[00:23:54.132]It's just, it's a more complex form of statistics.
[00:23:57.335]Overall, the biggest point I want you to know is,
[00:24:00.105]it's just maths, it's different.
[00:24:01.807]And I use maths, plural,
[00:24:02.974]because it really is lots of different fields of math
[00:24:05.777]that are coming together.
[00:24:06.978]It's calculations, it's computing.
[00:24:11.116]This is why computers are so good at it.
[00:24:14.219]But remember that a lot of what we do in our brain
[00:24:16.788]isn't computing.
[00:24:18.089]Our brains are actually bad at computing.
[00:24:20.725]So when I give you that analogy
[00:24:22.327]between natural intelligence and artificial intelligence,
[00:24:25.664]remember, artificial intelligence
[00:24:28.200]is really just about that computing side of it.
[00:24:32.304]So keep that in mind.
[00:24:35.941]Which makes you question things like this.
[00:24:38.710]So this is a reporter, sorry, I'm forgetting now,
[00:24:41.847]I'm pretty sure he's with The Economist.
[00:24:43.915]And he said, "I'm sick of these videos."
[00:24:45.817]And this was a video by Boston Dynamics,
[00:24:47.719]about the robot that could do parkour.
[00:24:51.056]You may have also seen the dog robot that could open doors,
[00:24:54.292]or dance.
[00:24:56.328]You know, you see all of those things.
[00:24:58.697]And so his point was,
[00:24:59.931]I'm really sick of seeing these videos,
[00:25:01.366]because they're garbage.
[00:25:03.802]The robot can't actually do parkour.
[00:25:08.740]And his point is, they're highly scripted.
[00:25:11.243]This isn't AI.
[00:25:13.278]And I totally believe it.
[00:25:14.079]And by the way, the Boston Dynamics people will tell you that.
[00:25:16.615]Are there aspects of it that's AI?
[00:25:19.017]Can that robot see that there's something there?
[00:25:22.287]Yes.
[00:25:23.021]But can it actually judge how it's supposed to jump up,
[00:25:26.091]how it should land?
[00:25:27.492]No, that's all scripted.
[00:25:29.294]There's a lot of cool stuff there, by the way.
[00:25:32.030]That doesn't mean that this isn't cool,
[00:25:33.665]there's a lot of cool stuff in robotics
[00:25:35.000]that has nothing to do with AI,
[00:25:36.768]it's really mechy.
[00:25:38.370]And so that's to take nothing away from that.
[00:25:41.973]But when people see a robot reach out and turn a knob
[00:25:45.810]and open a door,
[00:25:48.580]they think it's just like them,
[00:25:50.348]and it's not.
[00:25:52.183]That's a very, very different thing.
[00:25:55.820]So, someone says,
[00:25:58.056]"Agree, but don't dis them, because they're not garbage.
[00:26:01.226]"This is really cool stuff going on."
[00:26:03.762]But the point is, exactly,
[00:26:06.865]but people need to make that context.
[00:26:09.834]And what I'd say is what you can do
[00:26:11.336]is ask for that context.
[00:26:13.605]You can ask, "What's really going on here?"
[00:26:17.943]You can ask companies, you can demand.
[00:26:22.347]Think of all the microphones that we have now
[00:26:24.182]with the social media,
[00:26:25.216]and think of what's happened with that,
[00:26:27.052]which is exciting.
[00:26:28.553]You have that right to push people,
[00:26:30.789]and I hope that people more go and do this.
[00:26:33.625]I know Boston Dynamics actually,
[00:26:35.327]they're great guys,
[00:26:36.428]this is not a judgment on them,
[00:26:38.296]they did come out and actually explain
[00:26:41.032]how one of the videos was made,
[00:26:43.001]and how much time it took for them to actually do those.
[00:26:47.839]They're all scripted.
[00:26:51.042]It can't walk into a random room, a random hotel,
[00:26:53.979]and open the door and greet you at the door (laughing).
[00:26:59.117]That can't happen right now.
[00:27:01.052]So, where are we good at AI with that?
[00:27:07.359]Yann LeCun who's at Facebook,
[00:27:10.095]and one of the top researchers in AI,
[00:27:12.864]say that we're really good at perception,
[00:27:15.533]but not contextualization or prediction of cause and effect.
[00:27:19.971]So you can imagine as you think about those robots
[00:27:22.007]that are going in and reaching out and opening the door,
[00:27:25.243]they're not able to really contextualize that.
[00:27:28.380]They can actually perceive that a knob there,
[00:27:30.715]they can perceive how much they need to squeeze it or not,
[00:27:33.451]they can perceive that they need to turn it.
[00:27:35.420]But they're not actually thinking through all of that,
[00:27:37.489]they've just been trained to do all of those things.
[00:27:40.291]I think that depends on your measuring stick.
[00:27:44.496]Certainly we can say that some AI can read slides,
[00:27:52.504]so radiology slides, better than humans.
[00:27:55.707]What I would say is,
[00:27:57.475]our measuring stick is wrong on that,
[00:27:59.778]because humans will pick up things
[00:28:01.646]that the AI doesn't pick up.
[00:28:03.048]So while AI may be at 85% and humans are at 75%,
[00:28:07.552]the two together could probably be at 95%.
[00:28:10.588]And we pick up things that the machines don't pick up,
[00:28:15.293]and that's a really important difference.
[00:28:17.362]And it's been shown, it's called centaur in chess,
[00:28:23.268]that humans and computers can almost always beat computers,
[00:28:25.937]humans and computer combined,
[00:28:28.139]can almost always beat computers alone.
[00:28:32.477]And so that's more the measuring stick
[00:28:33.978]that I like to look at,
[00:28:35.313]and it frustrates me every time I see some studies like that
[00:28:39.017]that say, "Oh yeah, they beat the human."
[00:28:40.952]And I'm like,
[00:28:41.853]"Well, did they beat the human and the computer together?
[00:28:43.388]"Then we're talking something exciting."
[00:28:45.290]So, while I agree with this,
[00:28:47.592]that that's the most advanced area
[00:28:49.060]is on the perception side and not on the cognitive side,
[00:28:52.630]I still have issues with even saying,
[00:28:54.365]yeah, check, we're there, on the perception side.
[00:28:58.103]And I'll give you an example of that.
[00:29:02.507]If you can't tell, what those words say is poop.
[00:29:06.144]Does anybody know about this?
[00:29:07.479]Does anybody recognize this picture?
[00:29:09.614]It came from Arkansas, yeah.
[00:29:11.850]It happened in Little Rock, Arkansas,
[00:29:13.551]so I was very proud.
[00:29:14.953]This was called the Poop Apocalypse.
[00:29:18.456]And it was done by a Roomba vacuum cleaner.
[00:29:23.094](audience laughing)
[00:29:23.928]And this was someone's diagram of their living room.
[00:29:27.665]So, sorry, it's gotten a little bit washed out,
[00:29:30.635]but there's, you know, there's a rug in there,
[00:29:32.403]there're bookshelves.
[00:29:33.805]It was in, you know,
[00:29:35.373]it had splattered up onto the bookshelves.
[00:29:37.942]So when we say that computer, or that AI,
[00:29:40.845]is really good at perceiving things,
[00:29:42.847]one of the easiest things for humans to perceive,
[00:29:45.016]I don't know if you guys know this or not,
[00:29:46.885]is poop. (audience laughing)
[00:29:51.256]So that's where I said, you know,
[00:29:52.690]it really depends on the measuring stick that you're using.
[00:29:57.195]There's a lot of ways of detecting poop,
[00:29:59.931]but we still haven't gotten the AI to do that.
[00:30:02.433]And I'm not saying that that's the biggest problem
[00:30:04.169]that we should solve.
[00:30:05.236]But just realize,
[00:30:08.173]I'm not allowed to say the S word on the stage,
[00:30:14.045]computers don't know poop.
[00:30:15.413](audience laughing) (Jana laughing)
[00:30:19.017]So, let's get down to the really important things,
[00:30:22.520]which is how can you get involved.
[00:30:24.255]So hopefully I've given you some context
[00:30:26.057]of what's going on,
[00:30:27.358]given you the status, so you really understand better
[00:30:30.995]about AI and what its knowledge actually is.
[00:30:36.100]How can you get involved?
[00:30:38.269]So, I would say the biggest thing to remember,
[00:30:43.074]is stick to the first principles of AI.
[00:30:46.110]And I call it the chicken, the egg, and the bacon,
[00:30:51.316]because it makes it easier for me to remember.
[00:30:54.586]I call it AI's holy trinity.
[00:30:58.590]The chicken is the algorithm, the eggs are the data.
[00:31:03.094]And the reason why I originally came up with that,
[00:31:05.230]I didn't have the bacon at first,
[00:31:06.397]but everybody needs bacon.
[00:31:08.299]The reason why I came up with that
[00:31:10.435]was 'cause people asked me,
[00:31:11.669]"Which do I start with?
[00:31:13.204]"Do is start with the data and find an algorithm,
[00:31:14.973]"or do I start with the algorithm and find the data?"
[00:31:17.842]And my answer was,
[00:31:19.177]there isn't really a great answer to that,
[00:31:20.979]it's kinda both.
[00:31:23.147]You have to, you're going back and forth between both
[00:31:25.883]on a regular basis.
[00:31:28.152]And so I deemed it the chicken and egg problem.
[00:31:31.022]But then I realized, actually there's a key component there,
[00:31:33.925]is the results.
[00:31:35.260]Bring home the bacon?
[00:31:37.028]I was like, oh, it goes with my chicken and egg thing!
[00:31:40.031]So that's what I'd say is,
[00:31:42.600]this is the whole is greater than the sum of the parts.
[00:31:46.170]So these three things together,
[00:31:47.905]and considering them together,
[00:31:49.540]and considering that when one changes
[00:31:51.276]you need to think about changing the others,
[00:31:53.778]and really, seriously considering that.
[00:31:56.347]So I hope that helps you
[00:31:57.615]as you think about how you get involved in AI,
[00:31:59.751]think about oh wait,
[00:32:01.286]is this a chicken, an egg, or a bacon?
[00:32:03.621]Now, let me teach you a little bit about chickens.
[00:32:06.924]The chicken's the algorithm, remember?
[00:32:08.126]There are many, many breeds of chicken.
[00:32:10.561]This is from Pedro Domingos,
[00:32:13.231]who's out of the University of Washington,
[00:32:15.500]and he wrote a book called The Master Algorithm.
[00:32:18.202]I think it's really good,
[00:32:18.770]I think it's very accessible;
[00:32:20.171]a lot of people I know have read it and said,
[00:32:23.808]"Wow, I feel better, I understand more."
[00:32:27.011]And he breaks it into five tribes.
[00:32:30.848]I think what's cool that's happened since then,
[00:32:33.551]and he kinda talks about this too,
[00:32:35.720]is it's becoming more of an orchestra,
[00:32:38.156]so people are using different tools and combining them,
[00:32:41.793]and they're coming in at different,
[00:32:43.594]the instruments are coming in at different times,
[00:32:47.532]and together at the same time.
[00:32:50.234]But I do think this is a good overview to talk about.
[00:32:54.505]So talk to people about,
[00:32:56.641]"Hey, what algorithm are you using?
[00:32:58.376]"Hey, can you, on this chart,
[00:33:00.878]"can you point to me and show me which one it is?"
[00:33:03.681]And it will help you understand in a pretty basic way,
[00:33:08.086]I think all of these,
[00:33:09.153]if you talk about, you know,
[00:33:12.056]over there on your right hand side,
[00:33:14.726]that's really, for the most part, what Amazon is using.
[00:33:20.598]That's what drives the
[00:33:22.133]people who bought this also bought this, right?
[00:33:25.436]So you can get,
[00:33:26.771]and get an understanding of,
[00:33:28.339]oh, when does that algorithm come into play and talk,
[00:33:31.642]you know, what are examples
[00:33:32.810]of who uses that kind of algorithm?
[00:33:34.979]And the other point that I wanted to make clear
[00:33:40.351]is they aren't all black boxes.
[00:33:42.487]So what's really a black box is the most popular one,
[00:33:45.990]so that's that blue line right here,
[00:33:47.859]that has taken off just recently.
[00:33:49.794]That's why when people say AI is a black box,
[00:33:52.430]that's mostly what they're talking about,
[00:33:55.032]and it's deep learning neural nets.
[00:33:57.201]And I'm gonna show you in a minute
[00:33:59.203]that they're not exactly the black boxes that people think.
[00:34:02.774]But a lot of these other tools that Pedro covers,
[00:34:06.511]genetic algorithms,
[00:34:07.912]I actually switch from using neural nets at Los Alamos
[00:34:10.915]to genetic algorithms,
[00:34:12.183]because of how they worked,
[00:34:14.619]I could actually do what I'm gonna show you next,
[00:34:16.587]which is sensitivity analysis, a little bit more easily.
[00:34:19.524]So that's why they're not as black box
[00:34:21.458]as neural nets, for what I was using it for,
[00:34:24.295]you can argue some of the other way for other things.
[00:34:27.665]Expert systems are really the whitest box of all,
[00:34:31.268]because expert systems,
[00:34:33.237]you're actually hand-coding everything.
[00:34:35.505]That's why they never took off,
[00:34:37.375]is because they take a lot of labor to write,
[00:34:40.478]and also they're very brittle;
[00:34:43.815]so they'll break a lot when you need to change the rules,
[00:34:46.284]it takes a lot of maintenance to do.
[00:34:48.652]Bayesian's kind of in between,
[00:34:52.255]the color actually didn't show up on that one,
[00:34:54.257]but it's the one, the second from your right.
[00:35:00.264]Because Bayesian's kind of the same thing
[00:35:04.602]as an expert system in a way,
[00:35:05.803]but it's based on probabilities;
[00:35:07.438]which way would you go?
[00:35:08.840]And sometimes you don't know what the probability,
[00:35:10.875]or we as humans don't deal with probabilities.
[00:35:12.877]The point is not to get into all of that.
[00:35:15.179]We could talk about, that's a whole talk unto itself.
[00:35:17.748]The bigger point is,
[00:35:19.250]these tools have been around for a long time,
[00:35:21.419]they're actually very well understood.
[00:35:23.187]There's nothing wrong with using one of these,
[00:35:25.256]rather than neural nets.
[00:35:27.492]It's just that neural nets, as you can tell,
[00:35:29.327]is the very popular, en vogue thing to do, right?
[00:35:32.296]So, you know, if you're using one of these others,
[00:35:34.665]you may look like you're wearing plaid pants
[00:35:36.400]with bell bottoms,
[00:35:38.102]but eventually those are gonna come back in.
[00:35:40.471](audience laughing)
[00:35:41.939]So don't worry about it,
[00:35:43.007]'cause they've actually been used quite a bit,
[00:35:44.575]and a lot of people know a lot of things about it.
[00:35:46.677]And just the kids these days
[00:35:47.945]don't quite know as much about them.
[00:35:50.348]But there's even night vision goggles for deep learning.
[00:35:54.785]So I agree, deep learning does a lot of great things,
[00:35:58.289]particularly on that perception side,
[00:36:00.291]it's really, really good.
[00:36:01.826]So it is a great tool to use.
[00:36:03.361]And the bigger thing is,
[00:36:06.230]if you need explainability you can get it,
[00:36:08.533]and that's where I started.
[00:36:10.101]I had the chemists that I was working with
[00:36:12.570]out at Los Alamos,
[00:36:13.404]they wanted to know why when I gave them an answer,
[00:36:15.573]they wanted to know why it was that way.
[00:36:17.408]And basically what I did
[00:36:18.609]is what's called a sensitivity analysis.
[00:36:20.678]And that's what the graph on the left shows,
[00:36:22.947]is what parameters that are going into this
[00:36:26.350]are most sensitive to the input parameters
[00:36:29.520]to change the result.
[00:36:31.389]And so that's what I did.
[00:36:32.456]The problem with that is,
[00:36:33.791]it took me a lot of compute power to get there,
[00:36:36.761]and I didn't know the directionality of the input,
[00:36:41.265]and that's why genetic algorithms actually gave me
[00:36:43.334]a better way of controlling and guiding the input,
[00:36:47.638]and figuring out where I could get
[00:36:48.839]to the optimum answer.
[00:36:50.074]Like I said, these are some pretty complex topics.
[00:36:53.077]You don't have to understand all of that
[00:36:55.279]to ask the right questions and get involved.
[00:36:57.982]There's a lot of, what's great is,
[00:36:59.750]different from when I was doing it,
[00:37:01.819]there's a lot of great tools out there now,
[00:37:04.055]for figuring out where is this sensitive?
[00:37:06.924]You may have heard
[00:37:08.292]about the what's a wolf and what's a dog example.
[00:37:11.762]It was probably about six or seven years ago.
[00:37:16.067]It was out there where it's like, oh, AI is so smart
[00:37:19.804]it knows the difference between a wolf and a dog.
[00:37:21.739]Well then people started playing,
[00:37:23.374]there was a whole paper published on it.
[00:37:24.875]People started playing with it.
[00:37:26.143]And it turns out it was detecting the background,
[00:37:27.812]rather than detecting an actual dog versus a wolf.
[00:37:32.817]And so that's what a sensitivity analysis does.
[00:37:37.154]And this is what it's saying is,
[00:37:39.123]with this tool, it's called Space,
[00:37:41.892]you actually could figure out where those red zones,
[00:37:45.763]and sorry it's a little washed out here.
[00:37:47.531]What in the picture it's actually detecting,
[00:37:50.768]that really identifies this bird versus another bird,
[00:37:55.306]or this animal,
[00:37:56.474]in this case it was a meerkat and a mongoose.
[00:37:59.710]What tells them apart?
[00:38:01.278]That's a sensitivity analysis.
[00:38:03.147]So if you need explainability, you can get it,
[00:38:06.484]and don't let people say you can't.
[00:38:08.953]It may be you need to change the algorithm,
[00:38:12.490]and that's what you need to talk to people about,
[00:38:14.592]and understand that. But just don't let them say,
[00:38:17.094]"Oh, it's a black box."
[00:38:19.664]Another big thing is,
[00:38:21.065]also don't expect the computer to be right.
[00:38:23.901]You might remember in '97,
[00:38:28.272]Garry Kasparov lost to Deep Blue in chess,
[00:38:33.177]and it was a big deal, right?
[00:38:34.178]The Grand Master of chess came down to a computer.
[00:38:37.782]What came out later, in about 2010,
[00:38:40.551]and he's verified this,
[00:38:43.054]is that it was actually a bug that beat him.
[00:38:46.157]And how that bug beat him
[00:38:48.259]is it picked a move at random,
[00:38:50.895]because it didn't know what to do.
[00:38:54.465]So the computer didn't know what to do,
[00:38:56.600]so it picked a move at random,
[00:38:58.502]Kasparov looked at it and said,
[00:38:59.937]"Holy crap it knows something I don't.
[00:39:03.507]"What has it figured out that I don't?"
[00:39:05.843]And his mind was completely preoccupied
[00:39:09.547]with the fact that he believed it saw something
[00:39:13.918]that he didn't see.
[00:39:18.456]They can be wrong.
[00:39:20.324]Even when they're playing at that level,
[00:39:22.426]they can be wrong.
[00:39:25.096]So we've gotta remember that computers have bugs,
[00:39:31.936]and we can't always,
[00:39:33.471]even if they're right a lot,
[00:39:34.672]even if they've won against a lot of chess masters,
[00:39:38.576]we have to remember, (hands slapping together)
[00:39:40.711]that there's time that they're going to be wrong.
[00:39:43.414]So don't get caught up in the,
[00:39:46.217]well, it's been right all this time,
[00:39:47.918]so it must be right now.
[00:39:49.386]If it really doesn't look and feel right,
[00:39:51.222]you know, if Garry Kasparov could have let that go,
[00:39:54.191]and just played his game,
[00:39:56.160]we don't know what would have happened,
[00:39:58.863]but he could have won.
[00:40:00.965]So, the point on the chickens,
[00:40:04.235]is you can question and understand
[00:40:07.204]your chicken's fit to the problem,
[00:40:08.706]and don't be afraid to change or question your chickens,
[00:40:12.276]or your chicken's keepers.
[00:40:15.446]Just because you don't speak the same language,
[00:40:18.215]just because you don't talk in code,
[00:40:21.018]doesn't mean you can't ask some of these basic questions.
[00:40:25.089]I hope I'm giving you some examples to say,
[00:40:27.158]"Yeah, but you know, Garry Kasparov,
[00:40:29.894]"you know what happened there?
[00:40:31.128]"How do we know that this isn't just a random move
[00:40:34.265]"picked by the computer?"
[00:40:37.835]So (hands clapping),
[00:40:42.306]let's talk about the eggs,
[00:40:44.375]and in particular that egg on your face.
[00:40:46.977]Again, I'm not picking on Google, or Microsoft.
[00:40:52.650]They have some of the most,
[00:40:53.784]and are doing some of the most amazing work in AI.
[00:40:57.388]But it's also really important to see
[00:40:59.790]that even people like that,
[00:41:01.292]who have some of the most advanced resources
[00:41:06.063]outside of universities,
[00:41:07.331]'cause I still think universities...
[00:41:09.567]Thank you by the way.
[00:41:11.135]Universities still have the hold
[00:41:13.771]on the really advanced research that's happening,
[00:41:17.174]although these guys are really supporting it,
[00:41:18.943]so it's not to take them down.
[00:41:20.911]They make some really big mistakes.
[00:41:26.116]Identifying black people as gorillas.
[00:41:28.719]That's a big mistake.
[00:41:31.889]Tay being online for only hours,
[00:41:35.292]hours she became a genocidal maniac.
[00:41:39.864]What you might not know is,
[00:41:44.101]because Google said that they'll fix it,
[00:41:46.303]we're taking immediate action
[00:41:47.872]to prevent this result from happening.
[00:41:49.974]You might not know that it's almost three years ago now?
[00:41:55.012]The immediate action that they took
[00:41:58.048]was just stopping the algorithm
[00:41:59.216]from tagging things as gorillas.
[00:42:02.653]They didn't actually go in and fix the problem.
[00:42:08.058]They just kind of stopped it from happening again.
[00:42:12.129]Tay is, and has been, running in China for years
[00:42:17.101]before and after Tay was launched here in the US.
[00:42:21.438]The difference between the US and China
[00:42:25.709]is a really big difference in Internet culture.
[00:42:30.614]There is no 4Chan in China.
[00:42:33.918]There's no subversive element that's trying to go after,
[00:42:40.391](speaking in Chinese) is what it's called,
[00:42:42.393]and it works great.
[00:42:44.962]So what happened is Microsoft got lulled
[00:42:49.033]into a false sense of security, if you will,
[00:42:51.936]because of the success and the importance
[00:42:55.673]of (speaking in Chinese) in China,
[00:42:58.075]thinking that when they brought it over here it'd be fine.
[00:43:01.045]Now, both of these companies,
[00:43:02.279]and again, I'm not,
[00:43:03.580]I can't judge them 'cause I'm not there
[00:43:05.416]dealing with their business problems every day,
[00:43:08.419]but what I will say is if it was really a priority,
[00:43:11.288]both of these people
[00:43:12.890]could have made major advances and fixes.
[00:43:15.392]I will never say
[00:43:16.527]that they could have stopped the bad things from happening
[00:43:18.495]in every incidence.
[00:43:20.230]But there's a lot of training material
[00:43:22.099]to not talk about Hitler, right?
[00:43:25.035]Just like they trained her how to talk,
[00:43:26.770]they can also train her how not to talk.
[00:43:29.273]There's a lot of things that could have been done
[00:43:32.876]that weren't done.
[00:43:34.778]And again, I can't talk about their priorities,
[00:43:36.680]and why they did that.
[00:43:38.382]But as a technologist,
[00:43:39.783]it does bother me,
[00:43:41.952]that these people that have the most power
[00:43:44.355]and the most resources in this technology,
[00:43:46.523]aren't making some of those steps
[00:43:48.625]in being very transparent about it;
[00:43:50.461]they just make statements like this,
[00:43:51.895]saying, "We'll fix the problem,"
[00:43:53.998]and then they go dark.
[00:43:55.766]And I think that's a problem for all of us,
[00:43:57.468]and I think, again, as a community,
[00:43:59.103]we need to demand more.
[00:44:06.677]This is a whole talk, this slide.
[00:44:08.545]Too many of these slides are whole talks unto themselves.
[00:44:10.881]But I will just, the thing that I wanna say about this,
[00:44:14.518]is when I said I didn't know
[00:44:16.453]whether the chicken or egg came first,
[00:44:18.288]it's the egg;
[00:44:19.656]it's the data.
[00:44:21.025]And that's good news for you,
[00:44:22.693]because that's more of what you're familiar with
[00:44:25.095]than the technology.
[00:44:26.430]So learn more about the data, it's more understandable.
[00:44:29.133]Learn about pieces of the data,
[00:44:30.667]and just know that.
[00:44:33.170]And if anybody's interested in this,
[00:44:34.371]I'm happy to come and give the whole talk
[00:44:36.407]about the difference between software development and AI.
[00:44:40.244]This is, if you believe in the lean startup,
[00:44:43.047]usually where you start is the idea.
[00:44:45.182]What I'd say is what we're moving, whoops!
[00:44:48.285]What we're moving into is an era of the data leading.
[00:44:51.922]And so we're gonna start with the data,
[00:44:53.690]and formulate ideas off of that,
[00:44:55.959]so that's just doubling down on that point.
[00:44:57.661]What is the problem with that,
[00:45:02.266]is I was really excited when HBR, a couple of years ago,
[00:45:06.437]came out with this article,
[00:45:07.871]which was, how Unilever built an insights engine.
[00:45:11.975]And I look at that and go, that's an engine,
[00:45:13.677]that's technology.
[00:45:14.912]I like technology.
[00:45:16.046]So I'm very excited and I start reading this.
[00:45:18.248]And I'm like, "This has nothing to do with technology,
[00:45:20.117]"at all."
[00:45:21.919]It's all about their organization
[00:45:23.821]and how they built their organization.
[00:45:27.024]And I read that and I'm like (hand slapping forehead),
[00:45:29.026]"Oh my gosh!
[00:45:30.360]"This is the thing I haven't been talking
[00:45:32.096]"to our customers about."
[00:45:33.697]I've been so excited about our technology
[00:45:35.766]and telling them about our technology,
[00:45:37.935]I didn't bring it back to yeah, but how does that impact?
[00:45:41.171]And here's the important thing about that,
[00:45:42.873]on how it impacts,
[00:45:44.074]is it goes back to my point.
[00:45:45.309]This is why this is under the egg section, right?
[00:45:48.779]You probably heard that AI is eating the world.
[00:45:53.617]You are what you eat, by the way.
[00:45:55.586]So software was eating the world,
[00:45:57.821]and then Marc Andreessen came back and said,
[00:45:59.723]"Now AI is eating the world."
[00:46:01.558]I think both are true.
[00:46:03.093]But AI is fed by data, right?
[00:46:07.197]Well who feeds data?
[00:46:09.032]It's your organization.
[00:46:11.435]So how do you get to a WOW cycle with AI,
[00:46:15.272]is understanding, in my opinion,
[00:46:17.274]understanding this food chain.
[00:46:19.877]Understanding how important your org is
[00:46:22.679]in curating and cultivating that data.
[00:46:26.150]So the key point on the egg side is,
[00:46:28.519]keep your eggs free range,
[00:46:29.920]meaning don't block them off.
[00:46:31.455]A lot of eggs are in silos in your organization.
[00:46:34.391]Keep your eggs free range,
[00:46:36.660]and set your organization
[00:46:38.795]to be the cultivators of those eggs.
[00:46:41.398]Now the best part, the bacon.
[00:46:45.602]You've heard for breakfast,
[00:46:47.471]the chicken is involved but the pig's committed.
[00:46:49.740](audience laughing)
[00:46:52.576]Right?
[00:46:55.012]Well what does that mean?
[00:46:56.580]The pig being committed says,
[00:46:58.315]your AI and the AI you develop and throw out into the world
[00:47:03.120]reflects your values.
[00:47:06.089]That's how committed you are.
[00:47:10.060]You're committed to the results that you're driving.
[00:47:13.864]Are you sure you're driving the right results?
[00:47:17.134]Let me give you some examples.
[00:47:19.703]Again, this happened in Arkansas.
[00:47:23.240]An algorithm cut healthcare and medicare benefits.
[00:47:28.745]And we can go into, like,
[00:47:31.281]this was more statistical analysis
[00:47:34.418]than what most people would call AI,
[00:47:36.853]but AI can do the same thing.
[00:47:38.455]This is recent, 2018.
[00:47:41.024]This is what really bothered me the most
[00:47:44.161]about when this happened is,
[00:47:46.196]this is the engineer that said,
[00:47:47.531]remember we talked about black boxes?
[00:47:49.499]He said, "There's no breast practice for alerting people
[00:47:52.502]"about how an algorithm works."
[00:47:55.906]We duck that all the time.
[00:47:57.107]We're like, "Oh, it's really complicated.
[00:47:59.209]"It's black box."
[00:48:00.744]So there's no way of doing that.
[00:48:02.946]And then it says, he says,
[00:48:05.182]"It's probably something we should do.
[00:48:07.684]"Yeah, I should also dust under my bed."
[00:48:10.187]This is what this engineer said
[00:48:13.423]about taking away people's medicare benefits.
[00:48:17.661]"I should probably also dust under my bed."
[00:48:20.631]That's how trivial that person took this.
[00:48:25.769]Is that the values you want your organization to represent?
[00:48:34.111]Be represented by?
[00:48:36.313]Probably not.
[00:48:38.582]And that's why it's really important that you're in the room
[00:48:41.518]and having these discussion,
[00:48:43.053]you're part of that community,
[00:48:44.921]so that an engineer doesn't just say it's hard.
[00:48:48.892]There's lots of hard problems that we've solved.
[00:48:52.829]Let me give you another example.
[00:48:55.098]The Uber self-driving car,
[00:48:56.500]what you may have heard,
[00:48:57.734]and it killed a woman,
[00:48:59.036]you've probably heard about that.
[00:49:00.537]I don't know if you also heard,
[00:49:02.306]that, I mean, they say right there,
[00:49:06.009]a software bug led to the death.
[00:49:08.278]The problem is, that the sensors, the AI actually...
[00:49:11.315]They say the sensors, but the sensors don't detect,
[00:49:14.484]they just bring in the information,
[00:49:16.486]the AI actually detects, that's that perception thing.
[00:49:21.091]It actually saw the woman.
[00:49:22.759]It actually saw an object even earlier,
[00:49:24.594]it's like, six seconds it identified a bike.
[00:49:27.731]At two seconds before impact it identified her as a person.
[00:49:33.503]Why didn't it stop?
[00:49:37.474]Because they had disabled the feature to say stop,
[00:49:43.046]because it led
[00:49:44.648]to too much jerkiness (hands clapping)
[00:49:46.083]in the car.
[00:49:48.051]So the AI was saying too often, stop.
[00:49:51.988]And so I am quite sure
[00:49:53.590]there was not a conversation that said,
[00:49:55.959]"How much jerkiness would we take for a human life?"
[00:50:00.597]These are really hard conversations to have,
[00:50:04.134]and oftentimes we ignore them,
[00:50:06.370]except in cases where we know we're supposed to have them,
[00:50:09.005]like in medical devices.
[00:50:10.674]So we have areas that we know really well
[00:50:13.076]how to have those conversations,
[00:50:14.811]we're just not used to having them in the broader context.
[00:50:17.647]It was much easier to say to an engineer,
[00:50:19.616]"Hey, let's not brake as often."
[00:50:22.352]Okay, I won't brake as often.
[00:50:25.489]And I can tell you,
[00:50:27.491]I'm not saying the engineers were thinking,
[00:50:29.226]and we'll kill someone, but I won't tell you that.
[00:50:31.695]They weren't thinking that
[00:50:33.296]because they got focused (hands slapping together)
[00:50:34.965]on that result of not braking too often.
[00:50:38.301]We need to have these conversations where we say,
[00:50:40.170]"Okay, I know we want that, but what does that really mean?
[00:50:43.407]"How do we unpeel that?"
[00:50:47.310]So my point here
[00:50:48.145]is this really isn't about artificial intelligence,
[00:50:50.547]it's about human intelligence.
[00:50:52.883]It's about you being in the room,
[00:50:54.718]and you asking the questions,
[00:50:56.753]and not feeling afraid to say,
[00:50:58.722]"Well I don't know anything about AI
[00:50:59.923]"so I can't contribute to this."
[00:51:02.726]You can contribute to it,
[00:51:04.728]and you can challenge the thinking that's happening,
[00:51:07.464]so that we're in a better place together.
[00:51:10.400]So the point here is,
[00:51:11.902]metrics have no moral compass,
[00:51:13.336]that's why we need you.
[00:51:16.006]The metrics of not braking too often is a fine metric;
[00:51:18.975]that's not a moral compass.
[00:51:22.145]We need you in the room.
[00:51:25.248]So, I brought up the whole point about light,
[00:51:30.120]so how do you make that light shine?
[00:51:32.722]Algorithms are introverts.
[00:51:37.694]Those chickens, they're not out there in your face saying,
[00:51:40.464]"This is what I'm doin'!
[00:51:41.798]"This is how I'm workin'!"
[00:51:44.134]Remember that.
[00:51:45.335]Understand their job description,
[00:51:47.204]and make sure you've hired the right algorithm
[00:51:49.206]for the problem you're trying to solve.
[00:51:52.175]Learn from Kasparov.
[00:51:56.680]Remember you need to expect mistakes,
[00:51:58.748]just like kids.
[00:52:00.517]Have your kids made mistakes?
[00:52:03.019]If you were a kid, did you make any mistakes?
[00:52:06.256]Again, think about that,
[00:52:08.959]and don't just think the computer's smarter than us,
[00:52:11.127]because it's not.
[00:52:12.162]It's different from us.
[00:52:13.063]It's artificial light, not natural light.
[00:52:17.634]Start small, this is really where the eggs come in,
[00:52:20.637]so you can understand what's going on,
[00:52:22.506]and that's hard because a lot of AI needs big data.
[00:52:25.909]What's the smallest amount that you can work on?
[00:52:27.878]And by the way, we're improving the algorithms
[00:52:29.513]so they don't need as much big data.
[00:52:31.681]But still carry that big vision;
[00:52:33.049]where do we wanna go?
[00:52:34.684]Start with something you can start to understand,
[00:52:36.753]get your team,
[00:52:37.721]start having those types of conversations,
[00:52:39.923]it's really important.
[00:52:41.491]Results are bacon, focus on those results,
[00:52:43.727]but make sure you know what the results
[00:52:45.529]you're really trying to drive,
[00:52:46.763]and what trade-offs you have in that.
[00:52:48.598]And then, you know, poop happens.
[00:52:51.768]Make sure you're bringing in some expertise.
[00:52:54.804]And some people, whether that expertise is ethicists,
[00:52:58.575]which you may need help having some of these discussions
[00:53:00.343]around ethics,
[00:53:01.378]because they're tough discussions
[00:53:02.746]and they're not ones that we're used to.
[00:53:04.481]It can be technical.
[00:53:06.049]It could be data.
[00:53:07.150]It could be how is this data generated?
[00:53:08.718]Where is it coming from?
[00:53:09.886]Who is the person that knows that the best?
[00:53:12.122]So, expertise is not just technical expertise,
[00:53:14.691]it's expertise in all of these areas
[00:53:16.459]that I've been thinking about.
[00:53:18.461]So where I'm gonna end is,
[00:53:20.397]this is the other big question.
[00:53:22.966]So hopefully that helps you and gives you some things to do.
[00:53:25.368]But I wanna answer the other big question;
[00:53:27.938]so sorry if somebody asked it already,
[00:53:30.140]is, you know, the question is normally
[00:53:33.243]what do you worry about?
[00:53:35.478]I'm much more where Stockdale is.
[00:53:40.383]I'm not an optimist.
[00:53:43.420]But I'm very much a realist.
[00:53:45.355]And the reality is we have so much to gain from AI.
[00:53:49.826]I really think AI is gonna be the cure to us solving cancer.
[00:53:53.797]I am not gonna trade that off for some of the risks.
[00:53:56.866]So I'm gonna deal with my current reality,
[00:53:58.902]that there's risks that are involved,
[00:54:01.004]while I am pushing to solve some of those bigger problems
[00:54:05.842]that I know can't be solved without AI.
[00:54:09.112]Fear is pessimistic.
[00:54:11.448]I tend to focus on the reality,
[00:54:13.650]which is more sunlight.
[00:54:17.454]So sunlight is not only just seeing what's happening there,
[00:54:20.757]sunlight's also a great disinfectant (chuckling).
[00:54:24.127]So this is that transparency point,
[00:54:25.862]this is having these open conversation,
[00:54:27.697]that's what's important.
[00:54:29.099]More patience.
[00:54:30.467]We're not to the hype,
[00:54:32.435]so be patient and let the kid fall and make a mistake,
[00:54:36.406]but put boundaries around that,
[00:54:38.141]just like we do with our kids.
[00:54:40.043]More diversity;
[00:54:41.111]we need more people involved,
[00:54:42.178]it can't just be the drunk mathematicians
[00:54:44.848]and computer scientists that I talked about, right?
[00:54:47.417]We need you coming in and helping us,
[00:54:50.220]and asking the question.
[00:54:51.421]Which means it needs more of you.
[00:54:53.356]So that's what I think about when I think about AI.
[00:54:55.592]I worry about,
[00:54:56.926]how do I get people like you?
[00:54:58.528]That's why I'm here tonight,
[00:55:00.330]is this is not about my business,
[00:55:02.132]this is about me wanting to bring more people in AI,
[00:55:06.469]because I know that's the only way we're gonna be stronger.
[00:55:10.407]So, that's it. (audience applauding)
[00:55:12.509]And I look forward to questions.
[00:55:13.410](audience applauding)
[00:55:23.520]MIKE ZELENY: Thank you so much Jana.
[00:55:24.821]At this time, Jana will take questions from you.
[00:55:27.390]You may submit questions on Twitter
[00:55:29.426]using the hashtag ENThompsonForum,
[00:55:31.494]or simply write your questions on the note cards
[00:55:33.763]provided by the ushers.
[00:55:35.365]Jana, we've got an active Twitter feed already tonight.
[00:55:38.334]Let's start with a gentleman whose question you answered
[00:55:41.037]about the workforce,
[00:55:42.572]and he appreciated you responding to that.
[00:55:44.541]His follow up question is,
[00:55:47.043]can you address the role universities and community colleges
[00:55:49.612]will need to fill to provide affordable workers,
[00:55:53.149]retraining for those displaced by AI?
[00:55:55.385]A couple questions around that topic tonight.
[00:55:57.253]JANA EGGERS: Yeah.
[00:55:58.121]I think that's a really good point.
[00:55:59.823]One of the things that I say...
[00:56:01.658]So, as I mentioned, I was in logistics,
[00:56:03.426]so I actually know a lot about truckers
[00:56:07.130]and truckers themselves.
[00:56:08.965]And they're one of the people that are talked about
[00:56:10.900]as being displaced.
[00:56:12.168]And what I'd say is these are some people
[00:56:17.040]that most of them did not wake up as a kid and say,
[00:56:20.810]"I wanna grow up to be a trucker."
[00:56:22.645]They did it because it's a great living to be make,
[00:56:27.350]and that's our societal values;
[00:56:29.486]we value truckers.
[00:56:31.554]And we can value other things.
[00:56:33.790]So I think it's gonna require a shift in values,
[00:56:36.860]and the universities and community colleges
[00:56:41.297]can help us respond to those shifts.
[00:56:44.267]And so as we as a society shift,
[00:56:46.102]them being aware, hey, what's going on.
[00:56:48.605]I do think there is a big shift going on
[00:56:50.373]with eldercare,
[00:56:52.842]because of the population shift,
[00:56:54.677]so that's an example of community colleges and universities
[00:56:58.681]really doing, expanding their trainings in eldercare,
[00:57:03.319]and understanding of eldercare.
[00:57:04.921]MIKE: All right. JANA: Just one example.
[00:57:06.356]MIKE: Thank you.
[00:57:07.557]We've got an audience member here at the Lied Center
[00:57:09.492]asking for more examples of where AI is being used today,
[00:57:13.329]and where should it be used,
[00:57:14.864]maybe other than eldercare.
[00:57:17.400]JANA: So AI is being used,
[00:57:21.137]I mean, you would,
[00:57:23.373]I don't know would you be shocked?
[00:57:25.308]I don't know, (hands slapping legs)
[00:57:26.309]depends on who you are whether you'd be shocked.
[00:57:27.744]It's used in so much that's happening today.
[00:57:30.980]Do you use Google search?
[00:57:34.284]There's tons of AI happening there.
[00:57:36.286]Everything with vision.
[00:57:37.554]You know, we were talking before, offstage,
[00:57:40.490]about the use in agriculture,
[00:57:43.893]for weed detection, for harvesting,
[00:57:46.262]for autonomous vehicles driving around the fields.
[00:57:49.432]Monitoring, it's being used for monitoring all the sensors;
[00:57:55.738]IOT has driven a lot...
[00:57:57.240]Internet of Things has driven a lot in the AI space.
[00:58:00.844]So it's really ubiquitous these days
[00:58:03.980]and being used all over.
[00:58:06.015]You know, most people just don't realize
[00:58:08.551]that that's happening,
[00:58:10.186]because it's not in their face.
[00:58:11.187]And that's really what's gonna happen with AI is,
[00:58:14.724]it's not the physical presence here,
[00:58:17.927]it's the computation that's going on behind the scenes.
[00:58:22.098]As far as where should it be,
[00:58:23.633]you know, the joke I always make is (chuckling),
[00:58:25.668]I'd just like it to manage my calendar.
[00:58:27.937]So I hope that some more things
[00:58:31.241]are coming out in those areas,
[00:58:33.476]though realizing that's a really complex problem.
[00:58:36.246]So I do think, as I mentioned in the talk,
[00:58:40.016]that healthcare is a huge area,
[00:58:42.318]and I'm really glad we do some work in there,
[00:58:44.020]but there's also a lot of really great work going on there.
[00:58:47.023]MIKE: All right, thank you.
[00:58:48.091]Also from our Twitter feed this evening, Jana.
[00:58:50.293]How may net neutrality or similar restrictions
[00:58:53.229]hinder AI to the public in the future?
[00:58:56.499]JANA: I wouldn't say net neutrality.
[00:58:59.135]Net neutrality has a lot of other issues,
[00:59:01.337]I don't think that it really impacts AI that much.
[00:59:04.507]GDPR has a bigger impact on AI,
[00:59:10.546]just because of,
[00:59:11.581]and I don't think that's a bad thing,
[00:59:13.383]I'm actually very pro-GDPR.
[00:59:15.418]I think it's gonna teach us how to be more responsible
[00:59:19.055]with the use of data,
[00:59:20.089]and we should, that's not a bad thing,
[00:59:21.824]that's a good thing.
[00:59:23.393]Teaching us how to do that is absolutely good.
[00:59:27.030]I do think that, you know, the things that are gonna hinder
[00:59:31.100]can be some regulations on algorithms
[00:59:36.406]and what they can be used for.
[00:59:39.575]So I worry about regulations, but I'm not anti-regulation.
[00:59:43.246]I think we need more people that are doing that work,
[00:59:46.883]to understand AI and the implications,
[00:59:51.387]so that they can write better regulations around it.
[00:59:54.357]MIKE: Great, thanks.
[00:59:55.792]Do you think AI will ever become truly sentient?
[00:59:58.895]If so, how do we deal with that?
[01:00:01.230]JANA: I don't.
[01:00:03.766]I'll just be honest.
[01:00:05.301]I'm not in that camp.
[01:00:08.104]I mean, it's so far off that I just can't even predict it.
[01:00:12.608]But again, you know my analogy of natural light
[01:00:15.611]versus artificial light.
[01:00:16.779]That doesn't mean that it's not more advanced than us
[01:00:20.416]in some areas.
[01:00:22.418]As far as what do we do about it?
[01:00:24.620]You know, again, I'll go back to being on team Stockdale.
[01:00:28.024]I think we just deal with the present reality.
[01:00:30.927]I don't think it's gonna be a prison camp, like I said.
[01:00:33.997]But if it did happen to be that horrible,
[01:00:36.632]and the sentient beings decided not to take care of us,
[01:00:41.037]I mean, Fairabee said,
[01:00:42.972]"If there was super-intelligence,
[01:00:44.574]"they would have contacted us already."
[01:00:46.342](laughing) So, you know.
[01:00:49.579]But other people would say,
[01:00:51.481]"Oh, they have contacted us, and they left."
[01:00:53.449](audience laughing) (Jana laughing)
[01:00:55.585]So I just, I don't think that it's gonna be
[01:00:59.722]in any kind of form that we can predict,
[01:01:01.858]and that's where it's not a conversation
[01:01:05.461]that I think about a lot.
[01:01:07.830]MIKE: That's fair.
[01:01:09.198]Could you say more about starting with data and algorithms
[01:01:11.634]instead of ideas,
[01:01:13.102]in your lean startup example?
[01:01:15.571]JANA: So I think the data can bring you a lot of ideas,
[01:01:18.875]and AI can help you analyze that data in new ways.
[01:01:22.578]So the way data connects...
[01:01:26.449]It's a really long answer so I'm gonna try and shorten it.
[01:01:29.052]We build automatically with AI a knowledge graph of data.
[01:01:34.023]People, our customers, have learned thing from their data
[01:01:37.860]that they didn't know before,
[01:01:39.495]that they couldn't get with general statistical analysis,
[01:01:42.265]or they may have, but again,
[01:01:43.633]it's kinda like doing sensitivity analysis on neural nets
[01:01:46.869]rather than using genetic algorithms.
[01:01:51.040]So, I think the bigger point there is,
[01:01:54.243]what used to happen is we'd have an idea
[01:01:56.746]and then we'd start coding it,
[01:01:57.947]we'd write a minimum viable product,
[01:01:59.515]and then we'd launch that,
[01:02:00.983]and we'd get some data from people using that,
[01:02:03.453]and then generate.
[01:02:04.587]I think now we're gonna look at the data
[01:02:06.789]of what exhaust has happened,
[01:02:08.357]whether that's exhaust
[01:02:09.492]is coming from the Internet of Things,
[01:02:12.061]or whether it's coming from website;
[01:02:15.698]there's lots of data floating around out there.
[01:02:18.601]And so I think our ideas
[01:02:19.735]are more gonna come from looking at that data holistically,
[01:02:23.940]of I have this maintenance report here,
[01:02:25.942]I have this stream of IOT data here,
[01:02:29.045]I have an inspector that noticed something here.
[01:02:31.447]I think that merging those streams of data
[01:02:34.217]is where we're going to get more ideas.
[01:02:36.586]And then the algorithms again,
[01:02:38.054]like I said, my thing is,
[01:02:39.355]I really think it's the egg first,
[01:02:41.290]rather than the chicken,
[01:02:43.059]and the chickens will follow the eggs.
[01:02:45.328]MIKE: Excellent.
[01:02:46.762]And the bacon of, okay.
[01:02:47.630]JANA: And the bacon.
[01:02:49.365]MIKE: In what ways--
[01:02:50.099]JANA: Someone was listening!
[01:02:51.334]MIKE: Pig farmer, yeah.
[01:02:54.137]So some questions from our audience
[01:02:55.771]here at the Lied Center tonight.
[01:02:57.507]In what ways might we hold developers accountable
[01:02:59.976]for the job descriptions of algorithms?
[01:03:02.345]JANA: I think that's, it's a good question.
[01:03:05.114]I am not one that believes, as I said,
[01:03:08.584]raising AI takes a village.
[01:03:11.087]I don't think that we can blame it all
[01:03:13.055]on one of the villagers.
[01:03:14.524]If you are blaming the engineer,
[01:03:16.893]I would point back and blame you
[01:03:19.028]for not getting involved earlier.
[01:03:20.763]I would say that it's still realized
[01:03:24.567]that human tendencies and generalizations,
[01:03:28.404]what they are.
[01:03:29.705]Engineers tend to be more introverts.
[01:03:33.242]That doesn't mean
[01:03:34.810]that you can just let them be their introvert self
[01:03:37.013]and go off in a corner,
[01:03:38.014]it's up to you to engage them;
[01:03:40.349]engage us.
[01:03:43.452]Despite the fact that I stand up here and do this,
[01:03:47.156]there's a wall right here.
[01:03:48.791]You guys don't know,
[01:03:49.959]I will also go home and curl up in my little ball
[01:03:52.395]and have my introvert recharge.
[01:03:54.630]I do this 'cause I think it's really needed,
[01:04:00.303]and it's fun to explain things to people
[01:04:03.806]and have them say,
[01:04:04.874]"Wow, I feel like I understand something now."
[01:04:08.211]So I think that's a problem if we're blaming the engineer,
[01:04:11.314]but it doesn't mean that engineers
[01:04:12.648]don't have responsibility.
[01:04:14.517]MIKE: All right, that's fair.
[01:04:15.785]We'll perhaps broaden from AI a bit.
[01:04:18.120]Someone here in the audience would like you to expound on
[01:04:21.290]what you meant earlier
[01:04:22.458]about universities having a hold on research.
[01:04:25.027]JANA: Oh, I didn't mean hold, sorry if that came across wrong.
[01:04:28.497]I think you guys are still the epicenter of research,
[01:04:33.402]which I think is great,
[01:04:34.337]and it should be that way.
[01:04:37.106]I don't think that corporations are,
[01:04:40.409]they have a different purpose.
[01:04:43.946]And so I don't really feel like corporations
[01:04:47.717]should be the ones that are the primary drivers of research,
[01:04:50.686]and it actually makes me nervous,
[01:04:52.288]because I think they're constrained by other things,
[01:04:56.259]and they have different motivations than you do
[01:05:00.129]when you're doing more pure, scientific research.
[01:05:02.465]And that's doesn't mean that they shouldn't do any,
[01:05:04.333]but it does bother me when a lot of, like,
[01:05:09.972]the AI researchers are being sucked into large corporations.
[01:05:13.042]I think there is a place, place like Los Alamos,
[01:05:17.580]for pure scientific research,
[01:05:19.415]and there is a place for corporate-esque research,
[01:05:24.754]but those are also two different things,
[01:05:26.422]and we shouldn't expect corporations to do the research
[01:05:30.726]that universities do.
[01:05:33.696]So I didn't mean like, you have a hold on it
[01:05:35.531]and you won't give it up.
[01:05:36.799]But you have an advantage and you should use that advantage,
[01:05:41.437]and we as society should value that advantage.
[01:05:45.474]MIKE: Okay, you studied math
[01:05:46.709]and computer sciences as an undergraduate.
[01:05:48.878]So besides math and computer science,
[01:05:50.579]what areas of study in college
[01:05:52.548]will best prepare today's young people
[01:05:54.250]for working in an AI world?
[01:05:56.585]JANA: Definitely business,
[01:05:58.220]because business is the results part of the equation.
[01:06:01.390]Also there's a lot that needs to be done with design,
[01:06:05.194]particularly with AI, it's really changing.
[01:06:07.697]You know, when I was growing up in the field,
[01:06:10.733]there was no thing called user experience,
[01:06:15.004]it was all what color is the button,
[01:06:17.907]so that people notice it.
[01:06:21.143]It was really just about the button design,
[01:06:24.580]and so you're really talking about graphic designers.
[01:06:27.717]That became a very different field with user experience.
[01:06:31.954]So it went from graphic design to user interface
[01:06:35.925]to user experience.
[01:06:37.860]And those are really different things.
[01:06:39.395]And I do think we need more work around,
[01:06:43.065]what does it mean to design with AI?
[01:06:45.568]So that's another field that involves more of the arts.
[01:06:49.672]I also think ethics, and someone asked me earlier today
[01:06:53.676]about who's driving the policy.
[01:06:55.511]I mean, people that can understand the technology enough
[01:07:00.883]to really drive the policy and have that passion,
[01:07:03.152]you know, they need to spend some time
[01:07:05.054]understanding the technology,
[01:07:06.622]but it's understandable enough.
[01:07:09.592]And so people, some people doing that as well.
[01:07:12.661]So I think there's a lot of areas.
[01:07:14.663]I guess the basic point, I'd say,
[01:07:16.832]is technology is gonna impact everything.
[01:07:19.101]So being comfortable with technology,
[01:07:21.437]you can leverage that in any field
[01:07:25.808]that you want to go to, almost.
[01:07:28.310]MIKE: Great, okay.
[01:07:29.512]So much for not going political tonight, but.
[01:07:31.614]Can you give us an example of algorithms
[01:07:33.349]used for election outcomes?
[01:07:35.050]Perhaps Brexit?
[01:07:36.752]JANA: I mean, fake news is the biggest one, right?
[01:07:40.156]And people were generating fake news with algorithms,
[01:07:43.492]for knowing what people would react to,
[01:07:46.529]and then share more.
[01:07:48.130]So AI did have, unfortunately, had a part in that.
[01:07:52.601]It also was helped out,
[01:07:54.804]it was kinda like a, you know,
[01:07:57.973]it didn't mean to help,
[01:08:00.743]but the Facebook algorithms.
[01:08:02.978]It wasn't that they were creating the fake news,
[01:08:05.948]but they were creating the environment
[01:08:08.350]to where it would be shared more,
[01:08:09.585]because again, that was their profit
[01:08:11.487]that they were trying to do.
[01:08:12.421]You know, Facebook didn't generate their algorithm
[01:08:16.292]to spread fake news,
[01:08:17.693]but it ended up spreading
[01:08:18.961]because their algorithm was all about
[01:08:20.629]what makes people interact with this;
[01:08:22.965]what makes people share it?
[01:08:24.633]So sadly, AI did have a role in propagating it,
[01:08:29.371]and it also was, helped write and support it too.
[01:08:34.643]MIKE: Okay, from our Twitter feed tonight, Jana.
[01:08:36.479]Will AI be proprietary or controlled by big business?
[01:08:41.149]JANA: I think it's really hard.
[01:08:43.519]I believe that it's gonna be hard to control it right now.
[01:08:48.491]And going back to my point about,
[01:08:50.926]yeah, Google and Facebook and Microsoft,
[01:08:55.965]and all the big guys are doing a lot of AI research,
[01:08:58.734]but they don't have a hold on it.
[01:09:00.269]There are still plenty of universities
[01:09:03.072]who are trying to train people in AI.
[01:09:05.140]So I don't think that we have to worry about,
[01:09:09.411]oh, three people are going to own
[01:09:11.779]all of the AI in the world.
[01:09:13.916]I think it's way too far out there.
[01:09:16.118]What used to happen is,
[01:09:17.386]only people like a Los Alamos,
[01:09:20.890]big research institutions,
[01:09:22.957]had the compute power and the data.
[01:09:24.760]But now that that's free,
[01:09:26.761]I think that's really...
[01:09:29.098]And by free I don't mean it doesn't cost anything,
[01:09:31.399]but it's out there, it's available.
[01:09:33.269]I think it's gonna be really hard for anyone to really,
[01:09:36.971]including any nation, I get asked that a lot,
[01:09:39.608]you know, are we gonna lose to China?
[01:09:44.279]Might we?
[01:09:45.014]Yes. But is it a forgone conclusion?
[01:09:47.316]No.
[01:09:48.417]And so I don't think
[01:09:51.120]that there's gonna be that kind of control in any one area,
[01:09:54.623]unless we really screw up regulation.
[01:09:57.159]MIKE: Okay.
[01:09:59.028]Is your contrast between artificial intelligence
[01:10:00.896]and human intelligence
[01:10:02.231]also applicable to the condition of ethics?
[01:10:04.400]Do you think there's a link
[01:10:05.568]between the intelligence and ethics of both sides?
[01:10:08.304](Jana laughing)
[01:10:12.641]JANA: Yeah, there probably, wow.
[01:10:15.044]I've never been asked that question in that way,
[01:10:17.446]and there, you know,
[01:10:20.349]I went to the whole, like, we don't understand ethics,
[01:10:22.885]which is some of what I was saying is,
[01:10:26.422]we're not good at having those conversations.
[01:10:28.958]We feel like our ethics, we understand,
[01:10:32.795]and they're kind of hard lines to us.
[01:10:35.164]But the person next to us may feel differently.
[01:10:38.267]And so we don't often have those conversations.
[01:10:41.203]So yeah, I would say they're very highly linked,
[01:10:44.540]because we're not good
[01:10:45.841]about talking about just our own intelligence,
[01:10:47.710]and what we would do in an ethical situation,
[01:10:50.646]because we feel like that's taboo.
[01:10:53.048]And so therefore we don't do it with machines.
[01:10:56.685]And maybe, actually, I hadn't thought about it that way,
[01:10:59.221]but it could be kinda cool in that,
[01:11:02.124]maybe having it kinda be something different,
[01:11:05.828]and it's not us,
[01:11:06.996]and talking about it in the technology sense
[01:11:09.131]will actually help us talk about it,
[01:11:10.799]ourselves as human, with human intelligence.
[01:11:14.770]Thank you.
[01:11:15.838]MIKE: Interesting.
[01:11:16.705]On a recent 60 Minutes
[01:11:17.673]it was stated that China
[01:11:18.607]is much more AI research oriented than the US.
[01:11:21.610]Do you agree?
[01:11:22.745]JANA: No.
[01:11:23.746](audience laughing) MIKE: All right,
[01:11:24.680]next question. JANA: Next question.
[01:11:25.414]Search for chicken, eggs, and bacon,
[01:11:29.184]and Jana Eggers, and you'll see.
[01:11:31.787]Actually, Kai-Fu Lee wrote the book, called AI Superpowers,
[01:11:36.325]and it just came out a few months ago,
[01:11:38.093]and that's why it's being talked about a lot.
[01:11:39.862]And I go through, actually, in that article.
[01:11:41.864]That's what got me to the chicken, egg, and bacon thing.
[01:11:44.033]I actually go through his main points.
[01:11:47.269]Now, I wrote that before he wrote the book,
[01:11:50.906]'cause he had written a World Economic Forum post
[01:11:53.575]that is a very brief summary of his book
[01:11:56.345]before he wrote it.
[01:11:57.613]He has some other points,
[01:11:59.415]I have a whole presentation.
[01:12:00.516]He has added some other points.
[01:12:03.152]And I do think some of those points
[01:12:05.587]are a bit more on,
[01:12:06.855]as far as why China could win,
[01:12:09.591]but I just, I don't think that's gonna happen.
[01:12:12.327]MIKE: All right. At the end of your talk, Jana,
[01:12:14.863]you mentioned that you hope AI could help cure cancer.
[01:12:17.933]But cancer and disease processes are just algorithms
[01:12:20.235]that operate without sufficient regulatory oversight
[01:12:22.838]and disrupt homeostasis in their host organism.
[01:12:26.408]How can AI solve cancer without being a cancer?
[01:12:29.378](Jana laughing)
[01:12:31.447]JANA: Wow. (audience laughing)
[01:12:33.348]Is there an ethicist in the room?
[01:12:34.950](woman calling out)
[01:12:37.486]Biologist.
[01:12:39.254]So I think it's a good point.
[01:12:44.293]Part of the challenge there is,
[01:12:47.730]we may be able to solve that biological problem,
[01:12:51.767]but in doing that we create something
[01:12:54.036]that causes some other problem that isn't biological.
[01:12:59.141]Does that mean we shouldn't solve the biological problem?
[01:13:04.446]So that's how I would look at it, personally.
[01:13:07.583]Again, I'm not a pessimist on AI,
[01:13:11.487]so I'm not gonna look at it and say,
[01:13:14.723]"Well gosh, if I cure cancer,
[01:13:17.392]"I'm really in a worse position,
[01:13:18.961]"because I've created this horrible thing
[01:13:20.763]"that's also going to do X."
[01:13:22.765]And I don't even know what that X would be,
[01:13:23.999]but I can imagine there would be an X.
[01:13:26.368]But I would say we'd figure out ways of controlling that.
[01:13:31.039]I mean, you know, I worked at Los Alamos.
[01:13:33.475]We dealt with the nuclear weapons issue all the time.
[01:13:36.478]I regularly got stopped on a run where,
[01:13:39.281]'cause they were moving plutonium around.
[01:13:42.818]We've handled that.
[01:13:45.687]It doesn't mean it's solved forever.
[01:13:48.390]But we do have ways of handling things
[01:13:51.059]that do become dangerous.
[01:13:53.695]Doesn't mean we'll always win,
[01:13:55.864]but I think it's worth the risk to solve cancer.
[01:13:59.101]MIKE: Thanks for thinking through that with us.
[01:14:01.770]Before Jana takes her final question this evening,
[01:14:03.872]I want to thank each of you for joining us here tonight
[01:14:06.008]at the Lied Center,
[01:14:06.575]and via our web stream.
[01:14:08.777]I also want to encourage you to watch for more information
[01:14:10.779]about the exciting next season
[01:14:12.748]of the E.N. Thompson Forum on World Issues,
[01:14:14.583]and hope to see you back then.
[01:14:16.585]Jana, one final question this evening,
[01:14:18.220]following your fantastic presentation.
[01:14:20.489]How about the role of AI in government?
[01:14:22.991]And are you aware of any increased usage
[01:14:24.927]of the government regarding AI?
[01:14:27.062]JANA: I was hoping you were gonna ask me something,
[01:14:29.164]like about Razorback football, or (laughing).
[01:14:31.900]MIKE: That's not quite the color here, but--
[01:14:35.170]JANA: I thought, you know, Boston sports.
[01:14:37.773]You know, I thought we were gonna have something fun,
[01:14:39.808]but we're talking about government regulation.
[01:14:42.311]As I mentioned, I'm actually a big fan of GEPR.
[01:14:50.219]And my reason is, as I've said before,
[01:14:52.721]I think data and us being good custodians of data,
[01:14:57.793]and I'm not saying it's perfect,
[01:14:59.061]but I think it's a right step going forward,
[01:15:01.363]to tell people what you're collecting on them,
[01:15:05.934]allow them to opt out,
[01:15:08.370]and be able to take it actually out of the algorithm
[01:15:11.206]and the impact on the algorithm.
[01:15:13.141]I think those are all really good things to do.
[01:15:15.744]So I do believe that we can come up with smart regulations.
[01:15:20.215]I do worry
[01:15:22.451]that the people that are gonna make the regulations
[01:15:25.087]aren't gonna understand the technology,
[01:15:27.322]and that's where I think that we have to be involved.
[01:15:29.892]I'm not good at that,
[01:15:30.826]'cause I have little patience for bureaucracy.
[01:15:33.395]That said, I know some people that are great at it,
[01:15:36.465]and I do give them my point of view (laughing),
[01:15:38.500]on a regular basis, whether they ask me or not.
[01:15:41.670]So yes, I think government has a role in it.
[01:15:44.506]I don't think that they're going to be the great overseers.
[01:15:47.409]I get nervous, for example,
[01:15:49.244]when Elon Musk says, "Yes, government should regulate!"
[01:15:52.848]Because I think he wants to be the person that regulates.
[01:15:55.951]And so I worry about that,
[01:15:58.520]because I'm not in the same place on technology as he is.
[01:16:01.623]He thinks it's gonna kill us.
[01:16:02.991]He thinks we're summoning the demon.
[01:16:04.559]I don't think we're summoning the demon.
[01:16:07.763]And so that's where I get nervous about regulation,
[01:16:12.634]is that the wrong people would be doing it,
[01:16:16.371]either because they don't understand
[01:16:18.307]or because they do understand
[01:16:19.775]and they're gonna do it for their own advantage.
[01:16:22.878]MIKE: Thank you.
[01:16:24.079]Jana, we're much smarter and much more future-focused
[01:16:26.415]as a result of your presentation.
[01:16:27.683]Ladies and gentlemen,
[01:16:28.684]please join me in thanking Jana again.
[01:16:29.985](audience applauding)
[01:16:31.553]Thank you so much.
[01:16:32.387]JANA: Thank you.
[01:16:33.388]Thank you very much.
[01:16:34.356]Thank you guys.
[01:16:35.490]Thanks.
[01:16:36.358](upbeat music)

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Yes, You Have a Critical Role in Raising Artificial Intelligence.

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments