Latest / The Road to Accountable AI / Mitch Kapor: AI Gap-Closing

Transcript

Mitch Kapor: AI Gap-Closing · The Road to Accountable AI

• 0:01
  Mitch, it's always wonderful to talk with you.
• 0:02
  Welcome.
• 0:04
  Thank you, pleasure to be here.
• 0:06
  For those who don't know, you were one of the key people in kicking off the whole PC software era with Lotus 123 back in the 1980s, and then also one of the key figures in the
• 0:20
  whole digital rights conversation as a co-founder of Electronic Frontier Foundation.
• 0:26
  So fast forward to today with AI, what's your perspective on what you see out there happening given that history?
• 0:34
  Well, it is another enormous wave of disruptive innovation for all of the good and all of the bad that those bring.
• 0:46
  This is the latest in a series of my lifetime, my career, know, four or five, six of them, which really changes everything in huge, huge ways, changes daily life, changes who gets
• 0:59
  ahead in business uh and is quite
• 1:03
  unpredictable at the outset where it winds up and we're in the middle of another one these.
• 1:10
  Do you think that we can learn at all from the experience we had with the internet?
• 1:14
  We had all this excitement and then we realized all these problems and so forth.
• 1:18
  Do you think we can gain from that experience in terms of how we're dealing with AI?
• 1:23
  I hope so, but I'm not sure.
• 1:26
  There tends to be over optimism on the part of technologists who think that technology is the answer to human problems.
• 1:36
  ah And it isn't, and there tends to be a lack of realism about the downsides of what happens when you unleash powerful new technology.
• 1:46
  mean, we certainly, we are living through that in the world of social media, but.
• 1:50
  I could tell you as someone who was around in the early days of the internet, no one was really worried about that, or virtually no one.
• 2:00
  Yeah, so people like you and me with white hair always like to say that, but then people who were younger like you and I once were, you know, in this era always want to say, well,
• 2:11
  no, we're so excited we go forward.
• 2:12
  How do we break out of that cycle?
• 2:15
  Well, I don't think anybody has figured out a way to break out of that cycle.
• 2:19
  Quite honestly, that doesn't mean we shouldn't keep trying, but, the more thoughtful interaction we can have before everything is cast in concrete, the less we have polarized
• 2:37
  conversations about AI is wonderful.
• 2:40
  AI is terrible.
• 2:41
  Uh, the better it's going to be.
• 2:45
  And hopefully it will influence people who are making the new technologies and applying them, at least to think a little bit about the consequences of what they're doing.
• 2:57
  Yeah.
• 2:57
  At least what I see is because there is a fairly broad recognition you mentioned with social media that there wasn't enough thought early on to the problems.
• 3:07
  And then we saw all the problems happen.
• 3:08
  Now people are aware of the problems.
• 3:10
  At least it seems to me that there's some follow through with AI that people start with that recognition.
• 3:18
  I think so.
• 3:20
  But then where do people go?
• 3:23
  So, yes, it's great that everybody seems to be talking about AI safety.
• 3:32
  But should we really trust these large, sprawling empires of open AI and anthropic to do the job themselves without oversight?
• 3:46
  I think that incentives to aggregate power, to create wealth, to advance self-interest are so powerful that you can't let people regulate themselves.
• 4:02
  It doesn't work.
• 4:04
  And so that's an area where I think pressure from the outside of various kinds is super important.
• 4:12
  The other big issue today is the political dimension.
• 4:17
  And you've been working for many years, both as an investor and a philanthropist on promoting underrepresented founders and equity in technology.
• 4:27
  uh Now there's this processor saying that anything about diversity is harmful.
• 4:34
  Yeah, how do you think about that?
• 4:36
  And how do we get beyond that?
• 4:38
  Well.
• 4:41
  If you look at what we've actually been doing at Kapor Capital for the last 15 years almost, it's really not a diversity play.
• 4:54
  And I'm going to explain that we're interested in investing in startups, is a venture fund, that with the outcome of a successful investment, the business gets created.
• 5:09
  does close gaps of access or opportunity or outcome for low-income communities or communities of color.
• 5:16
  But it's completely agnostic about who the founder is and who gets invested in.
• 5:23
  So we're not and have never advocated taking a demographics first approach to things.
• 5:31
  It doesn't make sense.
• 5:32
  We should be interested in outcomes and in solutions and
• 5:37
  you know, formally speaking, that's a race neutral sort of thing to do.
• 5:42
  Now, it turns out, as anyone who knows the startup world, that um oftentimes entrepreneurs scratch their own itch.
• 5:52
  So if their lived experience, if the lived experience of the founder, say, uh because of what they went through, their family or the community, their school or the, you know, what
• 6:04
  they saw in their community, uh
• 6:07
  has to do with the structural inequalities because they went through that and they live through that.
• 6:13
  They're more likely to want to start a business that's going to do something about it.
• 6:16
  So it's not surprising that there's disproportionate, ah there's lot of diversity in the portfolio that we put together.
• 6:25
  But uh we're
• 6:29
  I'm amazed that even though we've been saying nothing but this for 15 years, people still assume we're sort of demographics first or diversity first.
• 6:39
  it's, we've never, never done that and don't believe in that.
• 6:44
  And we're not, that helps us stay out of the crosshairs of this, you know, anti DEI backlash that has taken over.
• 6:59
  How do you define gap closing?
• 7:03
  Well, I mean, we try to be very concrete.
• 7:08
  So there can be different kinds of gaps.
• 7:10
  That's why we say access, opportunity, or outcome.
• 7:12
  But if you look at a specific uh vertical, uh say, education, uh let's look at access, which high schools offer courses in computer science and AI and which don't?
• 7:32
  Our foundation did a study now probably six or seven years ago showing that the more black and brown the student population was, the less likely it was to offer those courses.
• 7:45
  If you don't offer the courses, how can you get on the ladder that goes to a STEM career?
• 7:50
  How can you participate in the economic upstart that comes from doing good things in the tech world?
• 7:58
  So that's a lack of access.
• 7:59
  So in evaluating startups, we say, well, if this thing succeeds, who's going to be better off and in what way?
• 8:08
  Is it going to help people with subprime credit scores get credit and participate in it?
• 8:13
  Is it going to help uh low income workers advance up a job ladder?
• 8:22
  So it's about access.
• 8:24
  It's about opportunity.
• 8:25
  And it's also about outcome.
• 8:27
  So if you look at differentials in health equity, for instance, uh mortality among infants and among Black women, much greater than the population as a whole, maternal care, uh
• 8:44
  there isn't a good reason for that.
• 8:46
  so uh that's what we mean to close gaps is to get to a world where those differences.
• 8:54
  uh
• 8:55
  diminish or are eliminated entirely.
• 8:58
  I was interested that you actually did a master's degree recently at the MIT Sloan School on gap closing investing.
• 9:04
  So I'm curious what you learned from that experience.
• 9:08
  Well, so that was interesting because I dropped out of MIT 45 years ago to seek fortune in Silicon Valley.
• 9:17
  as we know, things got distracted and it took a little while to close the loop.
• 9:22
  And so it was a great opportunity to think through and write up uh in a systematic way the approach to investing that we had taken.
• 9:34
  Where did it come from?
• 9:35
  How did we practice it?
• 9:37
  were we giving up financial returns in order to do good, to consider all those issues and to look at all of the evidence.
• 9:45
  you know, it gave me real sense of completion.
• 9:49
  Personally speaking, here's the thing that I learned that was surprising, which was in tracing back the origins for what we did to impact investing before that, to socially
• 10:02
  responsible investing before that, to the
• 10:06
  kind of investing done by some of the Quakers and other uh religious organizations in the 19th and 18th century.
• 10:16
  Going all the way back, all the way back, I found amazing resonances in the Jewish tradition, I'm Jewish, about business ethics and their conduct that I had absorbed that
• 10:30
  were just part of my DNA.
• 10:31
  Never been taught that, never had any formal education like that.
• 10:35
  And I said,
• 10:36
  my goodness, there's really something here.
• 10:37
  there's a little bit about that in the thesis, the Jewish origins of, uh or the connection ah that it had for me.
• 10:46
  And I had a lot of other material that I didn't actually put into the thesis because it didn't fit in.
• 10:52
  But that was interesting and surprising to me.
• 10:56
  Yeah, no, it's fascinating how this comes through different directions.
• 10:58
  As you mentioned, the Quaker, I am at an institution that was founded by a Quaker, Joseph Wharton, and that was a big piece of why there was a need for this kind of institution.
• 11:07
  Yeah.
• 11:08
  So these issues about how you conduct business and how to do it in a way that isn't in accord with from set of, uh, you know, ethical principles, it is a perennial issue and
• 11:21
  we're, we're, we're, facing it yet once again.
• 11:24
  I mean, it's never gone away.
• 11:26
  And, and, and here we are again with, with AI.
• 11:29
  Yeah.
• 11:30
  So what are some of the gaps that need to be closed for AI?
• 11:36
  Well, let me make the positive case first, which is we see big opportunities to apply AI in education, in fintech, in uh the future of work, in ways that actually uh improve
• 11:57
  access and outcomes.
• 11:59
  And uh we're seeing in some the investments that we're making and other
• 12:05
  you know, other people are making.
• 12:06
  And in that sense, AI really isn't that much different than other technologies.
• 12:13
  You say, okay, can we, you know, create a product or a service that, let's say, in education is engaging for kids, measurably improves learning outcomes, has a path to be
• 12:28
  adopted within the public school system, within Title I schools in particular.
• 12:34
  And, um, you know, we're finding that there is a set of entrepreneurs who are doing this.
• 12:43
  will say it.
• 12:45
  It is harder than it looks.
• 12:46
  It's not just a question of hooking up chat GPT to something and, know, popping it out.
• 12:53
  It's going to be like previous waves quite a few years before the stuff is sufficiently.
• 13:01
  mature and well done that we really see the benefits.
• 13:05
  And I think the market is missing that.
• 13:08
  And there's going to be some correction around that.
• 13:13
  This is some, if the bubble maybe not popping, but deflating, but that's its own subject.
• 13:20
  We're bullish though on the long-term about the use of AI in...
• 13:26
  uh
• 13:28
  improving, you know, in addressing a number of these social issues and problems.
• 13:34
  I'll give you just one anecdote, which is so interesting.
• 13:37
  You know, we've created this
• 13:42
  a highly regulated environment by which we convey benefits to people, know, snap and setting aside the shutdown for the moment.
• 13:54
  And we regulate at the state level, we regulate at the federal level.
• 13:59
  And when it comes to actually, you know, conveying some social benefit to some population, there are literally thousands of pages of regulations which are
• 14:11
  constantly changing that have to be rationalized.
• 14:15
  And we spend enormous amounts of money on consultants uh just figuring out what goes with what and is this in contradiction?
• 14:24
  Can we do this?
• 14:25
  And does this person qualify?
• 14:27
  And I've seen AI, I call this Doge for good, go through this and sort it out in at one, one thousandth the cost at one, thousandth the time.
• 14:40
  The result of which is that people who are eligible and deserving for a particular social benefit that's part of our safety net can actually get what they deserve.
• 14:50
  And without AI, it wouldn't happen.
• 14:53
  And there's more efficiency and less fraud.
• 14:55
  This is a company called Promise, founded by Phaedra Ellis Lamptons.
• 15:00
  That is something we have an investment in, full disclosure.
• 15:05
  But seeing the way AI works there to really
• 15:09
  cut through all of these regulatory overreach.
• 15:14
  And I say this as a diehard progressive, but who also believes in the kind of things that people talk about, the abundance agenda.
• 15:23
  I think AI has a role to play.
• 15:28
  The Kapor Foundation just put out a set of responsible AI principles.
• 15:31
  This is one of the reasons I wanted to talk to you now.
• 15:33
  Walk us through what you saw as the need for that and how it's different from all the other principles that are out
• 15:40
  So first of all, credit for initiating this should go to Dr.
• 15:45
  Allison Scott, who is the CEO of the foundation, who really a couple of years ago flagged AI as a very important frontier area for philanthropy.
• 15:58
  And these principles, which were developed after a great deal of consultation with other organizations, are uh part of the tangible fruit of it.
• 16:10
  One of the reasons why we did this is that the philanthropy sector has been lagging in figuring out what to do about AI.
• 16:19
  When you look at surveys of, know, foundations, are they on top of this?
• 16:26
  Is this a factor in their grant making?
• 16:30
  they know how, they go, we don't know how to think about this.
• 16:34
  So it was to help close that gap that, uh
• 16:39
  these principles were developed and in the formulation, which has been published, is in particular oriented towards philanthropy.
• 16:48
  We are in the process of adapting them and for investment.
• 16:54
  so it'll be the same.
• 16:55
  It'll be intellectually consistent, but they'll take a different form about what does it mean to us to be a responsible A.I.
• 17:03
  investor?
• 17:05
  One of the
• 17:06
  good things that is happening in philanthropy.
• 17:08
  And I would think this is in part a recognition, there's always been a recognition that philanthropy should be more collaborative, but foundations have been working in their own
• 17:18
  silos is with the, in the current administration is sort of force the issue that foundations realize they have to work together and that the overall impact will be much
• 17:31
  greater if they do so that we're part of this.
• 17:35
  organization called humanity.ai, which has a $500 million commitment to uh do responsible AI grant making.
• 17:44
  And that is something new.
• 17:45
  We haven't seen that type of cooperation on the scale before.
• 17:51
  So that in some sense, for people who have been living in pioneering the world of responsible AI, like yourself and others, you could look at what we're doing and go, well,
• 18:01
  this is, you we like this, but
• 18:03
  what's really new here.
• 18:05
  And I just point to this is laser focus right now on the philanthropy community, particularly as a government, federal government is stepping back from all sorts of
• 18:14
  things.
• 18:15
  There's obviously more and more of an urgent need.
• 18:19
  Yeah, was interesting what you said.
• 18:21
  What's different about responsible AI in a philanthropic context, an investment context, and then a corporate context where a business is trying to deploy?
• 18:33
  What's different?
• 18:35
  Well, I guess when you're talking to philanthropists, you're trying to help them see how responsible AI is consistent with their mission and values, which were obviously created
• 18:56
  beforehand.
• 18:57
  And those...
• 19:00
  tend to be all over the map, but they're all about promoting social good of one kind or another.
• 19:08
  So there's really bridge building to be done and education.
• 19:12
  And it's not so much coming up with new ideas as helping philanthropists who have ways of thinking about things and doing things to embrace an appreciation of the
• 19:30
  the benefits and the risks and dangers at both.
• 19:34
  And on that point, I would say that philanthropy is a lot of it, not all of it, has been tech backward, tech fearful.
• 19:46
  And so the initial instinct is how do we protect people from the bad effects?
• 19:54
  And there's certainly, that work has validity.
• 19:59
  But if that's all you do, you overlook the opportunities for AI for good and how you use AI in ways that positively promote the kinds of things that, kinds of outcomes that they
• 20:11
  would like to see.
• 20:13
  And to do that though, they need to become comfortable with the idea, that it's possible and how that works and then how they make decisions to embrace that.
• 20:24
  mean, there's just a lot of...
• 20:26
  on the ground work that has to be done to realize the potential of what philanthropy can actually do about responsible AI.
• 20:33
  Because we're not talking just about white papers here, pronouncements.
• 20:39
  We're talking about putting resources in the hands of people who are actually going to do something that will make a measurable difference, measurable outcomes.
• 20:50
  But how can uh foundations that don't have that technical DNA, your tech entrepreneur, you set up your foundation, how can they get that skills and the people that are able to
• 21:02
  understand it?
• 21:03
  Well, it's certainly, it's a tall mountain and a multi-step process, but you could start with improving and increasing AI literacy, which is actually one of the principles.
• 21:18
  And I will say, just happened to read today this great example in the New Yorker.
• 21:28
  of a piece by someone who is a software developer, but really explaining in a way that is more clear than anything I've seen what it really means when we talk about these large
• 21:41
  language models, thinking and understanding and how is it the same and how is it different from what humans do?
• 21:48
  And it's a non-technical, anybody can read it, but it brings the worlds together in a way that I thought had some.
• 21:55
  valuable insights and we need more of that and we need to disseminate that and it needs to be sort of baked into, you know, people's thinking.
• 22:04
  And some foundations should hire a chief technology officer.
• 22:08
  I've been saying this for, you know, it took, how long did it take government to decide that having a CTO here and there was a good thing to do?
• 22:18
  It didn't happen overnight.
• 22:20
  I think from when we started talking about it to what happened was a decade, maybe more.
• 22:25
  Well, here we are again with Flanth.
• 22:28
  Yeah.
• 22:29
  Well, and it took it took healthcare.gov to wake people up to the need to have tech teams in government to address them.
• 22:37
  That's right.
• 22:37
  No, that's, how we needed a gigantic failure to.
• 22:43
  Yeah.
• 22:43
  Um, so one, one more thing about this, I guess, you know, back to the political question, one of the principles you have there is supporting AI initiatives that shift power.
• 22:53
  And if I'm looking at that as, uh, someone who is in power right now, um, or someone who's skeptical that philanthropy is all, you know, on one side of the political spectrum and
• 23:06
  therefore should be dismissed, that to me looks scary.
• 23:08
  That looks like it's saying, okay, this is all about
• 23:12
  the left and not about something that's good for all society.
• 23:15
  I'm sure you don't think that.
• 23:16
  So what's the response to someone in that position?
• 23:20
  Well, you know, the first thing is I would say that today the right is as much or more about shifting power than the left.
• 23:30
  If you look at the revolt against the elites.
• 23:34
  So, I mean, as I say this, I realize, well, so do we really want to say shifting power?
• 23:42
  I might.
• 23:45
  I would interpret it this way, that what is valuable is to have processes that are more inclusive of more points of view and less dictatorial.
• 24:00
  We know what's good for you.
• 24:02
  And to really bake that in, to have a pro-social approach to developing your products and services, however you do that, and there's so many different ways that you could do that.
• 24:15
  to me a good thing and not it's it's not partisan.
• 24:21
  It just, you know, and I think.
• 24:25
  It's germane because.
• 24:28
  And it's hard to say this, these guys running these big companies and I used to be in that world in one of them, they're arrogant, they think they know better and they just go and
• 24:41
  do stuff.
• 24:42
  The heedlessness of it.
• 24:45
  ah is dangerous and highly consequential and shifting power, yes, it's a kind of a shot across the, it challenges that directly, but I think sometimes you have to speak up and
• 25:00
  say what's important.
• 25:01
  Doesn't mean you can't have a dialogue about it.
• 25:06
  Yeah, absolutely.
• 25:07
  No, and you mentioned humanity AI and that's $500 million, which is a big number.
• 25:13
  But at a time when Sam Altman is talking about open AI spending a trillion dollars or more on infrastructure investment, where are the leverage points to move things?
• 25:25
  Well, so the first thing is when he says things like that, the whole purpose he's trolling us, regardless of how much he actually spends, it's to discourage anybody else from
• 25:33
  competing.
• 25:34
  I mean, he's not worried about humanity, AI is worried about the competition.
• 25:38
  So it's fair just to apply an enormous discount factor.
• 25:41
  They are going to throw huge amounts of resources at this.
• 25:48
  So first of all, would say based on those of us who remember what happened in Web 1.0, most of this money is gonna be wasted.
• 25:57
  There'll be a few winners, enormous winners, enormous wealth creation.
• 26:03
  But in general, things will be overbuilt and not used and they're gonna get it wrong.
• 26:10
  it's what happened with the railroads in the 19th century.
• 26:13
  If you know the history of things are overbuilt, unnecessary competition.
• 26:19
  in creating what became absolutely crucial infrastructure.
• 26:25
  But let me answer the question more directly.
• 26:29
  uh You have to target.
• 26:32
  So Humanity AI and the Kepler Foundation, Alison has a seat on the steering committee, but let me be clear, I'm not operationally involved in the least.
• 26:43
  But I think you would want to target where it thinks it can
• 26:47
  specifically do good.
• 26:49
  It's going to leverage all the infrastructure of, you know, OpenAI and Anthropic and everybody else, hopefully to build or to fund solutions or interventions that, you know,
• 27:02
  make an enormous kind of difference.
• 27:04
  And to do that, you don't have to spend the same orders of magnitude kind of money.
• 27:12
  I mean, if you look at it, it's like startups.
• 27:14
  uh
• 27:15
  like Lotus or if you're doing applications, a relatively small amount of investment compared with what you need if you're doing all this infrastructure can still have world
• 27:28
  changing effects.
• 27:31
  So I think that's.
• 27:32
  the era of $50 million seed rounds.
• 27:34
  Yeah, well, we're in a bubble and I'm pretty sure that we will not continue to have $50 million seed rounds indefinitely.
• 27:46
  Are you still optimistic about the future?
• 27:50
  I make the following distinction.
• 27:52
  I'm not optimistic, but I'm hopeful.
• 27:55
  Meaning, I'm not making a forecast that says, probably things are going to be fine.
• 28:03
  That would be, to me, that's optimism in practice.
• 28:07
  I'm hopeful that we can do things that will make a difference and lead to better outcomes.
• 28:11
  That is always possible.
• 28:14
  uh
• 28:15
  But the fact of the matter is people of my generation, we should be getting off stage now.
• 28:20
  We're just, it is up to a younger generation to what we can do to help make a better future is to make sure that younger people are empowered to make the kind of positive
• 28:35
  differences we would like to see.
• 28:37
  We need to give up our seats at the table and support them.
• 28:45
  We have an interesting mayoral election today as we record this in New York.
• 28:50
  And to me, that is the most hopeful thing is I see leaders in the startup world and in philanthropy of a younger generation who really are ready to step up and make a positive
• 29:06
  difference.
• 29:06
  And that is what gives me hope.
• 29:10
  It's great to hear because now I think we need hope to go forward.
• 29:13
  And obviously there's so much excitement and so much potential in the time we're in now.
• 29:17
  Absolutely.
• 29:19
  Mitch, great to talk to you.
• 29:21
  Thanks so much for your time.
• 29:22
  Absolutely, you're most welcome.
• 29:23
  Great to talk to you.