- il y a 3 mois
Can We Have it “All” Safe, Profitable, and Ethical AI
Catégorie
🤖
TechnologieTranscription
00:00Good afternoon. I'm Jennifer Shanker, Editor-in-Chief of The Innovator, a global publication about innovation.
00:08Welcome to the session on Can We Have It All? Safe, Profitable, and Ethical AI.
00:17We have with us here today Meredith Whitaker, who some of you have just heard speak,
00:24who is president and chairperson of the Signal Foundation.
00:28We have Jonas Androulis, founder and CEO of Aleph Alpha, a made-in-Europe, large-language model company.
00:39We have Sneha Revener, who is the founder and president of IncoJustice,
00:45and Hanel Bhavja, a principal at Kriantam, an investor.
00:55So, to set the scene, I'd like to shine a spotlight on a few of the news stories that happened
01:03last week, in just one week.
01:06So, Google said, on May 14th, that it's going to turbocharge its search engine with its AI model,
01:15drawing on the technology to directly answer user queries at the top of results pages.
01:22Now, that may sound like a good thing, but for news publishers, many of whom are already struggling after steep
01:31traffic declines,
01:32this revamped search experience will likely cause even further decrease in their audience.
01:41In fact, you know, it is just one more thing that is helping to bury the free press, which is
01:49key to democracies and keeping business and government honest.
01:53Microsoft revealed that its emissions have risen by almost a third by 2020, as it tries to push out the
02:03infrastructure behind AI,
02:05threatening its own proclaimed climate goals.
02:08Mark Reed, the chief executive of London-based advertising group WPP, was the target of a deepfake scam in which
02:18criminals used advanced AI-powered voice clone software
02:22and public YouTube footage to set up a video meeting with his executives.
02:27The goal was to try to get personal information and money.
02:32It failed, but other deepfake scams are succeeding, fueling fears of how these deepfakes could negatively impact business and elections
02:44around the world.
02:45Meanwhile, AI is hitting the global labor market like a tsunami.
02:52The International Monetary Fund Managing Director said on May 13th at an event in Zurich.
02:58She said that AI is likely to impact 60% of jobs in advanced economies and 40% of jobs
03:05around the world in the next two years.
03:07And I quote from her,
03:10We have very little time to get people ready for it, businesses ready for it.
03:15It could bring tremendous increase in productivity if we manage it well,
03:19but it could also lead to more misinformation and, of course, more inequality in our society.
03:26So, I think it's safe to say the world has moved on from this conversation about AGI and killer robots
03:33to very real discussions about the impact that AI will have on our society.
03:40So, against this backdrop, I'd like to ask each of our panelists to discuss whether, in their view, the benefits
03:49of AI outweigh the potential downsides.
03:52And so, Meredith, let me start with you.
03:56Well, perhaps we have to caveat that and say the benefits to whom.
04:01Because I think the quick historical vignette I just gave illustrates that most of the technologies we are right now
04:10referring to as AI
04:12are these bigger is better, massive models that are accruing huge amounts of profit and speculative investment in a handful
04:25of companies
04:26and those who are kind of using those company resources.
04:30And these companies are largely jurisdictions in the U.S.
04:33So, I don't think we can think about social benefit without recognizing the concentration of power in the AI industry
04:41and the fact that that power is in the hands of a few U.S. companies.
04:46So, no, that is not a recipe for societal benefit.
04:49That is a recipe for the S&P 500, which is right now 50% tech stocks rising in time
04:57with AI.
04:58But, again, if we're talking about labor markets, if we're talking about the implications for every other country in the
05:04world
05:05which does not have this historical legacy that enabled platform monopolies, no, that is not a beneficial situation.
05:14And, you know, if we begin specifying who benefits and who loses, we get a very different picture from the
05:19rosy marketing that we often hear from those executive offices.
05:24Thank you, Meredith. Samir, what is your point of view?
05:26I think right now, candidly, we're heading down the wrong path, but that is entirely reversible.
05:31Fundamentally, this is a matter of human decision making.
05:33It's a matter of the values and priorities that we let steer the technologies that we build.
05:38It's a matter of whether or not we choose to move full throttle ahead without any breaks or insurance,
05:44or if we choose to think and be careful and be thoughtful about the civilization-changing technologies we're building.
05:51So I would say it's all about humans, and at the end of the day, for the benefits to outweigh
05:54the risks,
05:55we have to choose to act under a different paradigm, but right now we're on the wrong path.
05:59Okay, thank you.
06:01Hannah.
06:03I'm the investor.
06:05I don't know if this is on.
06:06Hopefully it is.
06:07Maybe hold it a bit closer.
06:09I'm the investor in the group, so I think in some ways it's my job to be the techno-optimist,
06:14or at least that's what I think venture capitalists like to tell themselves about AI.
06:19And I think it's a broad question, but maybe sort of just boiling it down to can AI, can the
06:28benefits of AI outweigh the cons?
06:31Absolutely.
06:32And I think we're seeing some incredibly exciting signs of that and also reasons to be cautious.
06:38But I am, you know, definitely optimistic about sort of the upside here.
06:45So let's turn now to the entrepreneur.
06:49So AI is definitely one of the wildest technological innovations we had in terms of magnitude and speed.
06:57But in the end, it's just technological innovation.
07:01So it's basically just the nearest of a long line of technological innovation.
07:05And technological innovation and innovation in general has brought tons of good.
07:10We lifted hundreds of millions of people out of poverty.
07:13And so a lot of good change has come.
07:16But I certainly agree with Meredith in the sense that we are shifting value creation away from labor to IP.
07:24And we're concentrating power with some of the huge companies that are not only monopolies,
07:31but have monopoly chains that they use to protect each other.
07:35So I think that's certainly something we should be mindful of.
07:38In general, I think, yes, the good guys will win.
07:41We will use tech probably more to the good than to the bad.
07:46But we should be careful of how this will change society and how we can kind of build our values
07:54into the future.
07:55So I think that's a great segue to my next question.
07:58So in an April 9th speech, European Commissioner Vice President Margaret Vestiger said,
08:06Probably never in history have we been confronted with a technology that has this much power but no predefined purpose.
08:14Neither good nor bad in itself.
08:17It all depends on how we humans shape it and use it.
08:21Just like in Oppenheimer's time, we're faced with what AI researchers call the alignment problem.
08:28When technology has the power to both serve and destroy us, how do we channel its development?
08:34How do we ensure this technology reflects the societies that we want to have
08:39instead of amplifying the flaws and injustices of the one that we already have?
08:47So, Meredith, let me turn to you and say, what do we do about this?
08:52How do we ensure that it reflects the society that we want?
08:56Right.
08:57Well, I think this is often the murky area these conversations tread into
09:02because, in fact, those are not technological questions.
09:05Those aren't really even questions about tech.
09:08Those are questions about political economy, about the way we organize our social and economic relationships.
09:17And, fundamentally, the answers to those questions will stem from the answers to larger pictures, right?
09:26Who is benefiting right now from these technologies when we have, I think it's 70% of Series A funding
09:34going into AI startups
09:35is from the large cloud corporations based in the US, often in the form of compute credits.
09:43So, they're not even giving them money.
09:44They're saying, hey, you can use our infrastructure because it's so, so, so expensive to develop this technology
09:51as a kind of first-order investment.
09:53And then we'll see about it, right?
09:55So, we are talking about a market that is extraordinarily warped by the concentration of resources in the hands of
10:02these companies.
10:03And if we don't change that, the answer to that question is, you know, well, no.
10:08We're not going to get social benefit.
10:10We're going to get incredible revenues and incredible growth stemming to a few companies
10:15that are, again, jurisdiction in the US at a time of political volatility in the country.
10:20So, we need to take that dynamic extraordinarily seriously as a political question, not as a technical question,
10:27but that will dictate whether we can answer yes to the idea of societal benefit.
10:33So, if we're not past the point of making this fixable yet, who can stop this?
10:41Who can change it?
10:42Will Europe be the policeman for the police person for the world again?
10:46Yeah.
10:47I mean, that's a, you know, the subjunctive tense is the right tense for that, right?
10:51I think we have the tools to fix this.
10:54We can structurally separate the infrastructure layer from the application layer.
10:59That would be a competition measure.
11:02A very moderate reading of GDPR could easily ban surveillance advertising,
11:09which is the kind of economic engine of these platform monopolies.
11:13We have the tools.
11:15It's a question of political will, and it's a question of mustering that will
11:19against an incredibly well-resourced and incredibly influential tide of marketing and lobbying
11:26and self-interest that has infiltrated Brussels and many other capital cities in this moment.
11:32So, Sneha, in the years since Encode Justice's founding,
11:37your organization has contributed to the passage of local ordinances,
11:43restricting surveillance in cities around the US.
11:47You've spearheaded advocacy efforts for more than 10 pieces of federal legislation.
11:52You helped shape the White House's Bill of Rights for an Automated Society
11:56and President Biden's executive order on AI.
12:00What do you think needs to be done to ensure that we encode justice into the technology?
12:08To encode justice into technology, we have to think about it and build it in a much different way.
12:12I know we're talking a lot of power concentration here, so I guess I'll illustrate that.
12:16Right now, we have a couple of people sitting around a table in Silicon Valley
12:20and essentially saying, this is the world that I want to see,
12:24and I'm going to harvest the resources and capital that I need to build the technology that could get us
12:29there,
12:29and I'm not going to ask anyone else if that's what they want to.
12:33I'm not going to respect actresses who don't want me to steal their voice from my models.
12:37I'm not going to stop and think about the risks that I'm externalizing on just about everybody else
12:41in my quest to transform everything and make a lot of money.
12:44And that is kind of the MO right now for a lot of these companies, and that to me is
12:49pretty harmful.
12:50So I think that fundamentally we have to reimagine the way that we're approaching AI development
12:55and move towards a much more democratic, a much more participatory approach.
12:59We need to bring youth to the table.
13:00I mean, this is the world that we're going to be inheriting, and we're on the front lines of this
13:04AI revolution.
13:05I would also say that we need some clear guardrails on this technology.
13:09For some reason, every other consumer-facing product is very closely regulated to ensure that we meet a minimum standard
13:15of safety.
13:16But when it comes to AI, people just can't seem to imagine how innovation and regulation might possibly coexist.
13:21So we need to move forward sort of behind those baseline concerns around regulation stymieing innovation
13:27and realize that it's possible to have both, and it is possible to achieve safe and profitable and ethical AI.
13:34Actually, a couple of days ago, Enco Justice, my organization, launched a platform called AI 2030
13:40outlining 22 recommendations for world leaders to secure our shared AI future by 2030.
13:46In it, we outlined policy asks to address a whole range of issues ranging from bias to disinformation
13:51to the future of work for our generation to the use of autonomous weapons.
13:54We're obviously hoping to offer some concrete ideas and move forward.
13:58But I think it really will go back to the question of, are we going to reimagine the incentives that
14:04are guiding AI development,
14:05the values that are guiding AI development, and move towards a different paradigm altogether?
14:09Because that's what we're going to need.
14:10Thank you. So we've just heard about how, you know, different regulatory environment is needed.
14:18But we also heard about the need for more competition.
14:24And Jonas, that's where you come in. You are building a large language model alternative for Europe.
14:30Tell us a little bit more about that and the difficulty of going up against these giants.
14:40Yeah, there was quite a wild ride. So we started in 2019. The first GPT style model we built was
14:47built on one Titan RTX,
14:49like one GPU, even before GPT-3 was out. And suddenly we found ourselves in competition with the best teams
14:57in the world,
14:57with the deepest pockets. And rightfully so, this technology is priority number one for all the big tech companies.
15:07They all focus on dominance here. And of course, this has, in a way, caused some problems for us
15:14in that we had to face up to that responsibility and build something meaningful.
15:19And I think we've succeeded so far. We've succeeded in funding this extremely sovereignly.
15:25We did no shenanigans deal of like compute or exclusively on that cloud or you can only use that chip
15:31or something like that.
15:32And we are handing this sovereignty to our customers. But of course, this is kind of, it's not, we haven't
15:39succeeded yet, right?
15:40It's still a fierce competition. There is a lot of question about where value actually is.
15:47So we can, we can go into that. And of course, regulation, while I, of course, understand that there's also
15:54this,
15:55there's always this idea that it would be fun or would be good if the state or if government would
16:01step in
16:01and kind of take care of all the things that can go wrong. This, of course, causes regulatory capture.
16:08And this is, of course, the reason why these tech giants are, have caused some of the doomsday panic
16:16that caused this kind of, dare I say, over-regulation or at least maybe a little bit excessive fear in
16:24some sense.
16:25So do you believe, let's talk about the opportunity for a minute, because yes, I mean, many people in Europe
16:33believe that it's important to have tech sovereignty and AI certainly being within that group.
16:42But do you believe that there will be a demand even beyond Europe for an alternative form of AI to
16:51the US offers and Chinese offer?
16:56Yes. So there's certainly a strategic element here. The challenge with strategic elements like sovereignty or something like that,
17:07it's always difficult to find somebody paying for that. And so that is why we have a lot of success
17:13on the, on the CEO level,
17:15basically telling them that if you use your tech sovereignty, if the crucial part of your knowledge work will be
17:23done by AI,
17:25that's a black box that is exclusively in one cloud you can never own and you can never control,
17:31then why do you think a lot of the value can be captured by you?
17:36Right. And because, I mean, we already see this to some extent with infrastructure.
17:41By the very time a hyperscaler or any cloud provider knows that you have no option, you cannot move out,
17:49pricing changes.
17:51Right. So that's certainly that. But of course, we're not just counting on our postcode or that we are almost
17:58exclusively European funded.
17:59We also have some proprietary innovation that we're designing our whole product with a different design paradigm
18:07so that we're not just trying to build a chatbot that's hopefully right.
18:11But we're trying to design a human machine paradigm that enables the human and gives the human superpowers.
18:17And that's why our customers are the kind of industries that are so complex and critical that responsibility is key,
18:25like government, health, finance, security. That's where we are very strong.
18:32Okay. Thank you. So let's also look at other opportunities for, you know, to create new businesses.
18:42And I'll turn to Hanel for that. Do you, you know, obviously there will always be a limited number of
18:50players on LLMs
18:51just because of the amount of investment that's needed. Do you see real opportunities at the application layer?
19:00I am. I really agree with your question of where does value accrue?
19:05And we've spent a lot of time discussing this, like how valuable will the application layer be?
19:12How sort of thin is that value over time?
19:15I think so maybe two things to quickly highlight.
19:18One is that I think LLMs and sort of the applications within language have been the first to market
19:26and are also the easiest to understand as humans, right?
19:29We can use chat GPT and get an amazing answer.
19:33But if you look at what some of the research labs out of places like Google and other are working
19:38on,
19:39there's so many things that this technology can do when it comes to protein discovery, drug discovery,
19:45the sciences, chemistry, new materials, new molecules.
19:49So I think this is one area which is sort of applying similar principles of compute and otherwise to other
19:56problems outside of language
19:58that can unlock new drugs, new pathologies, new materials, can redesign the way we make buildings, can maybe, you know,
20:05help solve the climate crisis.
20:07This is like one broad area that we're very excited about.
20:10And then I think when it comes to the actual sort of application layer on top of LLMs,
20:17so tools that companies are either building in-house or using to automate sales work, legal work, HR work.
20:24We haven't touched on open source at all.
20:27I don't know if we have time to talk about that in this discussion.
20:30But, you know, certainly a trend that we're seeing over the last six months, eight months are founding teams,
20:38building on open source models as a faster and more capital efficient way to get something in front of customers
20:45and consumers.
20:46I think the jury to some extent is still out on whether this is valuable long term and what will
20:51create defensible businesses there.
20:53But maybe these are just two ideas to share back.
20:56Okay. And, you know, when we're talking about the dangerous side or the dark side of AI,
21:05I think we shouldn't forget the role of large corporations because they also have a role to play in responsible
21:16AI.
21:17And, in fact, Stanford University's human-centered AI group publishes an AI index report every year.
21:31And in the latest one, a significant number of the thousand companies that responded admitted that they had only some
21:43or even none of the guardrails in place
21:46when they launched their AI services.
21:50So, Hanel, can you speak about, you know, what other opportunities this represents for entrepreneurs in helping ensure that we
22:01have responsible AI?
22:02You know, for example, AI governance software or risk management solutions like insurance, etc.
22:10Do you see companies coming up with these kind of solutions and helping to build the guardrails in another way?
22:20There are a lot of somewhat early stage startups working on this question of how to help companies mitigate risk
22:28when they have a model in deployment in a customer-facing application.
22:31I mean, I'd be curious for your view also on working with customers that maybe care more about this and,
22:38as such, are choosing different model providers that sort of fit that.
22:42So, I think the short answer is yes, but these companies are still early.
22:47And so, the jury's out again on sort of which ones will be durable and sustainable long-term.
22:52And it could maybe be some of the larger established companies like insurance companies, for example, that come up with
22:59completely new offers as well.
23:03So, there are some opportunities there.
23:06Now, I want to be, like, intentionally provocative.
23:10Last week, the New York Times ran an opinion piece by an investigative journalist and book author named Julia Angwin.
23:19And she, the premise of her opinion piece was that AI is not living up to its hype.
23:26And she said, look, for more than a year, the world has been asking what will be the consequences if
23:32AI becomes too smart.
23:33She says we're asking the wrong question. What will be the consequences if AI stays as dumb as it currently
23:41is?
23:42And I'm going to quote from her opinion piece.
23:46Should we, as a society, be investing tens of billions of dollars, our precious electricity that could be used towards
23:55moving away from fossil fuels,
23:57and a generation of the brightest math and science minds on what, up to now, have been incremental improvements?
24:05We can't abandon work on improving AI, she wrote.
24:08The technology, however middling as it is right now, is here to stay, and people are going to use it.
24:14But we should reckon with the possibility that we are investing in an ideal future that may not materialize.
24:23Do the panelists agree or disagree with us?
24:28I generally agree with Julia, and I think she's spot on here.
24:33Again, I think there is a real question as to whether the energy resources, which are extensive,
24:40whether the authority we're giving to those who develop and use these models, the companies we license the APIs from,
24:48is actually worth an email prompt, is actually worth a chat bot integrated into our messaging service that gets in
24:58the way of our talking to our friends.
24:59So I think there is a dynamic here in which AI is, of course, very useful.
25:05Statistically identifying patterns in large amounts of data can tell us things,
25:09and in a benevolent context we can act on those things for social benefit.
25:15However, I think one of the questions you asked previously about corporate use and about guardrails
25:24kind of gets to the heart of a real issue we're dealing with.
25:28These are very expensive technologies both to develop and to deploy.
25:33Inference, the process of using AI to get a result or a prompt, is much, much more expensive than traditional
25:40information retrieval.
25:41So we're looking at a very capital-intensive technology in which the infrastructure providers are, you know,
25:49have a kind of monopoly over start to finish AI development and deployment at this point.
25:55And the users of this technology are generally not us.
26:00We're talking about governments using it on their citizens.
26:04We're talking about militaries or police forces licensing it for, you know,
26:09different forms of law enforcement or social control.
26:12We're talking about employers licensing this to use on their employees
26:17or to justify displacing labor or degrading labor,
26:21something I think that, you know, the media industry is very familiar with.
26:24So we need to, I think, scratch the surface of this pronoun we
26:29and actually ask who is making these decisions.
26:32Who's, you know, again, who gets to use these technologies and who are they used on
26:36in addition to wrestling with the questions that Julia raises,
26:41which are really questions around should we buy the hype?
26:43And I think what we've seen over the last year in the kind of, you know, Microsoft-launched ChatGPT,
26:49everyone pretended AI was new even though it's very, very old.
26:53And a lot of weight has been put on that concept of AI.
26:59A lot of expectations that have us at, you know, certainly a kind of hype,
27:03you know, very, very high level of hype.
27:06That doesn't mean the technology is useless.
27:08But again, it means that most of these promises will not be kept.
27:12And we're going to see, in my prediction, a kind of deflationary turn because you can't, you know, you can't
27:17promise to be building God and then give me an incorrect answer in an email prompt and, you know, assume
27:24I'll continue to believe in your God.
27:28Yeah, I think two things here.
27:31The first is, as I mentioned before, you have a couple of people who are sitting around and who pretty
27:35much are imposing their vision for the future onto the rest of humanity.
27:39So I guess irrespective of whether or not AI is getting better, irrespective of whether or not it's living up
27:45to the hype, you have to interrogate who is controlling this.
27:48Why are we giving them that power?
27:50And, you know, why are they harvesting so much money, so many resources in the process of making that a
27:55reality?
27:55So that's like the first layer of the question.
27:57The second thing is, I mean, like, honestly, the answer to this is we don't know.
28:01And nobody can claim to know.
28:02Nobody can answer this with any degree of epistemic confidence because we truly don't know where this technology is headed.
28:09And to be honest, we have to approach that question with a great degree of humility and prepare our institutions
28:14to be resilient when so much is unknown and uncertain.
28:17I mean, there are these questions swirling around all the time.
28:19Will large language models get us to AGI?
28:22Will scaling laws soon entirely collapse?
28:24Is it possible that, you know, after a certain point, you could just be throwing together a bunch of compute
28:28and train the largest possible model that you can train and you're not really getting commensurate gains in capabilities?
28:35Yes, maybe that's possible.
28:36Maybe it's possible that you'll need some sort of new breakthrough for the hype to be fully realized.
28:41But at the same time, you can't miss the fact that these models are objectively a lot better than they
28:45used to be.
28:46And that rate of progress has been nonlinear.
28:48So the evidence is mixed.
28:50And again, as I mentioned before, nobody can answer this question with any degree of confidence.
28:54We really don't know where this technology is headed, what the next one year, five years, ten years will look
28:59like.
28:59So we have to approach this with humility.
29:01We have to approach this recognizing that we're inevitably going to experience a lot of uncertainty and learn to adapt
29:08to that kind of uncertainty without taking for granted that we're going to have an answer.
29:15Okay.
29:17I think, I mean, I really agree with the last answers.
29:21I think both feel sort of a cop out.
29:24But yes, you know, absolutely.
29:26The sort of pricing market, at least in the corner of the world I live in, which is sort of
29:32venture, is completely out of whack with reality.
29:35When you look at the pricing and valuations that many of these companies have raised.
29:41And so absolutely a bubble in that sense.
29:44And then on the other side, I think sort of the cycle of innovation and amazing researchers, talent, consumer, and
29:51corporate sort of eyeballs trying these solutions is happening at a pace that it wasn't just a few years ago.
29:58And so we'll see what the next years hold.
30:02You are working on the technology itself.
30:05So you must believe in it.
30:08Is it living up to its hype?
30:11Absolutely.
30:11I believe in the technology.
30:13But yes, there's a bubble.
30:14Of course, there's a bunch of hype.
30:17There is some funding rounds where I would never put my money.
30:22But this is working as intended.
30:25You may remember, if you're as old as me, everyone, you may remember the internet.
30:29And pets.com and all that, right?
30:32And the underlying insight was also true back then.
30:37The internet did change the world.
30:39And it created some of the most valuable companies out there.
30:43But of course, there was a lot of shenanigans and hype and completely overvalued companies.
30:48And this kind of now is an opportunity for smart entrepreneurs, for smart investors, and for dumb investors to lose
30:56their money and light a bunch of venture capital on fire.
31:00But this is fine.
31:01And this is the best way we have to find out what's going on and to find out how can
31:09we create value for everybody.
31:12And so I'm not concerned really about it.
31:15It's when you look at, for example, German government was a customer of ours.
31:20They have 400,000 open positions.
31:23And when you look at the demographic distribution, not only in Germany, we'll be falling off a cliff in the
31:29next years.
31:30It's really, I can really worry about this.
31:33So we need innovation.
31:34We need tech.
31:35And of course, we need to figure out how to best use it economically, but also ethically.
31:42And this is a process we're right in the middle of.
31:44And it's still miles better than crypto.
31:49Of course.
31:50But, you know, you talked about the internet.
31:56And, you know, no one at the beginning could imagine how it would change our lives.
32:02And, you know, we have certainly reaped many benefits from the internet.
32:06But also many, many unintended consequences.
32:10I was in the room in Half Bay Moon in California at a tech conference the day after the 2016
32:23election
32:25when Mark Zuckerberg stood up and said to the room, there is absolutely no way that Facebook had any role
32:33in, you know, in influencing the election.
32:37And we know that's not true.
32:39And yet we see, you know, we haven't learned from the past, it seems to me.
32:47Because the same tech companies are now racing ahead at great speed to get this technology out the door as
32:58fast as possible,
33:00whether it's ready for prime time or not.
33:02And I heard at another conference the chief economist for Microsoft, somebody asked him about unintended consequences.
33:11And he was like, well, you know, we just need to do it and then we'll fix it if things
33:18go wrong.
33:20But when we're talking about the kind of negative impacts that we could have on society, I think that's really
33:27dangerous.
33:28And I'm, you know, I wonder how concerned are the panelists by the news last week that, you know, the
33:35top people who were in charge of making sure that AI was safe at OpenAI have quit and said, you
33:44know, the company is just ignoring these guardrails.
33:49I am concerned.
33:51And I think, you know, I want to get back to this internet analogy because in some sense, yes, that's,
33:58you know, the internet or the commercial internet, our ability to access networked computation for commerce and communication shifted things
34:07radically.
34:08But the internet is a series of protocols and a method.
34:11It's not the business model built on top.
34:15It's not the surveillance advertising platforms that I talked about in, you know, my little talk before this panel.
34:22That was a choice that was made.
34:24Those were very particular visions for a very particular type of internet that is in no way natural.
34:30This is not the product of human innovation reaching an apex.
34:34These were decisions made by a very neoliberal government in the U.S. during the 1990s when manufacturing decline and
34:43an economic recession had people scrambling for some kind of answer to the U.S. economy.
34:51And, you know, in addition to a number of different forces.
34:54So I don't, I don't think we can simply take this as a natural analogy.
34:58Well, that, you know, that changed the world.
35:00It'll change it now.
35:01I think we need to recognize a number of the pathologies we are suffering from now as stemming from that
35:08decade.
35:08The unfettered surveillance that makes this the golden age of surveillance for corporations, for authoritarian governments, for law enforcement.
35:16And the fact that we have vanishingly few places for actual private communication, for autonomous decision making, as these often
35:26centralized technologies are infiltrating our workplace, our government and our social relationships.
35:34So that there's nothing natural about that.
35:36That can be changed.
35:38There is something very magical about the underlying methods, the ability to communicate instantly with someone across the globe.
35:45And before the Clinton administration locked in this regulatory vision for surveillance advertising as the engine of the tech economy.
35:53There were thousands of alternative visions for local networks built on top of global infrastructure, for a social wage that
36:01supported different applications, for the creation and collection of more mindful data about us in our communities that would reflect
36:08back at us answers to questions we asked intentionally.
36:11All of that was on the table.
36:13What was picked up from that table was a very particular set of regulatory proposals, like regulation by inaction, that
36:21led us to where we are today.
36:22And so I think we need to take some of that agency back and recognize, you know, kind of like
36:27Signal is doing, we can build it differently.
36:28We can push against that grain and we don't have to naturalize a vision handed down by multinational corporations.
36:35Staya, do you want to add to that?
36:37Staya, do you want to add to that?
36:37Yeah, I think absolutely I'm concerned.
36:39We've seen how our society has failed to rein in the power of exploitative social media platforms.
36:44As you mentioned, we've seen the impact on democracy.
36:47I've seen my peers become addicted to screens and be nudged to suicide.
36:50I mean, the impacts on the mental health of my generation, I think it's astounding.
36:55And the worst part is that lawmakers to this day are still working to write laws and to set rules
37:01of the road to hold these companies to account and to rein in their unchecked power.
37:05So definitely we've already seen some some massive failures of social media.
37:09And I fear that we're going to end up repeating our mistake with AI because we're investing so much power
37:13in the hands of a couple of people who pretty much are authoring the rest of humanity's fate.
37:18And we're seeing disinformation.
37:19We're seeing, you know, we're in a critical global election year and people are marching to the polls under a
37:24fog of disinformation.
37:25We're seeing deep fakes enable non-consensual revenge pornography.
37:29I mean, there are so many risks to be worried about right now.
37:31And I'm not saying that has to be the case.
37:33I mean, honestly, I'm young and so if I'm not optimistic, who would be?
37:37And I mean, I think there's so much potential in this technology.
37:40AI definitely could be the goose that lays the golden eggs.
37:43And there's so much that we can keep pushing for and trying to realize, but we're not on the right
37:48path right now.
37:49And I fear that we are just seeing a repeat of all the mistakes we made in the past with
37:54social media.
37:55So we have about five minutes left.
37:58And with those five minutes, I mean, we've heard about the downside, the dark side of AI.
38:05And, you know, at the same time, as Hanel mentioned, you know, if the incredible potential.
38:13So if AI, you know, can help us with the climate crisis, if it can help cure some of the
38:21worst diseases in the world, we have to do it.
38:25We can't abandon it.
38:27And we won't.
38:31So this is the crunch time where I want to ask the panelists, you know, if they believe, you know,
38:38that we're on the wrong path now, is there a way to get on the right path?
38:43And what are a few key action items that you'd like to see happen in the near term that will
38:52help ensure that we can have it all and an AI that is ethical and profitable and beneficial to society?
39:07So I'll start with you, Jonas.
39:09I don't think we are on the wrong path altogether.
39:13I think there's there's I mean, we exist.
39:16So that's a good that's a good sign.
39:19What I wish we would do more of is develop a vision out of strength.
39:25Don't be afraid.
39:26Don't think or don't let others fool us by thinking that this technology is so dangerous that only the world's
39:36biggest company can ever be entrusted with that.
39:39There's a lot we can do as Europe.
39:42But also this goes for every company.
39:45This goes for every country in the world.
39:48There's a lot of opportunity, but we should not be just afraid.
39:52We should build strength and we should shape the future because this is not just about value.
39:57It's also about liberal democracy.
39:59This technology goes to the very core of communication of thinking the world the next generations will grow up in
40:08will be built by AI.
40:10And if we don't have a positive vision that we are building towards, then eventually we'll basically just be the
40:17passenger, the paying passenger at the backseat, and we'll end up with the world we may not like.
40:23So I'm going to just ask you a quick follow-up, and that is if we want Europe to build
40:32an AI that reflects our values, what needs to happen?
40:38It's already happening.
40:40I mean, so what we need to do is we don't need more regulation.
40:45We don't need more panels.
40:46We don't need more smart people that basically shake each other's hands.
40:51We need to build.
40:53So we need to now think about everyone, and I think we're all the good guys, right?
40:59So we're all the good guys.
41:00We're all trying to make the world a little bit better.
41:02And every single person and every single company and every single government can now create a vision towards how can
41:10we build in a new era?
41:13How can we make the world a little bit better?
41:14This could be through personal choices, what kind of products we are using.
41:18This could be through entrepreneurial choices.
41:21How are we building new empires?
41:23What kind of new businesses can we create?
41:26And I think if we're doing this, then I'm super optimistic.
41:29Okay.
41:30Thank you.
41:31Hannah, what is your call to action?
41:34I will piggyback on your answer, which I thought was great, and just say more entrepreneurship and more building, right?
41:41Like more incredibly smart, driven, motivated, contrarian people shipping products, releasing them, writing code, finding a team, scraping something together,
41:55releasing it on the internet.
41:56I mean, I think this is the way forward and the way out and the way towards progress is just
42:02more fantastic entrepreneurs, not working for big companies, but starting their own ventures and coming together.
42:10Thanks.
42:11We're almost out of time.
42:12Sneha?
42:13I think if we want to build AI to solve problems, we should build AI to solve problems.
42:17We should not be building towards some kind of singular, all-knowing technology building towards some kind of machine god.
42:24If you want to build AI to help revolutionize education, let's do it.
42:27If you want to build AI to help revolutionize medicine, let's do it.
42:30I don't think we need to build any sort of single purpose technology to do all those things at once
42:34and ignore the dangers we could be creating in the process.
42:37Okay.
42:38Meredith, you have the last word.
42:39Yeah.
42:40Well, I agree with a lot of what I've heard at Signal.
42:43I know we certainly build every day and I think one of the answers adding to my co-panelists is
42:49we have to have the self-respect not to simply accept the vision that is handed to us by these
42:54companies.
42:55We can actually change things.
42:57Signal has done this.
42:58We have rewritten parts of the stack in order to reject the surveillance business model and we often offer those
43:05open source for other people to raise the bar.
43:08So I think we can transform the tech industry we have and actually, I think, realize some of the promises
43:15of the internet.
43:16But we can't do it if we naturalize the business model that was cemented in the 90s and is now
43:22causing so many problems in this volatile political age.
43:26Thank you very much.
43:28Let's have a nice round of applause for this fantastic group.
43:31Thank you.
43:31Your.
43:32Thank you very much.
43:34Thank you very much.
Commentaires