Siliconsciousness: Pope Leo Trolls Trump on AI…and much much more

right here on the DSR network.

“And there's no better way to enjoy the DSR network than by becoming a member.”

Members enjoying ad-free listening experience, access to our discord community, exclusive content, early episode access, and more. Who's code at DSR 26 for a 25% off discount on sign up at the DSR network.com. That's code at DSR 26 at the DSR network.com/buy.

Thank you and enjoy the show.

To sign up to the DSR network podcast focusing on the artificial intelligence revolution, politics, and policy. Hello and welcome to Silicon consciousness. I'm David Baroskov, your host and this week is everything we're going to talk about AI, but this week is special because we get to talk to one of our favorite smartest people we know. Alonja Nelson, as many of you know who've been joining us over the course of the past couple of years,

Professor Alonja Nelson is the Harold F. Linder Professor and founder and director of the science, technology and social values lab.

“Remember those at the Institute for Advanced Study.”

How are you doing today, Alonja? Great, David. It's good to see you. It's good to see you. You know, one of the reasons I like doing Silicon consciousness, I'll be honest, is I get to talk to you.

And, you know, let's just take a little bit of news and then we'll talk about a book that you've got out, which is super important. But let's just start with the breaking news because that's where these conversations typically start. And this one comes a little bit from left field, although it says something about where we are in the world,

because this weekend, the Pope, Pope Leo, the Pope from Chicago, put out his first big paper in cyclical.

And it was about AI and how we should deal with AI and how we should maintain human dignity and how we should be careful of AI. It was a very thoughtful document written entirely in Latin as you would expect from most AI documents. And I assume, because of your classical education that you read it in the Latin to begin. I am sorry to say, and my teachers will be very disappointed that I did not really read it in the Latin. I took Latin in the high school and my Catholic high school, but I read it most of it and not all of it in English.

And it was not only the document itself, but that it was actually a kind of, what we say in the 70s or 60s, you might have said, a happening, that people from lots of different walks of life, including my very Catholic mother, were all reading parts of this in cyclical or talking about it and talking about AI. So, you know, the Pope and the Vatican can't make laws for the rest of us, but they certainly held our attention and got lots of people talking about the stakes of AI.

“Yeah, and I think one of the things that is interesting about this is that, well, I mean, take it when you were first dealing with AI policy issues in the US government.”

That was a few years ago. And AI was considered a niche within an niche. There was tech and within tech, there was AI, and it was for big brains, and most people didn't understand it. It was something to watch, but nobody really quite had their brains around it. And now, the Pope thinks it's so mainstream that he makes it the subject of his first in cyclical and in American elections running into that where, you know, less than six months away from midterm elections, you have candidate for candidate who's talking about AI in real mainstream kitchen table terms, AI and jobs, AI and human dignity, AI and education.

And it's just, it's astonishing how fast this has happened, but I take from the people in cyclical the fact that, you know, this genius not going back in the bottle, this is now a permanent part of the discussion of how human beings on this planet live their lives. I definitely agree with that, and of course the Catholic Church is global, it's a global institution. I think also quite interesting is that the encyclical is really addressed to a public audience, so it's an external document, it's a document that's intended to be addressed in very plain language as you're setting in kitchen table kind of framing for people, for regular people, ordinary people about the implications of this for their life.

And that, you know, arguing that AI is different, but also, you know, really having this global address to a broad set of stakeholders, and so, you know, I am a couple of years ago served on this this UN body on AI governance, and you know, I think the UN imagines itself to have a kind of global address, but certainly a document from the Vatican does.

“You mentioned it was translated into English from Malatin, I imagine it was probably translated into a lot of other of the world's major languages besides.”

And so we are now having a global and the planetary conversation about the stakes of AI.

And what I appreciated about the document and a few things, I mean, one, you know, Pope Leo says, as many of us know that technologies aren't neutral, and that like the outcomes that we might want and hope for for AI.

“And he's, and he's, you know, not an optimist, but he certainly not a pessimist about AI that we've got to have a role, we as a sort of global community, and making those outcomes come true that, but AI for good is is not just going to happen.”

And also very much appreciated that his spectrum of issues that we needed to think about range from whether or not people could support themselves and have livelihood in the job market.

So is AI going to sort of help to expand opportunities for people or can strain them? Or is it, you know, or should we be worried about autonomous weapons and warfare? And he even has, you know, a line in which he's sort of saying, we might need to revisit the Catholic Church's approach to what just war is, because obviously, you know, it's failed us and we're not living in a time of sort of just warfare. So, you know, the spectrum of not only did he talk plainly to a kind of planetary public, he laid out the full spectrum, I think, of issues that many of us are both concerned about and, you know, some eagerly anticipating with regards to the benefits of AI.

I have recall back in one of our original conversations about all of this, we were talking about the impact of AI in the first world in the emerging world, the global south and so forth. And I seem to recall that one of the points that we touched upon was that what would have a lot of discussions about technology of AI and bi-technologists about AI and some discussions about markets and investments in AI or labor and AI. It's happening so fast that we haven't really been able to get in the discussions about the philosophy that we ought to embrace with AI and that in fact, you know, technologies should be made to serve human goals and work within human values.

To me that one of the things that's encouraging about this, and I don't read too much into it, but one of the things that's encouraging is there is a philosophical discussion going on right now, which is much needed and I just want to give, do you see it taking a roots elsewhere, do you see people beginning to try to grapple with these things and not just say, what is AI, but start to say what should AI be. Yeah, I think what Pope Lee have did is in part really set the table for that conversation because, you know, we're told that we can't even wrap our minds around it because it's so far in the future, so powerful, the singularity, sky net sort of pick your, you know, post human sort of sort of framing.

And, you know, if you're told that, like, where could you, where can you begin to ask questions, where do you feel like you can enter a conversation about what AI should be, what is the relationship between AI and humanity, you know, the pope sort of said sort of very clearly that humans have a soul, like AI doesn't have a soul that AI are tools that, you know, technologies are tools that we should be thinking about how we should use them.

“And so I think David, even to call it philosophy, although, of course, the, the, and cyclical, the what I read of it is filled with philosophy, not only theology and filled with references to sort of, you know, ancient and grease and, you know,”

fill us off, the law differs, but it also, again, is making it sort of like every day philosophy about, you know, what AI means for, for people's lives. And I think that is to be able to sort of, I think, rest, WREST, some of the attention away from far off futures, you know, sort of manifesto is about, you know, what the world must be. And bring it back to things that really matter for people's lives, including people's concerns for, you know, for their children and for, you know, how these technologies might be for used for good has definitely, you know,

Yeah, and of course, one of the things that Bob Leo does and I'll admit it, I find it appealing, some people may not, is that he is the master troll of Donald Trump.

“I mean, he has just on a regular basis, you know, he appoints an immigrant to be the bishop of West Virginia.”

He, you know, he has provided counterpoint to the critiques that he's gotten from JD Vance and Pete Higgs said about just wars matter effect. And he's just all done this extremely deftly and in a way that does not in any way diminish his office and actually adds to the kind of moral resonances got one of the things that I found kind of interesting was the juxtaposition of the pope putting this out.

“In the same league that the president was about to release a new AI strategy that talked about giving the government the ability to review some AI technologies before they were released presumably to ensure that they were safe.”

And just as he's about to release it and all the press releases are out and everything he gets a phone call apparently from David Sachs is AI guru who's also, you know, happens to be running, you know, venture capital fund investing in AI, no conflict of interest.

Anyway, so he gets a call and and the thrust of the call as we are led to believe was hey, you know, the people in the AI sector don't think this is such a good idea.

It's going to put the country behind if we add in this kind of layer of deciding whether the AI is safe in our race for AI supremacy. And so Trump goes, okay, we're not going to release it. And I was just wondering what your thoughts were about this as you've been involved in these kind of discussions at that level before. So a couple of things, I mean, on the one hand, as you've laid out beautifully, you have the pope who's advancing sweeping language real sort of bright lines about how technology is, you know, should be used and should not be used what we should be who should be involved in the conversations about the technology, you know,

that speaks, you know, that wears it's sort of philosophy on its sleeve, it's theology on its sleeve. On the other hand, I think, you know, what we saw with the Trump administration sort of backing out from this executive order is what is spoken and we've talked about this before, David, spoken as, you know, we just can't do this because we just can't regulate these technologies, right?

“Like that, you know, I think the president, the paraphrase was something like, you know, we just don't want to go there and we don't want to get in the ways of the companies, the sort of, you know, now sort of, you know,”

the Trump administration also has a philosophy of AI, but doesn't speak itself as a philosophy, it sort of speaks itself as doing nothing at all, when in fact, the administration's doing quite a lot around AI. So that even pulling out from the executive order is sort of saying, then our philosophy is that it is perfectly clear to you. You know, American companies to release quite risky technologies, not only to American companies, but to the world, and, you know, we don't have anything to say about that. So that is to say that the philosophy is that we think risky, okay, you know, risky AI should be the sort of priority of the United States in a certain way.

And, you know, as we discussed before, this is also happening in a week where we had some reporting, I think it was in the New York Times, that, that the Trump administration is taking more stakeholder shares on the behalf of the American public and more technology companies, in addition to to Intel, which happened last year. And we're doing all sorts of things that is giving kind of shape and form, the Trump administration to what the AI technologies look like, while acting like it's sort of this natural state of evolution, that, you know, these are just to go back to, I think, a point, another point you might say, to use your framing of the, the, the pope trolling president Trump is, you know, again, the technologies aren't neutral.

And that's just sort of leave them to grow in their natural evolutionary state. We're making decisions all the time about how the technologies will sort of exist in the world. And so, you know, I think what the in cyclical also does is really cause a call, you know, bring into relief the fact that sort of saying that you're not going to have an executive order is, is not the same as saying that the technologies are neutral or that you're not doing something about them.

Really having a very different from the encyclical philosophy and theory of what AI is and how it should be used in the world.

It's kind of Shakespeare. I don't know what it is. It's, it's, it's got kind of his story. It's difficult. Yeah, maybe it's biblical. It's got, it's got a lot of heft to it. It is not necessarily battle between light and darkness, but it is a mannequin kind of a battle go look up mannequin for those of you who don't have the classical education that I'll under it. But what it does is it says, look, we are all in a difficult position.

We've got this powerful technology contrary to what a bunch of people want. It's not going away. We can't legislate it out of existence.

It's going to be everywhere and indeed it's a hundred things. It's not just one thing. It's a thousand things. It affects jobs, it affects how markets work, it affects how more works, it affects how society works, it affects how politics works. It's going to be embedded in everything and we need to develop a vocabulary, but we also need to develop our own ability to assess it. And you think, well, if only there were a book that could help us assess AI for ourselves and add expert views. And then you find out that MIT Press is coming out of a book called Auditing AI, where a team of co-authors featuring a laundering Nelson, our friend here, is helping to provide that.

“And I think it's got, I don't know, several dozen perspectives on how to assess AI in different”

things. So, you know, incarnations and ways it can affect our lives. And the early response to the book has been fantastic. And that's enough of me talking about it since you're one of the co-authors. Perhaps you could tell us a little bit about it and the the raison d'etra.

Thank you, David. Well, a plus for that segue from the pope, from the the people in cyclical to an MIT Press book on AI, perfect. It's this is not my first radio.

“So, well, I was really pleased to have the chance to work on this book with computer scientists, legal scholars, legal advocates, other folks who are social scientists like me.”

It is an open access book.

So, you can buy it as part of the MIT Essential Knowledge series, or you can actually download it online.

And, you know, part of what we're trying to do is, you know, help the public understand that there are things as we're being told. That there's absolutely nothing that we can do, that these technologies are being used and growing parts of our lives and that there's literally nothing we can do to know how they're being used or to sort of understand what's happening. And really the point of the book is that we already have a proven tool for accountability, which is called auditing. It comes out of a period of time that before we even had algorithms from the 50s and 60s where we were trying to understand how to do systematic kind of evidence.

Based investigations of systems behavior.

“So, you might vary the inputs, and then you look at the outputs, and you kind of measure the results against a standard, right?”

So, it might be a safety benchmark, which is a lot of what we're doing around AI, or in the 60s it could have been a civil rights law. It could be today an accuracy threshold, so it's systematic, it's evidence-based, it's rigorous, and it's an empirical. And critically, given that from the era of social media to now, we don't always have access to what's happening inside underneath the algorithm, what's happening inside these large technology companies. And it can be done in a way that doesn't allow you to necessarily have to see what's inside the guts of the sort of algorithm or the black box, right?

That would obviously be ideal, but it doesn't have to happen that way.

And one of the first that we write about, and I think to our to the best that we could discern, the first regulation that we have at the federal level that uses the word algorithm is from 1983.

And it's the case of the Sabra Airline Reservation System. In this instance, American Airlines was caught or assumed to have been caught using a biased reservation system, which, and it was produced by American Airlines, but it was used by the entire industry. And what it was doing, you won't be surprised to hear, surfacing its own flights, burst, and search results if you were a consumer looking for a flight. So it was claiming, you know, the claim was that it, you know, other competitors were used losing millions, tens of millions a day, and American was gaining millions.

And demonstrate that it was biased in favor of itself.

“And the civilian, the civil, aeronautics board, responded by banning preferential display treatment.”

And, you know, this is something that we still work with, said, you were challenged with today, like should we have net neutrality or not. And not only in the use of internet technologies and broadband, but, you know, there was a recent paper by my former OSTP colleague, Assad Brahms and Alley. That said, that maybe we need, you know, this kind of neutrality, even with regards to sort of APIs for chapats for AI. So this is one of the kind of oldest algorithms that we're able to be done, like looking at inputs and outputs.

But, and but, and it's also an example that audits and regulations have to work together.

So it's, you know, if you just have transparency and we can just know that the systems are biased, which we know today about some of the advanced AI. Um, advanced AI systems, um, having the knowledge without the regulation, um, doesn't really help us. But audits do give us a way to, to look at systems that we're being told, um, that we can't look into at all. They don't guarantee accountability, um, and they've got to be constantly sort of updated as the systems are. But they give us a way and to begin to have a kind of scalable accountability tool that most people can, can really appreciate and understand.

This podcast is underwritten and part by the US Embassy of the United Arab Emirates. It's editorial content is completely independent and the views expressed are exclusively those of participating experts. It is presented live without editing. For further information about the UAE's efforts in the areas of artificial intelligence and technology, go to the website of the embassy at www.ue-embusy.org and search for UAE-us-tech cooperation. We thank them for their support.

We thank everybody who's supporting this podcast for their support, and we look forward to it developing and growing.

“Over time, because the issue is so important.”

Yeah, you know, it's interesting to me that it sort of resonates because you talk about going back to 1983 and the first use of the word algorithm, you know, regulation and it had to do with the American Airlines Saber system. But of course, the word algorithm goes back to the ninth century. And to a Persian mathematician, a brilliant Persian guy named Muhammad Ibn Musa al-Qawri's me. And his name was used to create the word algebra, algebra, but also algorithm, right?

And with regard to algorithm, it's basically, you know, sort of any step-by-step procedure for solving problems.

And hence, you know, there is a natural arc from that to these complex systems we're now developing to solve problems to how we manage those systems and sort of we need algorithms to deal with our algorithms. And I'm just wondering in looking at the other contributions to the book. You know, what were you or big takeaways? What coming from these other authors did you say, oh, you know, this is emerging as an issue. That is, you know, that the average reader is going to have see this as a headline.

“Yeah, so I think, you know, we started in 1983, I think the introduction to the book starts right there, but there's also been kind of other important audits that people have.”

I think that readers will be familiar with. So there was an audit that joy blew a meaning and ten that Gebur did in 2018 about facial recognition technology and how it was biased. And it's, and it's accuracy with regards to people with darker skin and particularly women with darker skin. We also have audits were done of the Facebook housing ads over, you know, a seven year period from 2016 to 2022. And this was done not, you know, so the, the facial recognition technology.

It was done by two academic researchers, right, but the face, the, the Facebook housing ads was done by journalists, but at ProPublica that revealed that, you know, Facebook's ad targeting interface was allowing advertisers to include or exclude audiences based on, on race and ethnicity.

You know, have been, have been kind of an important and enduring tool for us across the, you know, the interdaining years up into the present.

“I would also say that, you know, we don't offer the audit as like the only thing we need to do, right.”

I mean, what, you know, we need a very comprehensive response to get sort of responsible, governmental to AI governance and, you know, audits are just one of our tools that sits with, you know, risk assessments should we be. Doing, you know, that they, one of the sort of theory or one of the proposals that was floating around over the last couple of weeks. That was going to land in President Trump's executive order that didn't come to pass was giving the, what we used to be called the US AI Safety Institute, which is now called, I think, the Center for AI Innovation.

Advanced access to, you know, very capable algorithmic systems to take a look at them, so you can do pre-dependent risk assessment. You can do kind of risk assessment after a model has been released over the kind of delay of, of, of its use. And so, you know, there's lots of different things that we need to be doing and audits are one of them and audits are already really familiar to us. So I just also offer it as something that makes sense to people that folks are familiar with and when they. And I think when any of us think that getting a handle on a system and as powerful as these new as AI tools and systems and companies as powerful as, you know, Google and anthropic and open AI seems daunting, you know, this is kind of a first step that can be offered one tool among many.

That's not a replacement for steps into regulation, which is what we, which we also need, but certainly is a place to begin.

“It's a way in which we start to see some of us as experts, but it also can is, you know, audits also allow in, I think, non experts to sort of think about the, the reliability of these systems as well, and it's an important first step.”

But, you know, one of the things that is associated with AI and many people's site as a challenge amid many challenges that are associated with AI has to do with its impact on critical thinking.

And average individuals need to find a way through a process, if it's not called an audit, it's not called an audit to evaluate the information that they are getting and to evaluate the impact of a tool on their lives. And like last week, I got to email from one of my daughters, who's very smart and very. I'm an active, professionally, successful person and it said, beware of AI psychosis. I don't know why she directed that at me, but you know why she would think that that might be affecting me, but then I read, you know, yesterday the day before an article about.

I was looking at the way AI is in the algorithm designed to sort of make you feel better about what you're doing and flattery. And so, and so, you know, if you had a AI, you know, you fed your book auditing AI into a bit, you know, sort of big traditional LLM AI. So come back and I would say, oh, laundry, you guys are so brilliant. And this is such a great way to approach this book, get it. Emphasizes everything.

And, and the problem is that people are going, you know, should I break up with my wife.

And AI is going well, you know, she's done, she's done you wrong, and I was looking at your checkbook and and you're like, and people are following it, you know, they're doing it. I just want to make the connection here between organizations and governments auditing AI. And the burden that falls on individual users now and how they have to come up with their own kind of personalized approaches.

“Yeah, no, I think that's right. And I think, you know, what, you know, so we know that over the last couple of decades, trusted governments been declining certainly in the last few years, but what is government for.”

It is to not exactly, as you say, have the burden of things like this rest on us as individuals to figure out, you know, how not to have people who are, for example, addicted to the flattery and sick of fancy of AI systems. And so, you know, I think that we need to, you know, I think where the audit can be useful is that, you know, people can, I think, look at different ways that they prompt the tool, although they're going to be different to when the responses.

Even as we're talking about prompting, I mean, you know, companies, I think are encouraging, you know, prompting strategies, asking about your wife, for example, that are asking you to talk to talk to, you know, an AI system and a very kind of humanistic way, as you would talk to a person.

As opposed to doing a kind of prompting that as much more kind of systematic and would ask you to say, you know, I know, you know, that would assume, you know, I know that you are not a human being.

And I'm asking a question and I want, I want a frank assessment, I want, you know, an assessment that is most true, not the one that is intended to keep me most on this system, so I can use the most tokens that, you know,

“produce potentially the most profit for a company. So, I think, you know, you can think about it as an actual tool that you can use to look at the outputs of a system.”

But you might also think about it as a kind of posture that we all need to bring with these technologies while we're really waiting, awaiting the systematic response that we need.

That the Trump administration is failing to provide and that hopefully, you know, raise the alarm about this weekend.

Yeah, and I think, you know, for those who are listening or more active in AI and listening to some of our discussions earlier about world models and things that that have profound complexity added to them.

“I think, you know, they also need to acknowledge that in in big evolving complex systems using next generation types of AI, they're dealing with millions, millions of variables and millions of different kinds of scenarios and, you know,”

the gigabytes and terabytes of data that boggle the mind that within the formulas within the algorithms, there can be millions of biases of different sorts. And that the more complex the systems get, the harder it is to sort out where, you know, the the the creators of the algorithm or even the the algorithm can be created by machine are putting their thumbs on the scale of outcomes.

And so this this this issue is only going to grow more complex as we deal with more complex AI systems correct.

“Yeah, totally agree with that, but I want to guess I would also add that complexity should not be an excuse or justification for not dealing with bad outcomes, right?”

So it doesn't you know, in some ways it doesn't matter if you're thinking about the rule of law or people civil rights or civil liberties, for example, or, you know, just war theory, what the sort of causal mechanism within increasingly complex our algorithmic systems are, right? The outcomes are just or they're not just either the outcomes are fair to consumer or not, you know, etc. And so, you know, I, I, you know, I often try to caution us against. I think driven about how we want to govern AI by capability alone, so there's a kind of philosophy, I think of of AI governance that's like the technologies to just keep getting more capable and more complex. And so we have to figure out increasingly more complex ways of approaching them as opposed to you.

And this is one of the things I appreciated about the people in cyclical sort of creating just some like benchmarks, right, like social benchmarks moral benchmarks, right, not just technical benchmarks for AI systems. That sort of say they should do this this thing and not that thing and we don't know what's happening in the black box, but we're asking, you know, colleagues who, you know, Chris Ola who spoke yesterday at the kind of conference after the Encyclical was was released one of the co-founders of AI.

You know, he's an expert on interpretability, you know, we don't know how to interpret all the things that these systems are doing. But you had in this, in this end-thropic co-founder also someone saying that there's much much more that we could be doing. And so I think, you know, auditing becomes one of the tools if we're trying to do more than we're doing right now and there's a big window of opportunity there given that we're not doing much. You know, even in that in the, you know, in the context or against the backdrop of increasing the more complex models.

I mean, I hope that we will succeed John McCone and others with world models. And part because right now our attention and of course our market and all the financialization and all the new market deals and speculation have been captured by generative AI.

I think would be really important for a kind of regulatory marketplace in addition to a marketplace of ideas and not to mention the sort of economic speculation.

You know, I'm also a big supporter of open source models that are increasingly, you know, being used by on the enterprise level both in the United States and around the world. And these are models that, you know, allow us to know a lot more about their, their inner workings, even if we can't explain everything that they do.

“So we are very much at the beginning. And I think you weren't saying this, but I would certainly want to push back against sort of any sort of claim that the increasing complexity that's happening every day.”

And the world's engineering and computer science aren't excuse or justification to not have responsible models or an AI governance and forms of regulation.

Yeah, well, I'm going to, I would go a step further, which is, you know, when I had a conversation with a guy who's involved in this kind of stuff who's in a company that's using world models as a way to explore biological phenomenon have the rugs effect cells and so on and so forth. And they're building incredibly complex simulations that could also be super personalized. How does the drug behave within your.

“Um, a particular set of, of cells and data and so forth, but what, what struck me and I saw this as kind of.”

And I was kind of revelatory for me, although it may be obvious to, to thoughtful people like you, but is that we're entering an era in which the previously unimaginable in terms of complexity is going to be grappled with daily. And so for, you know, as a social scientist, you know, that, you know, what social scientist political scientists would say, well, let's predict what the following outcome is and we're going to create an algorithm and it's going to have eight variables in it and we'll try to get those eight variables and.

Yes, and right, right, but eight variables in a world in which there are millions of millions of variables and so you know from the get go that it's going to be.

“It's going to be the root of proxination of the simulations, but we're going to enter an era in which.”

And so we're going to be a critical deep complexity where we're, we're entering worlds where not only couldn't we imagine that, but we can't imagine it only machine assisted assessments are going to be enabled us to to imagine it. But, you know, it's going to create, I, I can, I'm concerned like you are labor dislocations and so forth associated with AI, but one area where I do see a lot of applications is where people are going to enter into the, the harnessing of these new simulations, the harnessing of the knowledge of this deep complexity and that it's going to be whole new worlds, right, but similarly.

To suggest to me that the auditing is going to have to be a constant process because every new algorithm, every new system, every new layer of complexity creates the possibility of embedding within itself. And so problems, biases, errors, and so forth. And, and it's not going to be static. It's not like, oh, yeah, I got, I mean, it's not even going to be like we're used to, right, where you goods, you know, I have, you know, Microsoft, you know, 25.1 and six months from now, then I have 25.2 it's going to be constant.

It's constant in a million. Yeah, right, and so that's going to create this, this area of auditing is is actually a burgeoning exploded kind of challenge we're going to face and regulators never had to do that, you know, it was like, okay, does the plan of two wings, okay, you know, can it, you know, good, you know, that we're going to prove the airplane because it is doing. This, this is going to have to be a constant process, and it's going to have to be a machine-aided process. For sure, I mean, all that. It's, you know, so, you know, I come to Winkie Manu and I, because I worked for sort of 12 or 15 years prior on the sort of and your mouth of the human genome project, right, which were some of our, some of the very first big, what we used to call big data, right, you know, that I think any computer science is working on complex systems right now would scoff at the point that the fact that we thought.

You know, 30,000 base pairs was, you know, big data, but you know, here we were. And, you know, those trying to sort of get a handle on the data that came out of the human genome project gave us some of our first, you know, really large, you know,

Even in the technology companies, like, you know, formerly Facebook, now meta, you know, you had sort of engineers doing sort of a, the testing, you know, kind of in real time sort of seeing what people liked, what they didn't like. So, we've already been sort of moving to a, to a, an acceleration and our algorithmic systems that I think right now, certainly with simulations, you can just be, you know, automating kind of science and potential insights, you know, sort of almost constantly. And certainly one of the challenges for, and I think why auditing isn't important in this moment is because, if it is the case, as we know, that, you know, AI companies, AI labs are sort of constantly sort of updating, we might think of some examples that are familiar that are clear to us because we see them happening in real time. So if we think about what happens with GROC on XAI, right, so we, you know, we had the, the mega hitler example. So, GROC is sort of spewing out of this, but we're not going to do that, and we're not going to do that.

We, you know, we had the, the mega hitler example. So GROC is sort of spewing out anti-Semitism and then, you know, users on X, say, you know, what's going on on Elon and sort of Elon Musk says, oh, we're going to fix that, right. And a way that you can kind of think of that as a social audit, right, you sort of say, this system is outputting discriminatory, you know, things and we are going to go and sort of tune things tune the knob a bit to use a metaphor for very more complex processes. And it's not going to do that anymore. So, you know, companies are doing forms of auditing all the time, but to your point about it and being to sort of be constant and live longer, even with things like world models, like we don't actually know what happens with them until they live in the society until you actually use them. So you can do quite a lot and a laboratory setting, you can do quite a lot of an assimilation setting to try to anticipate.

“And that's not even, you know, I think with the AI race that companies are not probably not even spending as much time as they could.”

And it's trying to anticipate some of these harms, but even if you do all of that, they live in the world and different things will happen. And so auditing is always going to be a really important part of that. Moreover, you know, I think we, you know, we, what we need kind of a new kind of tier of democratic institutions and that institutions like, you know, there was a nice report over the weekend about the UK AI safety institute now called security institute, which has been thriving since 2023.

And our parallel instituted in the United States has been without a leader for a year and is severely underfunded, so it's not writing in the same way.

“But you could imagine both of these are starting in 2023 as different as, as kinds of democratic institutions and new kinds of institutions that we need to be able to sort of govern AI at all and potentially govern it well.”

It's really about understanding what's happening, you know, with the technologies to the extent that we can and this instance, quite upstream before they're released or as they're released.

So, you know, we can imagine also lots of auditing, you know, companies they could be nonprofits, they could be in civil society, they can certainly be for profit.

But we're going to need to create all of these different institutional spaces as part of a way of like robustly governing AI.

“Yeah, and they're also going to have to come at it from a lot of different angles, because, you know, you talk about a social issue like racial prejudice, that's an angle.”

Obviously, when you are dealing with an issue like the world models that, you know, might influence how a drug is developed or ruled out there's health models that are issues.

There are market issues that are involved because one could easily imagine people building systems that have biases for their own products, right, or that biases for a certain kind. They are American Airlines example. Yeah. Well, right, and it's also diet coke, where, you know, you're most people don't know what's in the can and they don't realize that when they're drinking it, it makes them want more of it, right, and and that kind of thing can also be put into it. And so, in an era of deep complexity, you're going to need, you know, lenses that allow you to see within it, and you need those lenses to be constructed in a way that conforms with one's philosophical world view.

And if one does not have a worldview that says, you know, monopolies are bad, then then you're going to end up with these things promoting monopolistic behavior.

That's not the answer, right, that. I mean, the defaulting to, you know, a couple self-described geniuses and, you know, the tools that they have developed to keep themselves in power is not the answer. So, again, you know, you and I are going to have to have a lot more conversations because I do think that the implications of this for how democracy develops, how governments develop, how regulations develops, how people deal with that, what how much they can trust the government, how much they have to trust themselves.

And how do they deal with arbitrage is between governments overseas and the governments at home, between the federal and the state level. And so all of this is is is is is is making this an area that's going to demand a lot of analysis for a long time to come.

“I encourage everybody to start with this new book from the MIT press and Alondra and a bunch of other expert co-authors called Auditing AI. And of course, come back here to still consciousness follow what Alondra is writing because she is prolific.”

And why is and we'll try to coax her to come back periodically because we love this conversation. >> Anytime. >> For now, thank you very much Alondra. Thank everybody for listening and bye-bye.

>> This was Siliconjustness, a production of the DSR network.