Will AI Increase Freedom or Help Authoritarians?

0:00Today on the AI Breakdown, we're asking whether AI is more likely to promote freedom or help authoritarians. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information. Hello, friends. Welcome to this special holiday edition of the AI Breakdown. Now today, going along the themes of July 4th and the American Independence Day, we're asking what I think is one of the big, important society-level questions of AI. Before we get into that, however, if you are enjoying the show, I would ask that you do this one thing for me.

0:36One of the things that helps new shows get discovered the most, which helps them rise in the ranks, is reviews on Apple Podcasts. It would be so helpful for you guys who are enjoying the show to leave a five-star rating and just a couple words about what you like about the show. I would be endlessly appreciative, but with that, let's get to today's episode. Welcome back to the AI Breakdown. Today is July 4th, and that's a chance in my mind not only to celebrate America or American independence, but also to ask questions of the lessons of history to try to understand how the issues we face in the present might shape our future.

1:13Now, it will come as no surprise that when I look at the artificial intelligence space, I'm not wondering about it just from the standpoint of technology or business, but I'm also interested in the big questions of how it might shape the world going forward. So today, what we're doing for this special holiday edition of the AI Breakdown is asking whether AI can increase human freedom or whether it's doomed to simply aid authoritarians. Now, I will say as a starting point that if you are in the camp that AI is a tool for aiding human freedom, the broad view as represented in articles and tweets and other internet artifacts is not one that looks positive when it comes to AI and human freedom.

1:55Today, we're going to dig into why that might be and what I think some of the counter arguments are. I want to kick off our discussion with an excerpt from a piece written by George Soros last month called Can Democracy Survive the Polycrisis? It's short and I think does a good job of capturing the negative or concern side of this argument. Soros writes, we're living in troubled times. Too much is happening too fast. People are confused. The Columbia University economic historian Adam Tooze has indeed popularized a word for it. He calls it the polycrisis. The polycrisis has many sources. In my opinion, the main source of the polycrisis afflicting the world today is artificial intelligence.

2:33Now, Soros lists climate change second in the Russian invasion of Ukraine as third, but obviously for our purposes today, we're just going to read the section on artificial intelligence. Soros writes, AI shocked the world when Microsoft made ChatGPT freely available to the public through an associated company called OpenAI. That was in November 2022. ChatGPT posed an existential threat to Google's business model. Google went into overdrive to release a competing product as soon as possible. Shortly thereafter, Jeffrey Hinton, who is generally considered the godfather of AI, resigned from Google so that he could speak openly about the risks posed by the new technology.

3:06Reversing his previous position, he took a very dim view of AI. He said that it could destroy our civilization. Hinton pioneered the development of neural networks that can understand and generate language and learn skills by analyzing data. As the data grew, so did the capacity of AI's so-called large language models. This made a big impression on Hinton. He said, Maybe what's going on in these systems is actually a lot better than what's going on in the brain. As they become more powerful, they also become more dangerous, he claimed. In particular, he warned against fully autonomous weapon systems, killer robots, he called them.

3:35Soros quotes Hinton as saying, We've entered completely unknown territory. We're capable of building machines that are stronger than ourselves, but we're still in control. But what if we develop machines that are smarter than us? It will take AI between 5 and 20 years to surpass human intelligence, and it will soon realize that it achieves its goals better if it becomes more powerful. Soros continues, What Hinton said made a big impression on me. Indeed, AI reminded me of Goethe's poem, The Sorcerer's Apprentice. The apprentice is studying magic, but doesn't fully understand what the master is teaching him.

4:04When the master orders him to sweep the floor, he applies the magic words to a broom. The broom obeys him, but the apprentice can't stop the broom from fetching buckets of water to sweep the floor, and the house gets flooded. I grew up before AI was invented. That made me a great believer in reality. I realized at a relatively early age how difficult it is to understand the world I was born into, and I looked to reality to provide me with moral guidance. We, human beings, are both participants and observers in the world in which we live. As participants, we want to change the world in our favor.

4:30As observers, we want to understand reality as it is. These two objectives interfere with each other. I regard this as an important insight which allows me to distinguish between right and wrong. AI destroyed this simple schema because it has absolutely nothing to do with reality. AI creates its own reality, and when that artificial reality fails to correspond to the real world, which happens quite often, it is discarded as hallucination. This made me almost instinctively opposed to AI, and I wholeheartedly agree with the experts who argue that it needs to be regulated. But the regulations have to be globally enforceable because the incentive to cheat is too great.

5:02Those who evade the regulations gain an unfair advantage. Unfortunately, global regulations are unattainable because the world is dominated by a conflict between two systems of governance which are diametrically opposed to each other. They have radically different views on what needs to be regulated and why. I refer to the two systems of governance as open and closed societies. I define the difference between the two as follows. In an open society, the role of the state is to defend the freedom of the individual. In a closed society, the role of the individual is to serve the interests of the rulers.

5:28AI is developing incredibly fast, and it is impossible for ordinary human intelligence to fully understand it. Nobody can predict where it will take us, but we can be sure of one thing. AI helps closed societies and poses a mortal threat to open societies. That's because AI is particularly good at producing instruments of control that help closed societies to surveil their subjects. This is why I am instinctively opposed to AI, but I don't know how it can be stopped. Right now, nobody else does either, but most of those who developed AI recognize the need to regulate it. So does Congress and President Joe Biden's administration.

6:00But AI is moving much faster than government authorities. The Biden administration has taken some executive action, but Congress will have difficulties in enacting anything like an AI Bill of Rights. There is, however, a problem that cannot wait. There will be general elections in the United States in 2024, and most likely in the United Kingdom as well. and AI will undoubtedly play an important role, one which is unlikely to be anything but dangerous. AI is very good at producing disinformation and deepfakes, and there will be many malicious actors. What can we do about it? I don't have the answer, but I hope this issue will receive the attention it deserves.

6:31All right, so back to NLW here, and there are a couple things that I want to make note of. First is Taurus's language of reality. He said, I grew up before AI was invented. This made me a great believer in reality. I realized at a relatively early age how difficult it is to understand the world I was born into, and I look to reality to provide me with moral guidance. Now, this idea that there is a single reality is something that is being challenged more than just in the context of AI. John Esconis is a professor at Catholic University who's been writing a series of long-form pieces for the New Atlantis called What Happened to Consensus Reality?

7:05I actually had him on the main AI breakdown a couple months ago if you want to go check it out, but the long and short of it is that he believes effectively since the beginning of the internet, We have been moving away from the historical accident of a single consensus reality that was created, shaped, and maintained by a dominant central infrastructure such as the evening nightly news. I don't want to get too deep into it here, but it does matter in the context of people having a sense that reality as they know it is being threatened. It's a little bit difficult to disentangle, I think, how much of that is actual misinformation versus how much of that is just the inexorable shift away from the idea of there being a single consensus reality.

7:41Now, a second piece that I want to call out is Soros' almost visceral opposition to AI. He writes that AI has absolutely nothing to do with reality, and that made him instinctively opposed to AI. Now, of course, everyone brings their priors to any discussion, and AI is no different, but it is notable how deeply rooted, or maybe put differently, non-objective, non-intellectual, instead functioning on a gut level, this reaction is. But of course, Soros is just one person, and he comes from a very particular perspective. So to try to broaden this conversation, I polled my followers on Twitter. I asked two questions.

8:17The first was, is AI more likely to enable authoritarians or promote freedom? With around 350 votes, 75.9 % said enable authoritarians and 24.1 % said promote freedom. So clearly this skepticism that you're seeing in Soros is shared in substance, if not in specifics. Now the second question I asked, question two, are democracies or autocracies better positioned when it comes to generative AI? Interestingly, in that poll, 60.2 % said democracies and 39.8 % said autocracies. So in these two answers, we're seeing something of a tension. On the one hand, there seems to be a sense that AI can help authoritarians.

8:57We're going to get into a lot of reasons why that might be. But at the same time, the answers to that second question also suggest a sense that generative AI, which one might also frame as the democratization of the means of production, might be better suited to supporting democracies than to autocracies which rely on command and control. So now, with both that poll in mind as well as Soros' comments, let's talk about what reasons one might have to be concerned. In other words, why one might think that AI would be less good at promoting freedom and better at aiding authoritarianism. The first category let's call deepfakes and misinformation.

9:33This was clearly what Soros was talking about, that generative AI tools can be used to create highly convincing but fake, unreal information that can then be used to influence people's perspective. On my poll, Gigamesh wrote, authoritarians thrive in environments of mass distrust, and AI is much better at confidently lying than sorting out the facts. In February, Bill Drexel and Caleb Withers also wrote an opinion piece for The Hill which made a similar argument. The piece was called Generative AI Could Be an Authoritarian Breakthrough in Brainwashing. And in this piece, they argue that the real concern in some ways isn't just Russian bots trying to influence US elections, but is what happens for the populations within authoritarian countries.

10:13The authors write, however daunting Chinese and Russian foreign disinformation efforts look in a post-GPT world, open societies receive only a small fraction of the propaganda that Beijing and Moscow blasted their own populations. As examples, they write, in 2019, China's Xi Jinping ordered his party state to leverage AI to, quote, comprehensively increase the ability of the Chinese Communist Party to mold Chinese public opinion. Russia's Putin has similarly doubled down on AI-enabled propaganda in the wake of his Ukraine invasion, including a fake video of Ukrainian President Vladimir Zelensky calling for Ukrainians to surrender.

10:44So again, reason one to be concerned, deepfakes, and misinformation, both in the context of adversarial opponents trying to undermine democracies as well as trying to consolidate control in their own countries. A second reason for concern around AI and authoritarianism has to do with surveillance. Cosmo Crickster wrote, I voted enable authoritarians because combined with data scraping, I think governments will employ huge teams that deploy AI for unbelievable levels of surveillance, facial recognition, location patterning, buying habit patterning, etc. And privacy will become as scarce as time.

11:16Now, Cosmo is not alone in this concern. Authors like Daniel Oberhaus have written about how authoritarian regimes have an advantage in AI because they're willing to capture data such as advanced facial recognition data and use it in ways that open society simply wouldn't. But even in more benign contexts, surveillance is a real concern. One of the things that the White House has been paying attention to this year is worker surveillance. In the context of increasing remote work regimes, the White House is worried about the automated systems that employers are using to monitor their workers who are working from home.

11:46A third reason for concern about how AI might promote authoritarianism is that authoritarian societies or closed societies, specifically China, might use their global sphere of influence to get their particular slant on AI, which is designed to align with their particular motivations and goals, to other parts of the developing world and emerging markets. A Carnegie Endowment for International Peace essay reminds us that, quote, for most of the rest of the world, AI technology is better understood as an issue of economic development, determining a country's relative standing in the global technology race, rather than as a geopolitical or ideological preference.

12:20In other words, China exporting its CCP-aligned AI will be less about ideology and more about it being potentially lower cost than other options for a technology that the rest of the world understands that they need to keep up when it comes to their economies. A fourth reason to be concerned about authoritarianism as it relates to AI is the fear that it will convince people that command and control centrally planned economies weren't wrong theoretically, but simply didn't have the data to do a good job of central planning. In 2018, Peter Thiel and Reid Hoffman had a very famous conversation slash debate hosted by Neil Ferguson at Stanford, where Thiel famously said, crypto is libertarian, AI is communist.

12:59Now, what he was talking about was the tendency towards centralization versus decentralization, and something that many have noticed about AI, which is that it seems to prioritize centralization, the ability to aggregate large amounts of data. But as you can hear in this clip, he makes a point about how it might be used to reinstate principles of central planning, which had otherwise fallen by the wayside. Two of the areas of tech that people are very excited about in Silicon Valley today are crypto on the one hand and AI on the other. And even though I think these things are underdetermined, I do think these two map, you know, in a way, politically very tightly on this centralization, decentralization thing.

13:40Crypto's decentralizing, AI is centralizing. Or if you want to frame it, you know, more ideologically, you could say that crypto is libertarian and AI is communist. And of course, we always hear only the first half because we're biased to the left. But AI is communist in the sense that it's about big data. It's about big governments controlling all the data, knowing more about you than you know about yourself. So a bureaucrat in Moscow could, in fact, set the prices of potatoes in Leningrad and hold the whole system together. And if you look at the Chinese Communist Party, it loves AI and hates crypto.

14:18So it actually fits pretty closely on that level. A fifth reason to be concerned about AI and its impact on authoritarianism versus democracies is the concern that capitalist competition will create AGI we can't control. This has obviously been the drum that Dr. Jeffrey Hinton has been beating ever since he left Google. Now, one thing that's worth noting, given how much Hinton comes up in this discourse right now, is his own comments about his political persuasion. In an interview with The Guardian in May, he wrote, I'm a socialist. I think the private ownership of the media and the means of computation is not good.

14:49If you view what Google is doing in the context of a capitalist system, it's behaving as responsibly as you could expect it to do. But that doesn't mean it's trying to maximize utility for all people. It's legally obliged to maximize utility for its shareholders, and that's a very different thing. And this is, of course, what Hinton has been talking about constantly, that the existential threat to business models that AI represents is forcing companies who otherwise might be good stewards of this technology to be reckless in their pursuit of catching up or getting ahead of their competitors. Now, I think if you play this out in terms of how it leads to more authoritarianism, a lot of that has to do with what feels like an inevitable reaction from the state.

15:24In the face of AI-led disruption, state power could increase in two very clear ways. The first is that people become more reliant on the state because the entire economy has to be reimagined in the context of AI doing a lot of the work that people do now. The most mainstream version of this conversation we've seen yet was during the last presidential campaign when Andrew Yang proposed giving every adult$1 ,000 a month as a way to offset job losses that came from technologies like AI. Even if well-intentioned, the universal basic income regime would increase the power of the state relative to people's economic lives.

15:55However, a second way in which capitalist competition that led to AGI could increase authoritarian state power ultimately is exhibit in how some of the most concerned AI safety and AI risk people talk about leveraging the power of the state to stop advanced AGI. Coindesk columnist David Z. Morris wrote, it took the AI safety people exactly six months to speedrun their cult trajectory all the way to saying the quiet part loud. Under that, he quote tweeted Aleph, at woke eight-year-old, who wrote, the idea that you should be able to do whatever you want in the privacy of your data center when there is a 20 % of ending the world is simply ludicrous.

16:28We will use the state to stop you and be fully justified in doing so. Now, I don't want to reduce the entire AI safety and risk conversation to this caricature of it that seems frothing at the mouth for state power to intervene. But I do think that it is a meaningful part of the discourse that we have to have, because if you follow many of the more extreme parts of that AI safety and risk community, there is absolutely a sense among some, not all, but some, that the trade-off of an authoritarian state is worth it if the other side is human extinction. Now that deserves an entire show on its own, so we'll leave it there, but I do think it is worth flagging as a part of this discourse about how AI could aid authoritarianism.

17:08Now, as I said at the outset of this piece, there is a much broader sense, it seems, at least currently in the narrative, that AI is likely to aid authoritarianism versus increase freedom. However, that's not universal, and I think it's worth looking at some examples of how AI could be used to promote freedom. A first way is that AI could be deployed to actualize values we've already proclaimed. An example of this came from just a couple of days ago when Fox ran a piece about Altana, which is an AI supply chain tool that helps identify goods that are made with inputs from Uyghur forced labor in China.

17:41If you haven't dug into the Uyghur situation in China, the TLDR is that this is a Muslim ethnic minority group who primarily reside in a specific northwestern region of China and who've been forced into what have been called re-education camps, forced to work in factories against their will, etc., etc. Now, last year, something called the Uyghur Forced Labor Prevention Act took effect, which is an attempt to stop companies from sourcing components or inputs that are products of forced labor. Now, this startup Altana said that an analysis that it did found that 10 % of all companies in the world were buying and selling physical goods that were using inputs from that region in China.

18:16The problem was that it wasn't as obvious as getting in touch with a factory in Xinjiang, but that the goods were, as the CEO of Altana put it, typically many hops removed. Altana has now helped dozens of companies flag these goods and inputs in order to allow them to make different decisions with their supply chains. A second way in which AI could promote freedom is as a new interface for the dissemination of information. All the way back in 2018, Global Voices was writing about a chatbot that was helping Russians who had been detained at protests request legal assistance. Now, of course, for these types of uses to have their full impact, there are a number of hurdles to jump.

18:50First, the information has to actually be good, i.e. not compromised through hallucination or through political pressure, which has dictated what information can and can't be shared. and two, that information or the interface for that information has to be accessible through problematic firewalls. But still, it's hard not to look at something like ChatGPT and not see the potential for the democratization of information. And indeed, if you go back to my poll, for people who thought that AI had a good chance to promote or increase freedom, it seemed like the biggest determinant that they were discussing is, as Chicago ZeroX put it, is there a good open-source, uncensorable version for the people?

19:25Fern Vanderkleed said something similar. It depends on whether we each get our own personal AI that has our interests as a primary function, or whether we all use a personal access to a communal AI that scrapes our data. AI entrepreneur Adam Levine wrote, If it's treated like the internet, then it's wildly supportive of freedom. If it's treated like weapons, technologies, then only militaries, the state, criminals, and big corporations with proximity to power get access to it, and then it's wildly dystopian. A final comment from my poll that I wanted to make reference of is someone who unabashedly was in the Promote Freedom camp.

19:54Natoshi B. Freedom writes, Authoritarianism requires the ability to control information flows and fear-mongering. AI, like the internet, unless withheld from the general public, makes it harder to control narratives. Open-source AI is even more liberating. The cat's out of the bag. Now, I wonder a little bit if we're thinking about these risks a little incorrectly. Specifically, I wonder to what extent people are underestimating the capacity of democracies and some of the things that we don't even like about democracies to be a counterweight to what authoritarian control of AI might want to do.

20:29Authoritarians, as Natoshi points out, want to control narratives. They want to dictate reality. In that same interview with The Guardian, Hinton made an argument that America's inability to agree on issues makes it susceptible to these problems. He wrote, This stuff helps authoritarian governments in destroying truth or manipulating electorates, And having to deal with these threats in a situation where Americans can't even agree to not give assault rifles to teenage boys, that's not a hard thing to think about. In Uvalde, there were 200 policemen who didn't dare go through a door because the guy on the other side had an assault rifle and was shooting children.

21:02And yet, they can't decide not to ban assault weapons. So a totally dysfunctional political system like that is just not the right system to have to deal with these threats. Now, I'm not at all interested in talking about gun issues right now or the Second Amendment. But to me, it seems like almost exactly the opposite of what Hinton is saying. might be the case as it relates to AI. In other words, the fact that there is so much disagreement about something that seems obvious to many suggests to me that AI is going to have an extremely hard time consolidating people around a single point of view.

21:32So yes, to the extent that the goal of authoritarians using AI is simply to sow discord in the American electorate and have us disagreeing with each other, of course, that feels like a real concern. However, to the extent that one thinks that the real authoritarian threat of AI is not lots of people disagreeing with each other, but lots of people believing wholeheartedly what the state says, it seems like that's going to be a huge, huge challenge in a system like the one we have. I think we may also be underestimating AI's value as a catalytic force in determining how we want to organize society.

22:03A recent Wired piece discussing the Reddit API updates and the blackouts that followed made the point that this is part and parcel of a labor battle between algorithms and the humans who feed them. And if you look across the early political response to AI, things like the Hollywood writer's strike, it's hard for me not to see those things as part of a re-emerging conversation around what we want out of technology and whether it serves us or not. Over the last few decades, we have gotten used to the idea that technology is just an inexorable march forward. But I think we may be starting to remember that ultimately, we do have some amount of agency and control to determine and dictate whether a technology actually suits our needs or not.