Can Worldcoin Help Solve AI's Big Economic Problems?

0:00Today on the AI Breakdown, we're discussing whether WorldCoin can solve some of AI's big economic problems. Before that on the brief, Stability has released two new open source LLMs. 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 about our newsletter, our YouTube, and our Discord. Welcome back to the AI Breakdown Brief. All the AI headline news you need in five-ish minutes or less. Today we kick off with another entrant into the open source AI LLM battle. And this time it comes from Stability AI, the company, of course, behind Stable Diffusion.

0:39Confirming that these guys grew up at the same time as I did, the model is called Free Willy. The model technically comes from Stability AI and its Carper AI lab, and it's actually two separate models, Free Willy 1 and Free Willy 2. Free Willy 1 is based on the original Lama 65 billion foundation model, but was fine-tuned with a synthetically generated data set, while Free Willy 2 leverages the Llama 70 billion foundation model, and they say that the performance after fine-tuning of Free Willy 2 actually compares favorably with GPT-3.5 in some tasks. In a lot of ways, these models represent a test of new training approaches.

1:15Stability AI writes in their announcement post, the training for the Free Willy models was directly inspired by the methodology pioneered by Microsoft in its paper, ORCA, Progressive Learning from Complex Explanation Traces of GPT-4. Now, they show a number of different comparisons of performance evaluations, and in a number of areas, including the LSAT's logical reasoning and the LSAT's analytical reasoning, as well as the SAT English, Free Willy 2 actually compares favorably to ChatGPT, which is GPT 3.5. Stability AI writes, quote, Both models are research experiments and are released to foster open research under a non-commercial license.

1:50While we have conducted internal red teaming to ensure the model remains polite and harmless, We welcome the community's feedback and help in further red teaming. In a separate tweet, Ahmad, the CEO of Stability AI, said that when they were ready for commercial release, they would likely do something pretty similar to the terms surrounding Llama 2. Now, speaking of Llama 2, Meta has received a little bit of criticism over the last couple days for calling Llama 2 open source. On the Voices of Open Source blog, a post reads, Meta's Llama 2 license is not open source. The piece starts, OSI is pleased to see that Meta is lowering barriers for access to powerful AI systems.

2:24Unfortunately, the tech giant has created the misunderstanding that Llama 2 is open source. It is not. Even assuming the term can be validly applied to a large language model comprising several resources of different kinds, Meta is confusing open source with resources available to some users under some conditions. Two very different things. The piece goes on. Open source means software under a license with specific characteristics defined by the open source definition. Among other requirements for a license to be open source, it may not discriminate against persons or groups or fields of endeavor, Meta's license for the LAMA models and code does not meet this standard.

2:58Specifically, it puts restrictions on commercial use for some users and also restricts the use of the model and software for certain purposes. Later in the piece, they write, OSI does not question Meta's desire to limit the use of LAMA for competitive purposes, but doing so takes the license out of the category of open source. The OSD does not allow restrictions on field of use because you can't know beforehand what can happen in the future, good or bad. That's what allows the Linux kernel to become popular in medical devices as well as airplanes and rockets. But the meta policy prohibits use in several areas that might be highly beneficial to society, such as regulated and controlled substances and used for critical infrastructure.

3:33Even something that sounds as simple as you must follow the law is problematic in practice. What if the law in different places is inconsistent? What if the law is unjust? So basically the piece isn't so much a critique of the way that meta is releasing this, in fact it validates them. It just wants to be precise about language because they argue language matters. Decibel partner and latent space host Alessio Finelli also wrote a piece about this called Llama 2 is an open source and why it doesn't matter. He writes, when Llama 2 came out, many of the folks I respect in the community were upset about misusing the term open source when referring to the model.

4:06While it's mostly open, there are caveats such as you can't use the model commercially if you had more than 700 million monthly active users as of the release state, and you cannot use the model output to train another large language model. Alessio continues, these types of restrictions don't play well with the open source ethos. But while I agree that Lama 2 cannot be called open source in the traditional meaning of the world, I also think that it doesn't matter. The term open source needs to evolve once again in the world of AI models. Alessio then goes on to point out how many different aspects of an LLM could be released with different levels of openness, and ultimately comes to the inclusion that, quote, The important thing is that more and more of this work is done as openly as possible.

4:43Now, of course, if you listened to the AI breakdown over the weekend, you heard me do a long reads all about this idea of open source AI and whether it actually creates new challenges in terms of safety. In a slightly different AI safety topic, OpenAI's head of trust and safety, Dave Willner, has stepped down. In a note on LinkedIn, Willner wrote, Some bittersweet news to share. I'm leaving OpenAI as an employee and transitioning into an advisory role. Now, Dave pointed to family time and specifically wanting to spend more time with their young children is the reason. And as someone with young kids, I totally get it.

5:13But that hasn't stopped there from being lots of speculation around whether there is something more going on. Now, it's actually kind of interesting to me how much hay is being made of this with a very extensive TechCrunch piece being written about it that goes deep into Willner's personal history with this, not only at OpenAI, but also at previous employers, including Facebook. And I think that this probably reflects just how hot button this question of AI safety is right now. Of course, last week at the end of the week, we got that announcement from the White House that seven companies, including OpenAI, along with Meta, Microsoft, Google, and others, had all committed to a set of voluntary principles around AI governance.

5:47At the time I last recorded about it, a few companies had started to release statements, but we've gotten a few more since then. OpenAI published a piece called Moving AI Governance Forward. OpenAI and other leading labs reinforce AI safety, security, and trustworthiness through voluntary commitments. OpenAI's note is mostly a further articulation of what it intends to do specifically around the commitment areas that had been previously articulated by the White House. One interesting note is that OpenAI makes sure to define the scope. It says, where commitments mention particular models, they only apply to generative models that are overall more powerful than the current industry frontier, e.g.

6:22models that are overall more powerful than any currently released models, including GPT-4, Claude-2, Palm-2, Titan, and in the case of image generation, Dali-2. So effectively, they're saying that this commitment is about the stuff coming down the pipeline, not the stuff that already exists today. Inflection, the company behind Pi, also wrote a note about their commitment. The piece was called the precautionary principle, partnering with the White House on AI safety. Again, mostly it's just an announcement, but one paragraph that was interesting is where they said, Speaking plainly, there's a huge amount of safety work ahead.

6:52So far, AI safety has been stuck in the space of ideas and meetings, stuck in bureaucratic wrangling, stuck in academic journals, stuck in breathless op-eds and Twitter threads. The amount of tangible progress versus hype and panic has been insufficient. At inflection, we find this both concerning and frustrating. That's why safety is at the heart of our mission, why we founded the company in the first place, and why we were working flat out to make it happen. Later in the post, they say, we will soon publish further work on safety and announce a number of significant collaborations. They then call out the example of elections as an area of specific focus.

7:21At inflection, they write, we believe it is vital that AI be kept out of the democratic process. Powerful but still nascent technology cannot be a part of electioneering. We should legislate to ban the use of AIs in chatbots around the ballot box. Next year is an election year in America and it will be the first in the era of generative AI. Preserving the stability and integrity of our electoral system means starting work right now. Now, one more very different announcement around ChatGPT. It is now making its mobile app, which up till now has been only available on iOS, available for Android users as well.

7:51Lastly, a little bit of AI intrigue in the transition away from the Twitter brand. Over the weekend, Elon Musk announced, quote, and soon we shall bid adieu to the Twitter brand and gradually all the birds. The brand, of course, is becoming X to better align with Elon's big vision of an everything app. And yesterday afternoon, new CEO of X or Twitter or whatever it is now, Linda Iaccarino, wrote a thread where she talked about the new business. Linda writes, It's an exceptionally rare thing in life or in business that you get a second chance to make another big impression. Twitter made one massive impression and changed the way we communicate.

8:27Now X will go further, transforming the global town square. X is the future state of unlimited interactivity, centered in audio, video, messaging, payments, and banking, creating a global marketplace for ideas, goods, services, and opportunities. Powered by AI, X will connect us all in ways we're just beginning to imagine. For years, fans and critics alike have pushed Twitter to dream bigger, to innovate faster, and to fulfill our great potential. X will do that and more. We've already started to see X take shape over the past eight months through our rapid feature launches, but we're just getting started.

8:55There's absolutely no limit to this transformation. X will be the platform that can deliver, well, everything. Elon and I are looking forward to working with our teams and every single one of our partners to bring X to the world. Now, of course, the part that the AI crowd picked up on was this powered by AI, X will connect us all in ways we're just beginning to imagine. So I will close today's AI Breakdown brief with a question. What do you think X will do? And specifically, what does it mean that it will be powered by AI? Leave a comment to tell me what you think, or come join us on the AI Breakdown Discord.

9:26You can find a link at bit.ly slash AI Breakdown. Thanks for listening or watching as always, and I'll be back soon with the main AI breakdown. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage. If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? that's where supermanage comes in supermanage ai magically distills your team's public slack channels into a real-time brief on any employee anytime catch up on contributions work in progress challenges they're facing sentiment everything you need to show up ready for a truly meaningful conversation and it's completely free visit supermanage.ai forward slash breakdown today to start making the most of your one-on-ones and thanks again to supermanage for sponsoring the ai Breakdown.

10:26Welcome back to the AI Breakdown. Today we are talking about WorldCoin. It is an ambitious and controversial project to establish proof of personhood and to pave the way for a global UBI, and its context is, of course, the new era of artificial intelligence that we're living through today. The question, however, is whether WorldCoin actually solves a set of problems created by AI, or whether, as some are arguing, it's part and parcel of a growing dystopian hellscape. WorldCoin was founded three years ago and has been subsequently in a beta period. However, today the financial network that is part of it has gone live, and so much of both the crypto world and the AI world are talking about it.

11:03Their announcement reads, if successful, we believe WorldCoin could drastically increase economic opportunity, scale a reliable solution for distinguishing humans from AI online while preserving privacy, enable global democratic processes, and eventually show a potential path to AI-funded UBI. In a thread on Twitter, Sam Altman, who is a co-founder of WorldCoin in addition to his work on OpenAI, writes, The goal is simple. A global financial and identity network based on proof of personhood. This feels especially important in the AI era. I'm hoping WorldCoin can contribute to conversations about how we share access, benefits, and governance of future AI systems.

11:38So WorldCoin is actually a couple parts. At core, it involves establishing a unique identifier based on scanning one's irises. This creates a world ID, which is unique and can't be spoofed again. In other words, each person on Earth can have one and only one world ID, and things that are not human who don't have eyes to be scanned cannot create world IDs. Now, part of the incentive for scanning one's irises is that people who do so get a distribution of the WLD token, which just went live. The current value of the amount of token that was distributed to people who had their eyeball scanned is around$60 US, assuming it was liquid enough to actually exchange that for dollars should people so want.

12:15So there are kind of two different problems generated by AI that this is playing around with. One is how to prove people are people and not bots. The second is how to support people in a world where many of the jobs, if not most of the jobs that are done today, can be done more efficiently by artificial intelligence. There has certainly been no shortage of discussion around universal basic income schemes stemming from job loss that people assume will come from AI. Of course, this was the entire premise of Andrew Yang's insurgent presidential campaign in 2020. The fundamental pillar of that campaign was something that he called a freedom dividend, a monthly payment to every American adult that was meant to supplement their income in the face of assumed job loss from automation.

12:55Now, Andrew Yang didn't win, but his ideas got a bit of traction. And of course, this was all pre-ChatGPT and the rise of generative AI, where the conversation has done nothing but increase. An essay from May in the The Atlantic says before AI takes over, make plans to give everyone money. The U.S. needs policies now to support workers made redundant by artificial intelligence. Artificial intelligence is coming for all our jobs, the piece writes. That is a prevalent fear these days and not one easily dismissed. The specific effects of revolutionary technologies are impossible to predict, and perhaps AI will turn out to be overhyped.

13:26But it really is different from past advances. The work it can do really is different. The job it threatens really are different. Its effects on the labor market might really be different. A pair of economists at Goldman Sachs recently estimated that two-thirds of American occupations are now exposed to AI-driven automation. In the coming decade, the technology will wipe out 300 million jobs they forecast. That's one in every 11 jobs on the planet. The piece continues. Here in the United States, an AI jobs revolution need not be anything for average wage-earning families to fear. If AI does not enslave the human race or destroy humanity outright, this generative technology can and should be a very, very good thing.

14:02It should lift productivity growth, increase our national wealth, and contribute to our commons of knowledge. Every person in this country can and should benefit from such an awe-inspiring invention. The problem is that we do not have any policies in place to support workers in the event that AI causes mass job loss. And of course, the rest of the piece goes on to articulate a vision of UBI and how it might work. Now, so far, the UBI aspect of WorldCoin is a little bit harder to understand exactly how it's going to work. Indeed, even in the announcement, they say eventually show a potential path to AI-funded UBI versus really articulating how that might work.

14:37Coindesk columnist David Z. Morris writes, In just one of many missing premises around WorldCoin, it is unclear how the WorldCoin token can be expected to have any value for recipients once it circulates. It's extremely difficult to imagine how what amounts to an Ethereum-based meme coin with no apparent tokenomic model is going to be exchangeable for essentials like food and shelter over the long term. Now, this was written before the launch of WorldCoin Today, which came with a tokenomics announcement. But the tokenomics that were revealed aren't necessarily impressing people either. The DeFi investor writes, WorldCoin is another VC-backed project with predatory tokenomics.

15:11WLD already has an insane fully diluted valuation,$27 billion, but less than 1.5 % of the total supply is currently in circulation. The project might have potential, but I strongly advise staying away from its token. Now, for those of you who aren't that familiar with crypto, fully diluted valuation refers to the idea that you can just multiply what the price is of a token by the total number of tokens that will be created all time, even if only a tiny fraction of those tokens are actually circulating. The problems with this are numerous. One, it basically just serves to make things look more valuable than they actually are.

15:45Two, even in the context of the price of the tokens that are circulating, it's very rare that with a new unestablished project, you're actually going to get that price if you tried to, for example, cash it out or trade it, because liquidity is usually so light that any attempt to sell en masse is going to actually impact the price and drive it down. And importantly, Fully Deluded Valuation, or FDV, has a history of being used in the last cycle as a way to juice the value of projects in a way that rewarded the early investors, but ultimately left retail holding the bag. Given how little has been said so far about the UBI aspect of this project, it might be worth shifting over to just focus on the other dimension of this.

16:22And that's the idea of proof of personhood. Going back to David Morris, he writes, the UBI element of the project is simply window dressing for its real goal, solving the problem of digital identity. And this really does feel like what WorldCoin's core focus is. For people who are supporting WorldCoin, it seems to be largely about the importance of attempts at the proof of personhood problem. Bankless's David Hoffman writes, valid attempts at a decentralized proof of personhood protocols are invaluable. Worldcoin investor CoinFund wrote, Worldcoin started with the question, how do we prove human uniqueness in a digital context and do so while retaining anonymity and privacy?

16:58There are a number of reasons for this question, but one that truly spurred the need was how to disambiguate AI from humans as AI improves. Ethereum creator Vitalik Buterin wrote extensively about this in a blog post today called What Do I Think About Biometric Proof of Personhood? In that post, he wrote, one of the trickier but potentially one of the most valuable gadgets that people have been trying to build is a decentralized proof-of-personhood solution. Proof-of-personhood, aka the unique human problem, is a limited form of real-world identity that asserts that a given registered account is controlled by a real person and a different real person from every other registered account, ideally without revealing which real person it is.

17:33Vitalik goes on to describe the philosophy behind WorldCoin as, AI is going to create a lot of abundance and wealth for humanity, but it also may kill very many people's jobs and make it almost impossible to tell who even is a human and not a bot. And so we need to plug that hole by one, creating a really good proof of personhood system so that humans can prove they are actually humans, and two, giving everyone a UBI. Now when it comes to WorldCoin's specific approach to this problem, Vitalik identifies four big risks. One is privacy. Obviously scanning one's irises comes with privacy concerns, especially when it's an individual who's actually doing it and checking the database to determine whether a person already has a world ID.

18:07Another problem that Vitalik points out is accessibility. Are there enough orbs for people to actually get to them easily? A third is a question of security and hacking. And a fourth and a big one is centralization. Vitalik writes, the orb is a hardware device and we have no way to verify that it was constructed correctly and does not have back doors. Hence, even if the software layer is perfect and fully decentralized, the WorldCoin Foundation still has the ability to insert a backdoor into the system, letting it create arbitrarily many fake human identities. Now in this same post, Vitalik does point out that it's important to distinguish between issues with WorldCoin's approach to this problem, issues with any generally biometric approach to the problem, and issues that any approach to the problem will have in general as well.

18:47Unlike many on Twitter, Vitalik doesn't come away cynical, but he also certainly doesn't make a ringing endorsement of WorldCoin's approach to the problem. If you're interested in a more in-depth look of some of the critiques around both the privacy dimension of this as well as the tokenomic dimension, go check out the main breakdown today. which you can get a link to at breakdown.network. I will say when it comes to the artificial intelligence side of this conversation, it feels like if nothing else, this is a good moment to start asking the questions in a more vociferous way around how to deal with economic fallout of AI.

19:20There are a huge number of assumptions embedded into WorldCoin that are worth pushing on if not outright challenging. Is UBI, for example, really the right approach to AI job displacement? Should instead those efforts be focused on re-skilling, re-training, adapting to new environments? And when it comes to proof of personhood, how urgent is that problem really? Perhaps a better question, in what context is it an urgent problem? And what are the trade-offs of other solutions as well? How do those trade-offs compare to the trade-offs inherent in WorldCoin? At the end of the day, I do not fault people for trying to push forward on areas of global import that don't have enough conversation yet, and that people see as big problems in need of solutions.

20:01I try to assume good faith until otherwise proven wrong. But if the AI community wants to see WorldCoin as anything, I suggest we see it as an invitation to a conversation. It's a conversation that needs more voices than just the crypto community, as much as I love them, or people who work on OpenAI as well. So to the extent that these problems matter to you, let's dive into it. As always, you can find a link to our Discord in the show notes, and if nothing else, come join the conversation there. For now, I want to say thanks again for listening, and until next time, peace.