Lane Rettig on saving DAOs with AI
DAO governance has been "disastrous," could AI agents bring a much needed evolution?
On day three of ETHCC I was starting to falter. The swathes of side events and days of panels, although fascinating, were starting to take their toll. So it was with joyful relief that on that Wednesday, extortionate but delicious iced matcha in hand, I was greeted by Lane Rettig’s warm smile and boundless energy.
Lane is an Ethereum Foundation veteran. He started out as one of the few public-facing voices on Ethereum’s core development team, known not just for his technical work on Ewasm but for openly engaging in the messy, human side of blockchain governance. He spoke publicly, launched forums to discuss the social impact of crypto, and wasn’t afraid to call out what he saw as Ethereum’s governance failures. His public critique of Ethereum’s internal governance, calling it a “failed technocracy” that made him a controversial figure, admired by some for his transparency but questioned by others.
After leaving Ethereum in 2019, Rettig joined Spacemesh, a decentralised layer-1 project aiming to make home mining accessible again. He stayed on as CTO through its mainnet launch, but stepped away after the project ran into funding and centralisation issues. These days, he’s at NEAR Foundation, leading research in the creation of House of State, a new governance system built to move NEAR from “a benevolent dictatorship for now” to a fairer, leaderless and decentralised system.
Having had some interesting ideas come up when discussing AI and web3, I was keen to get his opinion on if and how AI made any sense within the context of governance. Others I had spoken to looked on the concept with concern, one even questioning why we would want an AI involved in something as human as social organisation. But Lane made some compelling points (which you can see below) and now I can safely say that while I can see where AI could go wrong if not built carefully (and I have a lot of concerns about AIs with hidden agendas or out of control bias voting for humans without oversight), it could change how we do things like voting, possibly shaping how organizational systems are designed because of it.
Key takeaways:
DAOs have faced a slew of problems ranging from voter apathy to general mistrust, however web3 continues to have the potential to foster more genuine systems of decentralised participatory governance. The missing link to unlock this potential may be AI.
Historically, humans have proven they are bad at democracy and we are battling similar problems within democratic systems that have existed for years. When this has been brought to new governance systems such as DAOs it has failed. However, older systems like the citizen assembly and systems of rough consensus could still provide lessons which blockchain systems could build upon.
The ability of AI to identify patterns and broad access to information, whatever the language of a text, could enhance governance systems. By efficiently identifying human preferences, even those which a human is unaware of, AI agents may be well placed to become delegates for voting systems.
However, there are major things to consider when bringing agents into governance systems. For example, how much agency should we give an agent? If it makes an alignment mistake with human preferences, it could be for a critical policy proposal which should shape the future direction of a project. AIs can also carry algorithmic bias which may influence its ability to act fairly.
Blockchains are coming to resemble a new kind of nation state. With this shift comes a need to consider elements that all nation states have to focus on, such as GDP, productive capacity and attracting the best talent for innovation.
Food for Thought:
In part of our conversation, Lane mentions AI’s ability to ascertain humans’ revealed preferences regardless of their explicitly stated preferences. It’s a concept that has been explored quite a bit the last few years but something that I agreed with Lane could be a positive element of using AI in governance systems … and at the very least just interesting in terms of practicing introspection. To that end, I went down a rabbit hole and unearthed this paper that looks at this ability in terms of achieving AI alignment and this paper, which studies the divergence from LLMs’ stated preferences according to the context outlined in a prompt.
When thinking about AI in governance there are concerns around bias and algorithmic manipulation, but Lane also mentioned “sleeper agents” as being a possible issue – AI agents with differing agendas to those outlined that avoid detection through safety checks. He mentions a study that Anthropic conducted on AI sleeper agents to understand how they work, what can trigger a change in behaviour and how the sleeper agents respond to attempts to remove deceptive coding.
Another idea that comes up is that of “rough consensus” where Lane points to a specific document – On Consensus and Humming – published by the IETF. So, of course, I went and found it and now you have it here. It outlines a really intuitive approach to reaching consensus among a group of people which does actually include humming. I’d be really interested to see how it could be transferred to a digital system.
Lane mentions an approach to decentralised “user-owned” AI that NEAR AI (an offshoot of the NEAR Foundation focused on AI) is working on called Decentralized Confidential Model Machine Learning. I’ve included a link here to a piece co-founder, Illia Polosuhkin wrote about it, outlining why decentralised AI is becoming increasingly important and how it could preserve privacy.
Isabelle Castro
Ok Lane, ease us in with a little brief history of you and how you got to where you are today.
Lane Rettig
We have the full hour. Right. In terms of pacing, there's a short medium and a long answer, but I'll start with a short answer.
So I've been a software developer my whole life, my whole career. That's been like the common thread for me. I say I came of age with the Internet. I'm 42, so TCPIP protocol was finalized the year I was born. Right. And so that's been just like. It's always been my passion. I got my first computer when I was like 3 or 4 years old, and I've sort of applied those skills in a whole bunch of different domains over the years. I studied computer science in university. I worked in traditional finance after that for about six years in New York and Hong Kong, which was good, you know, discipline, good rigor.
Got a chance to understand how the financial system works, at least the one we now call TradFi. Changed gears, went back to school, did an MBA, became an entrepreneur, had a healthcare technology startup for a few years. I kind of rode the startup roller coaster and a very compressed startup life cycle experience compressed into about three years. So from raising money, building a team, going through an accelerator, building the first product, selling it, and then ultimately selling the company as well. This healthcare technology startup, and I mention it because it's how I got into this industry basically because doing healthcare in the United States at that time, this was 2013-2016, was really painful. It was very difficult to innovate because of regulation, HIPAA compliance, which is the specific body of regulation that healthcare technology needs to comply with. And then just the fact that it's a very risk averse industry, as you'd imagine. They say hospitals, airlines, education, governments, those are the hardest, slowest moving industries. And after that experience, just pushing so hard against an immovable object, trying to bring innovation into healthcare industry. I took some time off and then I was looking for something where I could innovate more freely. And so that's indirectly how I found this industry. I discovered Ethereum in late 2016, had a very rapid descent down the rabbit hole. Found myself at the third Ethereum Devcon event in Cancun, Mexico in 2017 and showed up as a volunteer. Shortly thereafter they hired me and I ended up joining the Ethereum foundation as a core developer.
I was working on Ethereum for two and a half years. And ostensibly my role was technology core developer. But what I found was that the technical side of the project was very well served at that time, but the social human side was not, which is not surprising given the personality profiles of the people building it then and bars are still true today, I think. And so I found myself being one of the only people who felt comfortable speaking to media about the project. And also I just saw that the social governance side was a total disaster at that time. I didn't have a ton of experience with governance, but I'd studied a little bit of corporate governance in business school. I had a little bit of experience with it for my startup. So I just kind of rolled up my sleeves and ended up very hands on for a couple of years there in Ethereum governance, which made me a few friends and a lot of enemies. Seems to be a rule of thumb in governance.
Anyway, fast forwarding a little bit, parted ways with Ethereum in 2019, left to build my own layer one blockchain, which I had a vision to build, spent five to six years working on that. I have a lot to say about that. But that project did not work out and then just eight months ago, I joined NEAR Foundation. I've known NEAR for a very long time. I've been close to the founders. I always wanted to join them. Timing never lined up previously, but we began speaking again kind of the middle of last year and they were looking for someone to lead this new governance initiative as well as just owning research.
So the timing finally aligned and then, yeah, I came on board with that project late last year and officially I'm head of research at the NEAR Foundation. But what excited me about it was not the governance piece. So. So my primary project, my mandate right now is this House of State project which is a from-scratch governance. I mean, it's a DAO.
We're building a DAO to govern the NEAR protocol in a really decentralized kind of community-first fashion. I had some very negative experiences with DAOs as well back in like 2018- 2019. I think DAOs tend to be disasters for reasons we can come back to. And so the governance piece of it wasn't as exciting to me, but the AI piece was what excited me. So I've been observing AI from the sidelines for the past few years while working on the project I mentioned a moment ago, space Mesh, and there was no role for AI in that project. We didn't get there. And I think the thing that's always fascinated me the most and excited me the most about blockchain cryptocurrency. Web3 is not the financial stuff and it's the tech, but not the tech by itself. It's actually the human piece. I've always found that people who are new to the space gravitate towards the financial side first because it's the easiest thing to understand. It's the first touch point. You buy some Bitcoin, that kind of thing, It's a great place to start. And then they often say, came for the gains, stayed for the tech. You know, if you're a software developer like me, if you're a computer scientist, there's a lot of exciting technology to play with. But like the ground floor of the rabbit hole is the social layer.
Ultimately it's all about people, it's about human systems. It's about human coordination. And that is the key innovation of blockchain. That we now have these new permissionless tools, starting with Bitcoin, to like organize human endeavor. That's the truly novel primitive here, in my opinion, and the most exciting, most powerful one.
So in conclusion, you know, that superpower combined with the superpower of AI, it blows my mind to consider what's possible. We can't take any of this for granted yet. We're just getting started. And I often also describe my mission as avoiding dystopian outcomes, avoiding black mirror outcomes, because I still think that's like the more likely path, kind of the path we're on right now.
But in short, if we can kind of harness these two powerful ideas, combine them together and solve some of the issues with governance that humans have been struggling with for literally thousands of years, the opportunity to do that full time, to do that with the support of an organization like NEAR foundation, to work closely with Illia, who is a legend in the AI field. And just like I respect him enormously, it was too good to pass up on.
Isabelle Castro
Absolutely. It's definitely kind of the forefront of where we're going. I understand that you kind of haven't started the research piece.
Lane Rettig
We're just in the earliest phases now. Yeah.
Isabelle Castro
What is the thing that most excites you about the Web3 and AI intersection when it comes to governance?
Lane Rettig
Yeah. I mean, I was reflecting on this before we met. There's kind of like the thing that excites me about Web3, the thing that excites me at governance, the thing that excites me about AI. And then there's kind of like various permutations there, and then there's kind of the intersection of all three of them.
How do I say this succinctly? I think let's start with Web3 and Governance. So we have a new set of tools. If you look at the way that humans have organized ourselves, whether it's into cities, nation states, on the political side of the spectrum, whether it's,corporations, joint stock corporations. On the kind of more commercial side of the spectrum. I mean, people don't even realize these ideas are both actually pretty new ideas.
The joint stock corporation is, you know, 500 years old, plus or minus with the Dutch to thank for that, I think. And the nation state is also about 500 years old, give or take. This came out of the Peace of Westphalia, and 17th century Europe, etc. The world looked very different before that. And humans existed for a very long time with a very different set of organizing dynamics. Obviously things were more hierarchical, feudalism, things like that.
I mean, again, this is the water we swim in today, this is the world that we know. But like, my point in saying that these are relatively new in the grand scheme of things is to say that they're not always going to be this way. We're going to find different and better ways to coordinate and solve problems.
I think we can all agree that democracy is far from perfect, but it's better than feudalism. It's a step in the right direction. So what is kind of the next step on that journey? And I think that broadly speaking, you know, tools like Bitcoin and Ethereum, which allow people to coordinate in a leaderless fashion. That's really the keyword, leaderless. We never had that possibility before. Every company has a CEO. Every state has a president or dictator or prime minister. This wasn't a possibility before. So it's a really powerful new idea. And of course there's trade offs. Right. You know, I mean, I would say hierarchies are better for some things, democracies are better for other things.
These decentralized topologies are better for yet another set of things. But so that's something that excites me about the core idea about crypto governance. People have been experimenting with things like Teal Organizations. I'm sure you've come across some of this stuff just based on what I saw in your writing. Do-ocracies, you know, just flat hierarchies, things like that. They traditionally haven't worked very well in the corporate world. There's big companies like Zappos is a very famous example that tried to do this and it just kind of didn't go very well and they ended up reverting to a more traditional style.
I'll try to give it to you in one sentence: It allows for a system of governance that is just like more inclusive, more democratic, more participatory, more open than anything we've ever had before. It's just like this core idea of permissionlessness that in the case of Bitcoin, anyone anywhere in the world without permission can just connect a new miner to the network, turn it on and you're off to the races. You're a full peer in the network, you're earning coins, etcetera. The same exact idea applies in our governance. It doesn't matter who you are, it doesn't matter where you come from, doesn't matter what language you speak, doesn't matter how old or young you are, anything. Right. It's very meritocratic. In theory. There's a lot of obstacles in practice, but in theory, anyone, anywhere can just dive in and begin participating.
The initiative I'm leading right now, this House of State project, this is our first in person meeting ever. The initiative is six months old, give or take, and we have people from 10 countries here that came from all around the world. And we all sat in a room all day yesterday to get to know each other, to build trust, and to begin putting this project together. And there's no other domain I can conceive of in the world. No professional setting, no, you know, no government. Where that would be possible. This just random group of friends from the Internet basically come together to like, build a thing and govern billions and billions of dollars of value. That's very unique. I haven't touched upon AI yet, but let me pause there for a second. That's the crypto governance part of the story.
Isabelle Castro
Okay, okay. So, how does AI fit into this?
Lane Rettig
How do we fit AI in here? I think that the key idea is that humans are bad at governance. I also don't think that's controversial to say, like, we've been muddling through. I love that phrase, for thousands of years. It's actually interesting. I had no previous exposure to political philosophy, just the history of governance right before beginning all this. But I did a little bit of crash course, going back and reading Plato, reading Aristotle, reading John Locke, reading, some of the philosophers, Machiavelli, all of it. You know, like thousands of years of kind of like, you know, European thought on governance as well as some kind of Eastern stuff as well. And yeah, you just realize very quickly you're like, wow, we've been grappling with the same problems for thousands of years.
And so many of the, like, you could go back as far as I said, as far as Plato, and like, literally he's describing like 2015 United States politics. Some of the details are different but like the core kind of challenges we're facing are exactly the same. And it's because humans have certain failings, right? We're selfish, we're short sighted, we’re very corruptible, we're very biased inherently. And I think that AI is different. I hesitate to say it's better. I mean, it's certainly better at some things. It is biased for sure. Like we know that, but I think it's less biased or can be less biased in governance.
And I think that between that and the fact that it's, we don't know for sure, but it shouldn't be corruptible the way humans are. It shouldn't be able to bribe an AI the way I can bribe you know, if you're in a position of power or authority. A friend gave me a good example the other day. In our governance system, in House of State, if someone were to submit a proposal, a governance proposal to spend whatever, some amount of money and it just happened to be written in Russian, you know, I don't happen to read or speak or write Russian, right? So I would look at that and I would inherently be biased against it. I'm voting no. I don't understand this. AI doesn't happen to have that particular bias. It would be completely comfortable with it.
So I think that's the first and most important point I'd make, that we need to run this experiment. I don't know yet for sure that it's going to work. I have my misgivings, my doubts about it, but I think we have a good chance of being able to build a governance system that is less biased and fair, more transparent, more participatory using AI.
I'll give you one other quick example. One of the hardest problems in governance is preference aggregation and there's real science here. If you look at for example Arrow’s theorem or social choice theory more generally we know that it's borderline impossible to like understand the preferences that the aggregated collective preferences of a group of people, almost regardless of the size, people don't even know their own preferences.There's this issue of revealed preferences versus you know, versus explicitly stated preferences.
I think AI has a role to play here as well. I think if we get to the point where we can have an AI agent acting on everyone's behalf, every stakeholder, every voter, every consultant constituent, and we're not far from that. I think actually you could imagine a scenario where you boot up an agent to represent you or your group, your family, your tribe, and it just interviews you, just like you and I are doing. We have a conversation and it susses out your preferences and then goes out into the forum, you know, and represents your interests, you know, lets you know when there's a vote coming up, lets you know if there's a particular issue that you need to be aware of.
And even you give it… So there's another concept we can talk about, which is agency. How much agency should these agents have? We're starting with close to zero and we're going to try to slide it up to maximum. But one further step on, it actually votes on your behalf. Basically just on autopilot. You could just sleep soundly knowing that it's out there representing you. So that's actually one of my visions. I think that's what we're going to try to build. There's a narrative globally right now. There was a headline in the New York Times a week ago that AI is beginning to erode democracy globally. There's some truth to that. Right? Deepfakes are affecting elections. They had to rerun an election in Romania earlier this year. Because there's a widely circulated deep fake of one of the candidates.
More and more this is happening and I get that's happening. But I, I think what I would say is that I believe very strongly the exact opposite. I'm beginning to believe this is kind of my mission in this project and maybe my mission in my career as well, to prove that AI has a constructive role to play in governance along the lines I'm beginning to describe.
Isabelle Castro
You said earlier that. I'm paraphrasing, but DAOs have been disastrous.
Lane Rettig
They've been disastrous, yes. In general.
Isabelle Castro
Can we go into that a little bit and then I'll probably ask kind of like how AI might approach that. So why would you say they've been disastrous?
Lane Rettig
Yeah, there's a lot of reasons. I mean, one very obvious reason to me is that for some reason DAOs are very hesitant to meet face to face, which I kind of understand. You know, I use the term like people call a DAO, like a group of Internet friends with a bank account or multi stake crypto wallet, whatever. I guess zooming out a little bit, I would say the core issue is the DAOists don't take this stuff very seriously, right? I think if you're serious about governance, you need to do what we did yesterday, which is get the stakeholders physically in a room together. It is the only way that humans build trust, period. No exception. There's no other way to do it. You cannot do it on Zoom.
You cannot do it, you know, with a pseudonym and like text in a telegram chat group, which is how most DAOs run. It's just, it just doesn't happen. I mean, it's just, you know, there's people begin to mistrust one another. So I think that's the first and most obvious thing to me. I just call me old fashioned, maybe it's my age, I don't know, maybe the zoomers would disagree. But I really believe that you need to do this the old fashioned way. Put butts in seats and get people in a room together, you know, have a beer together, build trust together. Right. And almost no DAOs, do this. I still don't understand why they have millions, hundreds of millions of dollars and they still can't manage to put people in a room together. And I get it's hard.
People live far away, I have family, it's hard to take my son out of school or people have passport issues, things like that. But it's doable. Been doing this for years. It's a test of the fact that it's really doable. That's number one. I think that alone would solve 80% of DAO problems. People who are very compatible online, again, they just sit down face to face like this and like 95% of issues just vanish.
Isabelle Castro
Yeah, true.
Lane Rettig
Yeah. That's one thing I would say. You know, like again, there's inherent trade offs here and I think that the trade off, when you lean heavily into decentralization, so the benefit in theory is you end up with better results. Right. And there's some science here, right? You have a more deliberative process that involves more, you know, more stakeholders, more people. Right. So in theory, kind of the outcomes are better. But the trade off here is that it's very slow to achieve any results. Obviously dictatorship is like the most efficient system. So I guess what I'm getting at is to some extent DAOs will always be messy by design in the same way that like, you know, bitcoin transactions are always going to be slow and expensive. Because you're paying for millions of computers around the world to effectively redundantly do the exact same computation. You know, DAOs are also going to Just they're never going to be as efficient. So they're not well suited for every kind of governance scenario.
I think other reasons DAOs have been a disaster. I mean, just a lot of the maturity, to be honest. It's like, you know, I'll be very frank, it's a bunch of like, you know, 20 something young men who tend to be on the autistic side of the spectrum who are not very socialized. Right. There's very few adults in the room. I mean, I'm a crypto grandpa, you know, even in my early 40s, it's kind of funny.
But even just the fact that I spent six years working in a really professional, really well run, really disciplined organization, like I notice it already sets me apart because I have some idea of what good governance looks like. And a lot of my colleagues, they've never been in that situation before, they've never held a real job before. So I think that's part of the issue.
I think we need more feminine energy. That's another factor as well. I couldn't think anything else jumps out to me about why DAOs have been so disastrous. I mean, I think the final thing I would say is like money screws everything up. People smell money and a lot of these DAOs are governing big treasuries and they are designed to do things like grants and public goods funding and they just attract the wrong types of people. They attract people who smell money. The incentives are not aligned. They attract mercenaries rather than missionaries. So I could go on, honestly, I could probably lecture for an hour on other issues with DAOs, but those are the top of the keep for me, I think.
Isabelle Castro
DAOs have been a disaster, but they are the best approach that you've seen so far, aside from kind of the more centralized approach.
Lane Rettig
Yeah, well there's definitely a centralized, decentralized spectrum. I'm not sure I would say that. I think we are now, it's the first DAO, 2016. Right. And then there's a hiatus and then they kicked off again around 2019. So we're, you know, anywhere between six and nine years into this kind of DAO experiment, depending where you start counting. I would say we're now for the first time at the point where DAOs are beginning to work. Beginning, just beginning to. And there's. I get this question all the time. You know, people say, “well give me an example.” Okay, you're sitting on DAOs a lot. Give me an example of one that you think is working. And I really, really struggle to do that. I mean there's kind of some Arbitrum is the one that comes up the most often. They're very large, they're reasonably functional, but like they've had a lot of issues, continue to have issues. And partly it's the money factor. As I said, there's just a lot of kind of money on the table. But I love them for really leaning into like decentralized governance. And they're definitely like a good example of functional DAO.
So I think we're on like 2 gen 3 or gen 4 now of DAOs. And I think that House of State we're building, I'd call it like a Generation 4 DAO, which is to say that we're learning from the mistakes of all the projects that have come before. And I could talk about some of the ways we're doing that. What did I want to say? What was the question you asked just now? Sorry, I lost my train of thought.
Isabelle Castro
Are DAOs kind of like the best approach?
Lane Rettig
Yeah, so just to close the loop on that. So I think what I would say is up to this moment, definitely not. I think that the most functional part, like the way that NEAR has been governed is BDFN. Have you heard this term before?
Isabelle Castro
No.
Lane Rettig
BDFN stands for benevolent dictator for now. So this comes from the Linux world, of the Python world. So you know, Linus Torvalds created Linux and he was the benevolent dictator for life. And then there was Guido van Rossen was the guy who created the Python programming language and he was kind of like the benevolent dictator for life. And both of them eventually, many years later, did step down and created more robust governance structures. And, and actually we learn a lot from Linux Foundation, Python projects like that are just open source projects. That's actually the progenitor of all of this. We're kind of downstream from that even to the point where EIPs, the Ethereum Improvement Proposal process is based on BIP's – Bitcoin Improvement Proposal. People don't know this, but Bitcoin improvement proposals are based on pips, Python improvement proposals. So there's a clear lineage there.
You can actually look at the first BIP was literally a copy and paste was a fork of the Python improvement process proposal process. And of course they've evolved from there, but I think that the best run projects have been a bit more centralized up to now, including NEAR. But it's clear that we can't do this forever for a whole bunch of reasons. I mean, one is that like the dictator in our case, Illia, just he's never wanted to be dictator, he doesn't want to be a dictator, you know, and literally his mandate to me is like, can you get me out of a job please? You know, I want to retire five years from now. And the other reason is it's just a funding factor as well, like foundations.
You know, the NEAR foundation is super well run, we have a very large healthy treasury, but it won't last forever, we want NEAR to last forever. So how do we put in place a sustainable governance and also funding structure so the whole thing can just sustain itself indefinitely as like Linux has, for example. It's probably the best, most positive example we have in the world of functional decentralized governance of software project.
Isabelle Castro
So does the AI piece, I appreciate that you're going to be experimenting with it, but does that kind of feed into the decentralization and do you feel like it's going to bring more decentralization? .
Lane Rettig
Yeah, yeah, I think so. That's the goal, right? So the most obvious example of this, I touched upon it a moment ago when I suggested like, you know, the Oprah Winfrey thing. You know, it's like you get an agent, you get an agent for everybody, right? That's the idea.
So one of the biggest issues in DAO is, I didn't even mention this one is voter apathy and on the other hand, control by whales. There's countless examples of DAOs where there's this high stakes vote and the vote is running for a month or two months or three months and in the last hour, and it's like 60: 40 or 70: 30 or something in favor of some proposal the community favors.
And literally in the last hour, the last minute, some whale comes in, some anonymous investor and just tips the scale in the other direction. It happens all the time. And I think as a result a lot of voters are very apathetic in DAOs or they're just holding tokens because number go up, not because they're excited about governance. So most DAOs have struggled to get participation into the double digits, actually, which is a huge issue because you can't, you don't have a quorum, you don't have legitimacy in that case. So I think the most obvious example here of the role AI has to play with respect to decentralization and governance is again, if we can get to the point where we have agents that people trust to act on their behalf, we could get theoretically participation to, I don't know, 100% is pretty ambitious, but like, let's say, you know, 60, 70, 80%, I consider a huge win.
Isabelle Castro
Do you foresee any kind of issues, I guess, or stumbling blocks that we might have from having agents as our kind of out default proxy vote, just them doing it for us and we just kind of like let them do it.
Lane Rettig
I mean, I'm describing the utopian version of the world, you know, which is like I always come back to that movie Wall E, you know, where all the humans are just like, you know, on recliners and, big bellies and they're just like sipping sodas and watching TV all day while, you know, the robots and the AIs do everything on their behalf. I mean, maybe that's not utopian, but you know what I mean? My utopian outcome is exactly what I'm describing, which is like, I have an agent that pings me once a year and says, oh, you need to refresh your preferences or something.
I don't know if this will come to pass, but I think it's worth trying. I'm sure there's a laundry list of potential issues. Like the most obvious one is just alignment issues. And I know that you've written about this recently. You know, this gets into areas of alignment and AI safety, which are huge topics outside of Web3. Just, just in kind of AI in general, you know, how do we know the AI is aligned with our interest? How do we know it doesn't in its present form, it's not actually thinking, it's not really reasoning, and it's certainly not doing it from a place of a moral code. It's a pattern matching machine. So even to the extent that it appears to be acting in line with our preferences, it's fragile in a way. There's all sorts of failure modes. You know that it agrees with you on every issue, and let's say it faithfully expresses your preferences. Every issue except one, but that one happens to be the most important one to you or something. That's one failure mode.
There's the issue of adversarial failure modes, you know, where, like, we're running models that we think we understand them, but we don't because we don't know how they were trained. We don't know the data they were trained on, we don't know the parameters they were trained on. And the companies that do this, we talk a lot about trust in crypto. Like the whole point of crypto is trustlessness, right? You don't have to trust what someone says because you have cryptographic proof of it or you can verify it yourself. That's not the case in AI today. So I can spend millions of dollars and outsource training a model to some one of these third party model trainers, like three or four companies that do this. And they come back and hand me a big blob of data, you know, 10 gigabyte file or something. There's no way for me to verify that they did what I said, that they even trained it on the number of parameters they said they did. And there's some really amazing research into these things called sleeper agents.
So Anthropic published a paper a couple years ago where they built a sleeper agent into a model and then it passed all the safety checks, but they were able to trigger it with a 50 particular trigger work. Right. So this kind of thing, like, you know, it doesn't matter so much in DAO governance. I mean, the failure mode there is relatively limited. This, by the way, is why I think we should be doing this experiment in DAOs and in crypto today. Because it's lower stakes. You know, they're not controlling airplanes or hospitals or trains or things like that, you know, electrical grids, like I think eventually they will be. But like this is low stakes in the grand scheme of things.
But yeah, I mean, if you were to be doing real world governance with these models and you know, I don't want to pick on any particular country, but like, you know, a model comes out of some country that you have an adversarial relationship with or something like, who knows, you don't know how it was trained. You don't know if there's not like a sleeper agent in there and that, you know, one of the examples in the Anthropic paper, they just flipped a year. They went from, you know, 2023 to 2024 and it's switched into evil mode.
Isabelle Castro
Oh, really?
Lane Rettig
Really. I mean again, there's like real scientific research. This is very doable and again, past every safety filter. So we have solutions to this, you know, verifiable model training, things like this. They're a little bit experimental right now, but this is some of the things we're going to be doing research on.
Isabelle Castro
I'm really excited to see where your research goes with this. I assume decentralized AI will kind of be the fundamental focus for you?
Lane Rettig
Not so much, actually. I mean, NEAR is a little confusing because we have the NEAR foundation, we have NEAR AI which is a separate company. So NEAR foundation is a Swiss nonprofit foundation. NEAR AI is actually a for profit American company. And there's a relationship between the two of them. But like NEAR AI is working on that. That's kind of their area of focus. That DCMML, which stands for Decentralized Confidential Model Machine Learning. So the fancy word for decentralized AI, decentralized training.. At House of State, like in my kind of corner, in my initiative. I don't actually care that much whether we're like running on centralized or decentralized rails. I think we'll kind of use whatever is the best solution. To be honest. Right now it is more centralized stuff.
Like I'm totally okay if we're using, you know, like ChatGPT or you know, Anthropic or one of those, or you know, Gemini or something. But in order to solve the bigger problems, I'm talking about like for example, this, you know, these are longer term issues, but the verifiable model training. One of the things I've done is I've done some BD alongside NEAR AI and I've had conversations with some sovereign actors. Right. So some like. And NEAR’'s been in touch with, you know, governments in Europe, in North America and in the Middle East and the very least maybe others that I'm not aware of. And for them it matters enormously..
If you're a nation state or city state, you want to have absolute control, complete sovereignty over the entire AI stack. You want to know how the model was trained, you want to be able to prove that. You want to know how the inference is running, make sure it's running faithfully, make sure that there's no exfiltration, nothing's kind of leaking out of the model. So that is again the kind of stuff that NEAR AI is working on. I think that stuff will be important for us down the road, but it's not so essential at the stage we're at right now because we're still in an experimental stage right now.
Isabelle Castro
So you do think at some point we will get to the point where a DAO or some kind of decentralized system may be able to govern cities, the larger scale things.
Lane Rettig
This is the network state idea, basically. It's a good question. I mean, I don't know if I agree with that framing. Like, I don't think it's the case that a DAO that starts as a DAO and is called a DAO, he's gonna what, like, plant a flag somewhere, like build, raise an army? How does a DAO run a city? Cities all have governments already. I don't see that happening. I see. I like that the network state idea a little bit more, which is more like kind of things happen in parallel. So we're kind of building an alternative structure.
There have been experiments of, not in crypto, you know, just kind of like default world experiments with things like running a city in a kind of flatter, more decentralized, cooperative manner here and there. You know, there's like famously the Paris Commune, you know, back in the day, that was like a communist experiment and they just hadn't really gone very well. It just tends not to scale very well. So I'm a little skeptical of the idea of DAOs actually running like real world infrastructure. I see it more that we kind of have these two parallel structures for the foreseeable future.
That's an interesting question though. I haven't thought deeply about this. Like, would that change what would have to come to pass for us to want to govern? I keep using the term default world, but kind of like off chain infrastructure. Cities, states, nation states, whatever. It might be using something like a DAO. I mean, I think it would probably be more like a colony on Mars. I just don't see it replacing what's happening on Earth anytime. Like in my lifetime or in your lifetime.
Isabelle Castro
Interesting. Yeah, it's really interesting because I've just had a conversation with Jarrad Hope.
Lane Rettig
Yeah, I've known him forever.
Isabelle Castro
Yeah. Yeah. And he really looks into DAOs in that context.
Lane Rettig
He's a visionary, He's a utopian. He's a big thinker. I wish him luck. I think my vision's a little bit different than his. I think I'm more of a pragmatist. Maybe that's coming across.
I mean, I think we use these tools for what they're good at, but I mean, and coming back to this, I just don't see a way for a DAO to govern a city without violence. I just don't think you get there. I don't see a city. I mean, again, maybe like, fine, like some small experiment, you know, a city of 20,000 people in, like, rural Idaho or something. Sure, that will probably come to that. They'll probably happen. Someone will create like a, you know, golf's gulch somewhere or something, you know, try to govern it with a doubt.
I think that's cool, like those experiments should happen, but I just don't see that, you know, scaling to the size of a major city or state or something like that.
Isabelle Castro
Oh, interesting. Why do you think violence would be involved?
Lane Rettig
Because, I mean, they are all like every inch of land on the earth's surface with one or two very tiny notable exceptions, are run by existing governments. And they're not going to step aside. And it's their vested interests, their entrenched hierarchies, and it's not in their interest to turn themselves into a DAO. I just don't see that happening without force. And I also don't see DAOs raising armies anytime soon. So that's what I'm saying. I don't see this kind of changing.
Isabelle Castro
Yeah, no, I agree. You probably can't speak too much because it's sovereign states but why were they contacting you about your work at NEAR?
Lane Rettig
Yeah, so it's not anything super, super aggressive or ambitious, yet. It's things like. Citizen concierge services is the most common. So you're seeing banks do this. You're seeing airlines do this right there. And it's already, you know, Singapore Airlines and Emirates are two examples of this. These are among the best airlines in the world. Right. Like, you actually go into their apps. They actually have already, like, AI concierge services built in where you can, like, chat and it will, like, help you book flights or change flights or if you need. Like Emirates has one where if you are a passenger with special needs, they have an AI concierge service that can be used to help you navigate your whole journey and, you know, make sure you have a wheelchair where you need it and this and that kind of thing.
And so Nation states, Yeah, I guess so. Including some larger ones. They're looking to build citizen concierge services to offer Just, you know, services to the citizens. So this would be things like renewing a driver's license or paying a fine or you know, booking medical appointments, things like that just, just through a single AI interface. But these are very sensitive things. Medicine is, you know, obviously like your medical record super sensitive. So that's why it's so critical to them that these things be kind of like on premise sovereign. They have full control of everything, that kind of thing.
I'm sure there's military applications too. I don't know anything about that.
Isabelle Castro
Yeah, well, even if DAOs aren’t the ultimate thing for city states or nation state, do you feel like early web3 communities could give network states or startup cities lessons? And if so, what would those be?
Lane Rettig
It's interesting you asked the question because most of the kind of pop up city, you know, experiments I've seen are run by crypto people actually. So it's like we're giving ourselves advice. Yeah, there's so much overlap. If you look at the Venn diagram, you have kind of Burning man people, you have pop up city people, you have crypto people. I feel like it's kind of one big circle. So much overlap. I mean this is the thing though, pop up cities are called pop up cities for a reason. I keep coming back to the scaling idea.
I go to Burning man pretty often. This will be my fifth year doing it. I think it's, I mean it's a great example because it's pre blockchain, it's pre crypto, et cetera. But it shares a lot of the same kind of ideas and principles.
It's a week and it's on the order of 100,000 people. I don't think it really scales either in time or in size beyond that. I mean it has kind of gradually grown to the size it may even kind of continue to grow linearly in terms of size. I'm just really skeptical and I should say it's a great example. It's one of the best examples we have. I mentioned Linux earlier of really functional, successful, decentralized, participatory governance. There is no leader in Burning Man. There's just an organization that's there to provide very basic rails and everything else is just people coming together and building it and taking it down. And it works really well. But again, it doesn't really scale.
At the end of the day, we probably share a lot of values. Burning man has a set of 10 principles. They're just really beautiful principles, and it's a great starting point for any community-led project. I'd start with that. I mean, my first piece of advice would be, a mistake I've seen a lot of DAOs make is they kind of put the. How do I phrase this? They put the governance cart before the foundational horse, which is to say, you need to get the basics in place first. Right?
They go off to the races and they immediately try to do things like giving out money and grants and things, but it gets very messy because they didn't take the time to create a constitution. I like the term North Star, Right. So this refers to mission, vision, values, principles, constitution, basic operating procedures. You know, just like a very kind of foundational set of agreements. You need to kind of, like, get that in place. We're doing this process right now with House of Stake, and it's actually fairly complicated because you need to, like, ratify them somehow. In order for them to have legitimacy and authority. They need to be ratified. How do you ratify them? You know, when you have this, like, community of thousands of people, It's a complicated, nuanced question.
I think blockchain ecosystems are beginning to resemble nation states with similar governance structures, similar policies, things like that. So we really need to be thinking about what our GDP is? What is the productive capacity of the ecosystem? How do we bring the right people, the right projects, into the ecosystem.
Yeah, I think just kind of start with the basics. I spoke to this before when I was talking about the dysfunction in DAOs, and I said that the thing that DAOs need to do is put people in seats together. The beauty of putting people in seats together is you get democracy by definition. You put some people in a room together, and everyone gets a vote. That's just like a very kind of natural way to do things. You can't do that online because of sybil attacks. Immediately. You get people creating fake accounts and things. This is how a lot of DAOs fail as well. But the advantage that you have if you're doing a city is you can do democracy, right? So I think that they will probably be more functional than DAOs.
I think actually I might flip it around and say we probably have something to learn from them. We should probably try to model a lot of our governance after, for example, things like charter cities. I just think that they're going to be more functional. They're going to use more like tried, tested, true governance procedures and things like that.
You want to do rough consensus. I don't know if you're familiar with that. Rough consensus is a very interesting governance mechanism that is an alternative to voting, right? So because we don't have democracy, because we don't have a notion of one person, one vote on chain, we have to use other options. And one of them is this thing called rough consensus. I really like it. That's actually an RFC. RFC stands for Request for Common Rights. It's actually a formal document published about eight years ago by the IETF, the Internet Engineering Task Force. It's called On Consensus and Humming in the IETF. I highly encourage you to read it. It's one of the funnest, most interesting documents in the world about governance. And it talks about how IETF governs all the Internet protocols. So all of our SMTP, POP3, the things that email is built on, HTTP SSL, the way our devices communicate securely with web servers, things like that. It's all governed by this agency, this international body called IETF. And they don't do voting for a whole bunch of reasons. But rough consensus, it's really very straightforward. You kind of put people in a room together virtually or in person, and then you have a proposal. Someone says, “We're going to do X. Here's my proposal for our standard for this new protocol. We're going to add this feature to a protocol. In the case of blockchain, we're going to disperse these funds to these folks for these reasons.” And you open the floor and let anybody raise an objection.
You just continue doing this for however long it takes until all the objections have been addressed and acknowledged. In some cases you can kind of update the design to factor in objections. In other cases, you kind of decide, okay, we're going to table that, we're going to come back to it later. This is not relevant. And at the end of this process, when no one has any more objections, the chair of the session says, okay, we have rough consensus and then this motion passes.
Yeah, I mean it's an alternative to voting and that kind of thing. I think we're doing a lot of this in crypto governance and sure, like you should try running a city. That'd be really interesting, right? Or quadratic voting is another example. Or liquid democracy is another example. There's like a lot of really fun experiments in, you know, in governance, some more democratic, some less democratic that I think it would be interesting to see them put to practice in the real world. There's a little bit of this happening. Like the Democratic Caucus of the state of Colorado implemented quadratic voting, I think in partnership with Glenn Weyl and some of those folks.
Taiwan is famous for having a very interesting, very participatory democratic process that's evolved over the last few years. Initiative called Gov0. I don't think you've got a chance to look at it. You should speak to some of those people. Speak to Audrey Tang, the digital minister in Taiwan who's behind this and close collaborator of Glenn's. So there's some of these experiments beginning to happen in real world political bodies.
Isabelle Castro
Right and conversely, we've got centuries of governance systems to pick from, but what are the top elements of governance historically that you are trying to input or are learning lessons from when you approach creating House of State.
Lane Rettig
Yeah, it's a good question. I mean, democracy comes to mind. But as I said, we tried this. Right? So the previous governance initiative in NEAR, which was called NDC, NEAR Digital Collective. It was a total disaster. And one of the reasons is it tried to be democratic. They tried to do one person, one vote. They tried to have some civil resistance mechanism, some collusion detection mechanism or something and it just didn't go well. People just like attacked the system with fake identities. So that's sort of off the list, at least for now.
Another one is checks and balances, right? We love to talk about, especially we Americans, you know, we have our tricameral system and we have the judicial branch, executive branch, the legislative branch.
There's a lot of logic in that. It's a system, it's very robust. You know, it's kind of proven to work, you know, I mean, you have to mention that thing was called about democracy, right? It's like the worst system of all, except the worst possible system, except all the others that have been tried from time to time was one of the British premonitions, I can remember. So yeah, that's a great idea. It's messy, it kind of works. But we also tried to do that. That was also part the previous NEAR governance project. And also it was just too complex. It was too much, too soon. I think at some of the courts are taking a much simpler approach.
I kind of like to move in that direction of introducing different governance bodies and checks and balances and things like that. But it was too heavy of a lift, I think for a DAO. Another issue with DAOs is they're really bad at execution. So we need an executive branch, which today is the foundation.
So sorry, I'm not answering your question. So I mean, I think that the honest answer, the thing that comes to mind, I'm kind of repeating myself here, is like the Citizen assembly, that's probably the oldest, the best example of a real world governance mechanism that I'm using. Literally that's what I did. I flew people here from around the world, put them in seats together and we just sat in a room and talked and raised our hand to express preferences and issues. We're definitely doing that. And we're going to continue doing that. It's really old and really boring, but I think it's really essential.
It's the only thing I know that works. It's expensive and it's unpopular but we're going to push back against that, I think.
I mean there's a lot of interesting experiments happening in crypto. I mentioned, I gave you a whole laundry list of them a moment ago. But there aren't like, there aren't actually that many interesting mechanisms in kind of default world off chain governance. You know, like there's only three or four systems of government globally. Right. Like you have kind of democracies, you have constitutional democracies, you have parliamentary democracies, you have, you know, dictatorships, you have, you know, communist states. That's most of them. I mean, the Swiss system is very interesting. It's a little bit unique, where they don't have a head of state, the federal system, I guess the Swiss system is also a federal system, but it doesn't have a federal government. They just have the cantons. I'm not sure there's anything above it, actually. It's a much more decentralized system. So something like that might be interesting to experiment with.
Isabelle Castro
Okay. And obviously you've had your kind of forum yesterday. What were kind of some of the conclusions that came out of that and where are you focusing?
Lane Rettig
First, we have four primary work streams. Right now. Let's see if I can get this right. So we have treasury management. That's a big one. House of State will be inheriting a Treasury that is on the order of seven, eight million US dollars to start. That's happening in a few days. And then there's a much larger pool on the order of US$100 million on the NEAR protocol. So we have a portion of our every block, a portion of our kind of issuance or inflation goes into a special vault, a special treasury that is earmarked for community governance that has not been touched yet.
But over time, if the DAO matures, if it kind of proves itself and that it can use this initial kind of pool of capital maturely, then it's going to inherit this much larger pool gradually. And the goal here is to come up with something that's truly sustainable, and can last indefinitely. So treasury management is a big important part of that. I learned this the hard way. I was very naive. Like, yes, you have a pool of funds, you can just spend it. But, actually, no, you need to invest them. You need to balance your revenue with your with your expenditures so that ideally it can sustain itself indefinitely. And we have a great team of treasury managers in the NEAR foundation that are doing this in a very centralized fashion. So we want to kind of like replicate their strategy in the DAO. So that's the first work stream. To some extent this might involve investing in other protocols, other projects, DeFi, things like that.
The second one is governance processes and procedures. This is like meta governance. How do we establish the governance mechanism? And the focus of that right now is, as I said, the North Star stuff. Mission vision, values, code of conduct, conflict of interest, policy, really foundational, important stuff. They're making great progress on that.
The third is network economics and security. So one of the main goals of the House of Stake Initiative is to improve the network. The economics of the NEAR protocol. The tokenomics. This is a very key thing. In the same way that a nation state has a central bank or Federal Reserve or something that manages monetary policy, fiscal policy, things like that, we need to do the exact same thing. And there's a lot at stake. The market cap is on the order of a billion dollars, right? So there's a lot at stake there. So for example, in your protocol, we have 5% inflation right now, which is not the highest. Some protocols are higher. Some are much lower. Ethereum is much lower.
There's a very big debate happening in the entire kind of crypto ecosystem right now about what is the right number, what is the right issuance or inflation number, how much of these protocols be paying for security, proof of stake protocols. And these numbers were all, I mean, I was part of the process. It's literally like stick your finger in the air. These numbers were kind of chosen in a relatively arbitrary fashion five, six years ago. So we didn't really know better. We have a lot more information now. And it seems to be the case that probably all these protocols are paying way more than they need to and this inflation number should come down. That's one of the goals of the DAO and that's one of the work streams. That's three of them. And the fourth one is ecosystem growth.
So this is kind of a bit of a catch all. And this overlaps the most with what the foundation is doing right now. Very broadly speaking, it's about talent. We want a talent pipeline. We want talented builders, talented founders to come build things in the NEAR ecosystem. I really think of blockchain ecosystems as nation states. It's not a perfect metaphor. There's obviously some ways in which we're very different. But in the same way that countries in the world are competing for talent, some are experiencing brain drain, others have really wonderful visa programs, you know, golden visas, things like this, where they want to bring talented people in to contribute to the GDP.
I think blockchain ecosystems are beginning to resemble more and more, kind of nation states with similar governance structures, similar policies, things like that. So we really need to be thinking about what our GDP is? What is the productive capacity of the ecosystem? How do we kind of bring people in, bring the right people, the right projects, etcetera, into the ecosystem. So that's what ecosystem growth is focused on. Debating questions like so they'd be responsible for things like giving out grants to projects to come build in NEAR. So how do we gauge the potential, the quality potential kind of return on investment of projects like that? So that's flavor of some of the stuff that's happening right now. And this is a super young project. It's only a few months old. We're already doing all these things. It's really intense.
Isabelle Castro
Yeah, it's amazing.
Lane Rettig
And I didn't even touch on any of the AI stuff because that's all happening too.
Isabelle Castro
Yeah, I was going to ask, where do you see AI kind of coming into that?
Lane Rettig
Yeah, so that's begun as well. So we actually had a demo yesterday of our first AI governance tool, which is really cool. It's a copilot. So we start with a very basic thing. And what this does is it plugs into our proposal system and you can just chat with it and say, “What proposals are on the table right now? What proposals are coming? I want to learn more about proposal number 17. How do you suggest I vote about it? What things should I be considering? What are the potential pros and cons of this proposal?” That's already live, it's already working. It's not technologically that hard. It's basically a GPT that has access to the proposals and some other resources about the NEAR project in context.
From there we want to build AI delegates. We have 11 delegates right now. These are our MPS, you could call them. These are the people who actually receive delegation on the part of other stakeholders, other NEAR token holders and basically vote on proposals. We want to expandwhat we've built so far, the AI chat bot copilot type thing, into an actual voting agent that can actually receive delegation and vote on behalf of the people who delegate to it. And we'll probably have multiple of these and the goal is for there to be maybe three of them that have what are called policy documents. A policy document represents the preferences of a particular group of stakeholders in our ecosystem. So you can think of one being, you know, more right leaning, one more left leaning, one more centrist, kind of like that.
And there's some, and there's some precedent for this. There's some projects that have actually built things like this. So Gitcoin, in their most recent round of grants, they actually did exactly this. They actually had three AI agents with these three policy documents and they did this clustering exercise where they tried to capture the preferences of the people in the ecosystem and then gave a portion of their public goods funding grant pool to these agents to allocate. So we're going to do that. That's kind of the next step.
And then the vision, the longer term vision is as I alluded to earlier, kind of like a swarm intelligence, which is where every single voter, every single person in the ecosystem has like an AI agent that kind of knows their preferences and acts on their behalf. But that's like at least a year away.
Isabelle Castro
I like that it's only a year away.
Lane Rettig
Yeah, it's a pretty impressive time frame.
Isabelle Castro
Well unfortunately we've come to end of our time. But thank you so much, this has been fascinating.
Lane Rettig
Thank you so much. Yeah, it has been a super thoughtful conversation. I really enjoyed it.