Alignment Is Intelligence: What Communities Figured Out Long Before AI | James Young - Collab Land

All right, can everyone hear me? Welcome, East Denver. My name is James. I have some technical difficulty, but we'll figure this out together. Today's talk that I'm going to be presenting is alignment is intelligence.

And if you look at the agenda actually the title of the talk was originally coordination is intelligence. Uh but over the last few weeks since the presentation submission there's some keen insights that I would like to share some alpha with the group here. So I changed the title of my slide. Now what I'm going to talk about today is trust and how trust is actually the core primitive that we need to solve for Dows specifically. So in a community context and for AI and how they both are going to converge and this is how AI and crypto are going to be integrated together.

Now, when it comes to trust, I'm going to be talking about two types of trust. Hard trust and soft trust. And first, I want to just say that Ethereum has solved the hard trust problem. And that is no small feat. And I would say and I would kind of emphasize that Ethereum's ability to have this coordination breakthrough is actually one huge achievement of our time.

But for Dows, it showed a problem. Who in the audience knows what a DAO is? Okay, a DAO is a decentralized autonomous organization and Dows have really shown us some uncomfortable truths. Oh, and I'm going to talk about that today. And what I'm going to talk about is the pairing of hard trust with what I'm referring to as soft trust and why AI cannot be ignored here when it comes to soft trust.

I helped co-write the Mollik Dao white paper in 2018 and for the last six years I've been running a project called Collaband with my co-founder and wife Aneli Young and this talk is hopefully useful in that I'm going to share my experiences with running collab for over six years in the context text of Dows. So, let's start with Dows and why they keep failing. And I think people have seen this before. You have a Dow, you're able to coordinate, you see a huge momentum and then quick decay. And there are many reasons for this.

And I believe that it's not a tooling problem. It's an organizational issue. And I believe that we have been looking at the wrong metrics. So what I'm stating here is that coordination is necessary but it's not sufficient. Who here knows what Goodart's law is?

Okay. So for those that are in the audience and in the recording that aren't familiar with Goodart's law, it's when a measure becomes a target, it no longer is a good measure. And I think this previous phase of Dows, we've been looking at the wrong measurements. I don't blame anyone. I think it's just natural because we have been taking onchain metrics like voting engagement, token distribution, TLV, forum activity, reposts, but these are market metrics and Dows and tokenized communities in general actually need more robust soft trust metrics.

And so we are in this trust gap. How I'm defining soft trust is behavioral alignment over time and onchain metrics are not good enough. So coordination is necessary. You need the infrastructure for Dows. You need tokenized communities.

We at Collaband help coordinate tokenized communities and Dows. Actually, Collaband in its first incarnation was a DAO tool and it's expanded beyond that. But what we need is in AI parliament what they call coherence. Now what is coherence? It's belief and actions sustained over a period of time.

And this is what is known as behavior. And with this organized behavior, you need to look at other signals because market signals are not enough. And this is where AI agents come in because when you think about Dows and the reasons why they fail is that they're able to initially coordinate but they cannot sustain this coherence or alignment over time. When you look at AI and a specific subset of AI called AI safety, it really boils down to a trust problem. This is why people refer to Skynet.

The AIs will control us and there isn't this long-term alignment over time. And before I get into uh talking about a reference implementation, something that we're doing at Collabland, I I want to take an aside for a second and get a little nerdy here. How many developers are in the audience? Okay, so all the rage right now is open claw with these prompt injections which is very similar to those in the developer that developers that know about crosscript attacks. It's these pre-prompt injections are very similar.

So you'll see these agents on the internet. Internet is a hostile environment. And you see, no matter how hard you try to secure a private key for an agent, eventually it leaks. You get these pre-prompt injections. So, what we've done at Collaband is partnered with a project called Taco.

And if you're a developer, if you're looking at the intersection of crypto and AI, I highly recommend to go to taco.build. What taco.build is, it's a key abstraction MPC network, meaning agents don't need to have a private key. They actually have smart accounts.

Who in the audience knows what a smart account is? Smart accounts allow for a different primitive. When you have EOAs, it's like having root access to a computer where you cannot change the password. So once it's leaked, you you can't recover. And so with Taco and their key abstraction network using smart accounts, you don't have to have an agent manage a private key.

What agents do is they are delegated fine grain permissions. So MetaMask in our research at Collablam has the best most finely grain permissions so that you can not only give delegated permissions to your agent and the agent can give those delegations also to a swarm of agents. With the MetaMask uh smart account kit, you have what's called caveats and caveat enforcers that before funds leave or are retrieved from a delegation, you have what's called caveat and caveat enforcers. This sets the groundwork for what I'm going to be talking about here that makes this AI and crypto merging not only seamless but scalable. And one implementation that we're looking at with AI agents in the context of Dows is having a digital twin.

And let me explain. Before I explain though, we now have all of the standards in place. So for the nerds in the audience and for those that want to understand with the foundation of key abstraction and smart accounts, you have standards like 8004 which was recently pushed to mainet. You have 4337 and then you have a constellation of EIPs like 7702, 7710, 7715 and there are a couple of others and I highly recommend the developers in the audience in the context of smart accounts and delegations to also take a look at these standards. Now, it can be somewhat overwhelming if this is the first time you're hearing about this.

So uh I highly recommend reading it and there are SDKs available where you can vibe code this. So imagine a DAO and this is how it works. Imagine a DAO with only agents and these agents are connected to an operator and these agents are going to be playing a prediction game with you similar to a prediction market. And what the agents do is they create a scenario and the scenario will have multiplechoice answers associated with it. And the agent before sending you these scenarios, it also guesses and uses a ZK proof so that it cannot be revealed what the agent would predict or guess you or you as an operator what your answer will be.

If it guesses it right, what we're calling coherence, your coherence score goes up. If the agent guesses it wrong, your coherence score goes down. And I don't have internet connection else I would show you a demo of this. I will at the end of this talk there's a QR code if you want to sign up and actually test drive this game. This was vibe coded in just a few days.

But how does this scale? So you have coherence with your a agent and once your coherence score reaches a threshold, the agent can now perform interactions on your behalf because now you have coherence with your agent. you have trust with your agent in the DAO setting. These agents now can have coherence with one another. So now you have a coherent DAO.

Once you have a coherent DAO, your DAO can actually combine and group together with other Dows, which effectively gives you this network effect of coherence. So, not only does it help keep Dows aligned, it not only can give us insight into AI safety and how you can trust the AI, but I firmly believe that this is how decentralized AGI will emerge. So, AGI is not going to be a frontier model that knows everything. It's going to be an emergence of intelligence from a network set of agents. So in all what I'm trying to implore everyone is that let's put the A back in Dows and if you want to test drive this scan the QR code access code is E Denverver if you want to get priority access.

Thank you very much.