AI DAO and its three paths of realization
Author: Trent McConaghy
secondary title
AI DAO and three paths to its realization
AI DAO and three paths to its realization
(AIDAO Series, Part I)
DAOs have arrived. When artificial intelligence (AI) is added to the mix, the results are explosive.
I was about 10 years old when I stumbled across a book on artificial intelligence (AI). It captivated me, immerses me and impresses me. Since then, I have been passionate about it. I have spent nearly 20 years as a professional AI researcher. Three years ago, I stepped out of the world of artificial intelligence to work on something equally important: decentralization. These worlds will soon collide."This article is the first part of a series. It considers how to build such an AI DAO. The second part gives details on how it was built, as well as recent implications. Section III describes human-scale effects ("), and a possible solution.
secondary title
Hype/Tech Timeline
Decentralization started with Bitcoin, then Blockchain, but it didn't end there. I've realized it's not just one wave, or even two waves. I currently think of it as five in a row, one wave followed by another with a crazy crash.
Trent McConaghy on Twitter
Hype/Tech Timeline. 1 Bitcoin 2 Blockchain 3 Smart Contracts, Decentralization 4 DAO 5 AI DAO We just exited 2 . 5 will be fun after.
The importance of each wave is not diminished by the waves that follow.
In the fifth wave, the AI will arrive with a vengeance. So, how did we ride these waves so fast? faster than expected.
This made me think more about AI DAO.
AI DAOs
secondary title
In order to describe what makes AI DAOs special compared to normal DAOs or normal AI, we have to start with an understanding of DAOs and AI."I think DAO is an autonomous computing process with resource manipulation rights on a decentralized infrastructure. The Bitcoin network and the Ethereum network are DAOs. DAOs can live on top of other DAOs; for example, The DAO runs on Ethereum. It's just a script, it's waiting for some transaction to trigger it, and it runs according to the logic of the code. but most people think"triggered
DAOs are still DAOs because they don't require any human intervention to function, and they can't be shut down because they run on decentralized infrastructure.
There are several definitions of artificial intelligence: a collection of fields that aim to achieve narrow intelligence (such as designing analog chips using deep nets); a field that aims to achieve more general human-level intelligence (artificial general intelligence); and Generated scientific and technological capabilities. Progress in narrow areas often leads to progress in general areas. One can also strive to directly advance the general field.
Therefore, AI DAO is a DAO using AI technology. In other words, it is AI technology running on a decentralized processing substrate.
So, when and how might we achieve AI DAO? When that happens depends on how, but it feels quick.
secondary title
three paths
There are at least three ways to do AI DAO:
AI at the (Smart Contract) Edge
AI at the center (smart contracts).
Let's explore.
1. AI at the edge (smart contracts)
Starting with a single DAO, it acts like a clearinghouse for decision-making at the edge. Each edge is an intelligent decision-making entity: direct AI (like now, in a DAO), a human-organized entity like a prediction market or reputation system, or an AI that doesn't require human intervention to make decisions. Just like human-based entities have various biases and decision-making frameworks, so can AI.
Consider if all edge entities were AI, and control of the AI had been ceded by its creators to the running AI itself. This will be an AI DAO. You can also have some shades of gray in between, some of the edges are AI.
image description
The easiest way is to do it on the DAO itself. If you are a token holder, simply relinquish control to another smart contract running an AI-like to make key decisions: for a given proposal, whether to join a quorum, whether to vote for/against, whether split, and so on. This AI-like entity can start out as outrageously simple. As a simple non-AI baseline, start by saying to everything"yes". Next level: use a random number generator, based on previous block hashes. The next level up: maybe a linear classifier that gives if some weighted combination of inputs passes a threshold"yes"yes
. You can get progressively more complex, like a feedback framework for sensing inputs, updating state, and making decisions, implemented in modern recurrent deep networks.
Imagine if all the token holders of the DAO gave up to AI bots. Also interesting, admin: we're going to have $150 million managed by an AI that you can't turn off. Each bot holding tokens can also become more sophisticated, with its own automated mini-marketplace to make decisions. DAOs could end up being fundamentally more complex and automated than we thought.
secondary title
In this case, a more complex AI entity would be the core of the smart contract. This AI entity will be a feedback control system. Its feedback loop will continue its work, accepting input, updating its state, and executing output, with resources to continue doing so. This framework is how most modern AGIs (artificial general intelligence) are set up, and also DeepMind's AlphaGo. These building blocks can be deep nets, recursive deep nets, genetic programming, or other artificial intelligence methods.
image description
An example is an AI marketing organization. Let's say you have a company that makes shoes. You give marketing dollars to an AI DAO. AI DAO finds out who in the human social network can best market your company's product and automatically sends funding and marketing requests to them. Over time, humans build a reputation for being good at delivering high marketing ROI, and they get more assignments. (Thanks to Meher Roy for the help!).
secondary title
My point is that there is a network of smart contracts, everyone is stupid, but great wisdom emerges automatically. This is how ants, bees, and other individual organisms behave, and their collective intelligence does something more complex, like building a nest or a hive. This has an equivalent field in artificial intelligence, which started more than two decades ago as ant colony optimization, but has since spread to swarm intelligence (which has a mathematical basis in reinforcement learning).
image description
The above are just three of these paths, there are of course many more, for example any one path can be combined with the others, and there may be abstract intelligence to move to higher levels.
in conclusion
in conclusion
This post has a follow-up (Part II) that explores the implications of AI DAOs, including dangerous DAOs. Wild, Wooly AI DAO. Also, part three is here.
appendix
This post is not advocating that AI replace DAO hacks or governance issues, AI DAOs will have a harder time succeeding.
thank you
thank you
