We are excited to introduce Gnosis AI, a division of Gnosis Labs working at the intersection of AI and web3. Currently, Gnosis AI is focused on agent-to-agent payments and prediction markets. We believe that AI offers Gnosis Chain and the Gnosis 3.0 roadmap incredible opportunities. Our vision is to make Gnosis the Schelling point for AI agents in web3.
What We Do
Our team is developing AI agents designed to thrive in the Gnosis ecosystem. The first use-case we are focusing on is prediction markets [1], an application where Gnosis has gained considerable experience since its ICO in 2017.
While prediction markets only have a fraction of liquidity from other DeFi categories like DEXes, they offer a number of downstream uses. These include forecasting, risk management and most interestingly decision making. This functionality is akin to acting as a “truth oracle”.
DeFiLlama Prediction Market TVL Rankings
The idea is that if prediction markets are well incentivized (having substantial liquidity) and efficient (with many rational market participants), they can accurately forecast event probabilities. For instance, if someone asks, "Will Ethereum reach $10,000 in 2024?" participants would bet on this market, driven by the available liquidity. Because the market is efficient, the event probability, indicated by the relative price of outcome tokens, would closely reflect the true likelihood of the event. This is the promise of prediction markets: they harness the wisdom of crowds and the price mechanism to arrive at highly accurate probabilities [2].
To realize this idea, the crucial missing ingredient is “rational market participants.” We believe prediction markets provide fertile ground for AI agents to be active [3], as they offer tasks to be completed (questions to be answered) and incentives (liquidity behind each question). Therefore, this is the use case we are exploring.
It’s also worth noting that if the “truth teller” concept becomes a reality, more users, especially humans, would be drawn to prediction markets on Gnosis due to their forecasting abilities. This would create a flywheel effect, with more liquidity and more agents competing for it. This is the goal of our work.
What we have been doing until now
Trader agent
We have developed trader agents with different specializations to participate in prediction markets. The basic idea is that they all have to answer “yes” or “no” to a given question (examples are available in the table below).
Each agent employs different strategies for answering the question, which is reflected in their performance (displayed in the chart below). Notably, the accuracy of the top performing agents is around 65% (more details in the Dune dashboard). Additionally, agents from our partners at Autonolas (https://registry.olas.network/gnosis/services) and Agentcoin (https://www.agentcoin.org/) are also tracked. Agents with different strategies can be found in our agents repository.
Social agent
We also have a unique agentcalled the “Social agent”, which acts as a Public Relations representative for one of the agents (given by address 0xC918c15b87746E6351E5f0646DdCAaca11aF8568). Feel free to follow it on Twitter and Farcaster.
The Social Agent’s primary role is to analyze recent bets made by its “client” (the trading agent it represents) and draw attention to the respective market. It accomplishes this by posting a tweet highlighting the rationale behind the bet, employing a more analytical tone. Additionally, the Social Agent serves as an "external memory" for the agent, providing insights into its past actions and decision-making processes.
Our Future Plans
We plan to regularly release updates on our work and achievements, both on the Gnosis blog and across other social channels. While our primary focus remains on prediction markets, we aim to enhance the autonomy of our agents. This includes enabling them to reason about profitability and execute actions aligned with goals such as earning money, learning and adapting, or pursuing other creative objectives. A primary goal is to enhance the accuracy of our trading agents to realize the "truth teller" vision outlined earlier in this article.
How to Get Involved
We welcome external contributors! All our work is open-source, so please feel free to check out our repositories below by opening or addressing issues.
→ Prediction market agent tooling (PMAT)
https://github.com/gnosis/prediction-market-agent-tooling
→ Prediction market agent (PMA)
https://github.com/gnosis/prediction-market-agent
In any case, feel free to reach out via Discord, we are on the channel #gnosis-ai
References
[1] https://worksinprogress.co/issue/why-prediction-markets-arent-popular/
[2] Vitalik’s post on Prediction Markets - https://vitalik.eth.limo/general/2021/02/18/election.html
[3] Vitalik’s post on web3 x AI - https://vitalik.eth.limo/general/2024/01/30/cryptoai.html