With all the fuss about AI agents allegedly creating their own memecoins or managing crypto portfolios, I’d like to propose an idea: a “trAIding competition.”
TL;DR
Different participants train AI models on financial data from a DEX and deploy them on the Internet Computer. The goal of these models is to maximize the funds they are allocated at the beginning of the competition within a given timeframe, e.g., three months. All models receive the same live trading data from a DEX deployed on the IC and act on this information by making trades that ideally increase their net worth. At the end of the period, each model’s performance – defined as net worth – is compared, and the model with the highest balance wins. This could be combined with a prediction market, where spectators can predict which model will win the competition. The deployer of the winning model would receive the remaining balances of the other participants as a reward and potentially a share of the volume facilitated by the prediction market.
Why would anyone do this?
It’s a very cool showcase of what’s possible on the IC.
It incentivizes people to create models that can be deployed on the IC and optimize them for peak performance, potentially driving the entire space forward.
This can’t be replicated on centralized infrastructure:
You couldn’t verify that the model isn’t updated during the competition.
You couldn’t confirm the model you claim is running is actually the one in use.
You couldn’t even be sure that any model is running at all.
Key materials signing transactions would need centralized storage, which could allow the model deployer to “rug pull” (imagine AI agents managing users’ funds).
You wouldn’t be able to settle the competition in a trustless, decentralized manner.
Do you think this makes sense? If so, who’s building this?
just for me to understand correctly what you propose. can you confirm that my understanding is correct?
anybody can train AI models on data from a (specific?!) DEX in a specific time frame before the actual competition starts
I am wondering if we need to define which DEX will be used for the competition if it’s targeting a specific DEX, I think the one with the best liquidity and highest trading volume should be used
the AI model will be deployed and not updated during the whole competition period
the competition period starts and a canister with some (pre?)-developed (shared?) logic will perform automated trades on-behalf of each deployed AI model
during the competition, the community could participate in a small dedicated prediction market and bet on models
the deployer of the winning AI model will win whatever we define
I like the idea! I have thought about automated trading bots quite a while already, but making such a competition in that area is even cooler
The training data is not defined, you can train your model on whatever data you like. But I think the DEX used for the competition should be defined to have the same environment for all models.
Yes.
The functionality to actually perform the trades should be contained in the AI canister submitted to the competition I think.
but that logic would be shared, right? or at least every participant would use the same logic to process the input data according to their individual model and then perform some common trading action (or not action at all).
if it’s targeting a specific DEX, I think the one with the best liquidity and highest trading volume should be used
