Proposal: Liquid Prediction Markets for AI Forecasting

Jesse Richardson

Background

Polymarket is a prediction market platform where people can trade contracts representing the probability of events occurring (e.g., "Who will win the 2024 Presidential election").
Trading volume on Polymarket is heavily influenced by liquidity rewards, a program which pays traders to place competitive bids and offers in various markets, thereby increasing liquidity and encouraging more trades. Polymarket currently spends an average of $17K/day on liquidity rewards, and can range from $1/day to as much as $5K/day on various individuals.
Polymarket pays for the liquidity rewards itself (thanks, VC money!), and chooses how high to set the rewards on each market.

The problem

Despite significant interest in AI forecasting, there is a lack of liquid high-quality prediction markets on the future of AI.
Polymarket has ~25 markets on AI, but most of them aren't very interesting (e.g. predicting the Chatbot Arena leaderboard) and they all have pretty low liquidity, which is downstream of low liquidity rewards. For example, I'm pretty interested in "Will AI win an IMO gold medal in 2025?", but the rewards are set at a relatively low $5/day and the order book is pretty thin as a result.
Other prediction markets sites such as Kalshi suffer from similar pitfalls.

The proposal

I've been speaking a bit with a member of the Polymarket team and they are interested in having someone fund higher liquidity rewards on AI markets -- these could be existing markets or new markets that are specifically requested.
There's not yet a formal process for this to happen, and it's not guaranteed Polymarket will be able to make it work, but it seems pretty likely.
As a result, I am looking to partner with individuals or organizations who want to help make high-liquidity prediction markets a reality for AI forecasting.
I am primarily interested in hearing from potential funders, but also from people in AI forecasting who want to come up with new markets that should receive liquidity rewards funding, and help write rules for those markets.
To begin with, I'm hoping to raise $55K, which could fund ten markets for six months with liquidity rewards of $30/day.

Will this result in better AI forecasts?

Boosting liquidity rewards on AI prediction markets will definitely increase the trading volume on these markets, of that there is no doubt. The question is a) how much, and b) will that translate to more accurate predictions.
My best guess, based on my experience with liquidity rewards, is that going up to around $30/day will probably mean a significant increase in liquidity, and there's some diminishing returns as you go higher.
There's some additional stuff you can do like starting the rewards higher (maybe $100/day) for a few days to get people's attention, and then lowering them down to a stable level.
One of the major reasons liquidity rewards sometimes doesn't translate into higher volume is high volatility, such that liquidity providers are likely to get run over and lose money. This is not likely to be the case for many of the AI forecasts we're interested in, where the probability of something important happening in a year's time doesn't generally change that much hour-to-hour or day-to-day.
If it's not clear why more liquidity would make a market more accurate, the basic mechanism is that it attracts more participants and in particular the most profitable Polymarket traders, who don't have spare time or attention for illiquid markets. If I see an illiquid market that looks like it's trading at a weird price but I'm not super sure, I'm probably not going to bother to research it more and try to make some trades, I'm just going to do something else. But if I see a reasonably liquid market and have a sense that the price is wrong, it's much more worthwhile for me to spend time looking into it, and I'm much likelier to end up doing some significant volume in that market.
Overall I don't expect there to be a lot of upfront costs associated with this proposal since the liquidity rewards are paid for daily, so it should be easy and low-risk to try it out as a pilot and cut it off if it's not working.

Conflict of Interest

I am not affiliated with Polymarket the company, but I am an active trader on Polymarket and I would stand to benefit financially from increased liquidity rewards, as it's a net flow of money into the site. If potential funders are worried about this conflict of interest, I am open to abstaining from trading on the relevant markets, or contributing to the project out of my personal finances.

Summary

In conclusion, I think we should have liquid prediction markets for forecasting AI, I think Polymarket is the most natural place to host these markets right now, they are interested in facilitating people making these markets more liquid (and create new, better AI markets) and I am looking for people to help fund this project, or to help write questions. Please get in touch if you are interested.

[-]the gears to ascension2mo20

Do prediction markets make the things they predict go better? Worse? Add variance? Reduce variance? How do we know?

For that matter, does more people being good at prediction make things go better overall? If so, under what circumstances? I expect there's a wide variety of circumstances where it helps. what are they?

What hypotheses do we have for what-has-to-go-right-at-once in order that making-a-person-more-able-to-win-at-prediction-markets to result in the actual increase of actual probability of good long-term trajectories for the things humans and their friends cared about, such as there continuing to be more humans, and a cosmopolitan moral outcome?

The path from "make this prediction system more powerful" to "and then things go well" needs very strong heuristic arguments I haven't seen in discussions on prediction markets - I currently think any argument that improved prediction implies improved outcomes needs to show why the improved prediction produces an asymmetric effect towards integrating the longest-term preferences of the humans involved, rather than, eg, being an asymmetric short-term boost for whichever have the most short-term need for information about their moves on the short-term game board without regard to whether their short-term play even optimizes for their own long term outcomes.

25