Numerai — Does everyone win?
BlockFellows Cryptoeconomics research on game theory and mechanism design
Bitcoin succeeded where other decentralized protocols failed, not because of Proof-of-Work, the idea of decentralized cash, or even fault-tolerant consensus, but because it incorporated cryptoeconomics at the core of its consensus protocol.
– BlockChannel
Numerai
Numerai is an artificial intelligence hedge fund that uses Numeraire (NMR), a token on Ethereum, to incentivize data scientists to improve the hedge fund through the improvement of its artificial intelligence.
A hedge fund is as “an investment fund that pools capital from accredited individuals or institutional investors and invests in a variety of assets, often with complex portfolio-construction and risk-management techniques”. The success of a hedge fund lies in comprehending market inefficiencies, exploiting them and get profit at the expense of others. This pretty much simplifies how Wall Street works. Simplistically, Wall Street is a zero-sum game: someone wins with someone’s loss. In most of the cases, money isn’t lost or made, it’s just transfered from one hand to another. Having this in mind, Richard Craib, the creator of Numerai, believes he can turn Wall Street competition into collaboration.
This fund’s trades are chosen by proprietary artificial intelligence algorithms and encrypted public datasets. This is achieved by crowdsourced quantitative models from more than twelve thousands of anonymous data scientists. The data scientists are rewarded with cryptocurrencies (NMR tokens or Bitcoin in the past) for their effort through a collaborative reward mechanism.
Machine Learning
Machine Learning (ML) models are widely used in prediction markets. It allows educated decisions based on statistic techniques and to “learn” progressively with data. Accuracy is a metric commonly used to evaluate these models — it basically means, the fraction of right guesses one model got right.
Overfitting means good performance on trained datasets, but poor performance on live and not tested datasets. Numerai tries to evaluate the models based on how well they behave on new sets of data, improving the overall quality of the artificial intelligent hedge fund.
Prisoner’s Dilemma
Game theory studies strategic interaction between individuals in situations called games. Situations in which the total gains and losses are greater or less than zero are called positive-sum or negative-sum games, accordingly. In contrast, if the total gains and losses add up to zero, it is called a zero-sum game. Positive and negative-sum games can also be referred as non-zero-sum games.
One cannot talk about non-zero-sum games without talking about the Prisoner’s dilemma. This problem consists on having two prisoners (Player 1 and Player 2), getting interrogated separately. Both prisioner’s can decide whether they confess (betraying the other) or stay silent (co-operating).
The rules are simple:
- If Player 1 and Player 2 confess, they both get 6 years in prison.
- If just one of them confesses, he will stay off prison and the other one will get 10 years in prison.
- If they co-operate by not confessing, they will get one year in prison.
As you can see, the better outcome for this game is if both of them do not confess, by co-operating.
From zero-sum to collaboration
The tech companies benefit from network effects where people behave differently because they are trying to build a network, rather than trying to compete.
– Richard Craib
Numerai’s mission is to make several thousands of data scientists from all over the world to build learning models and present market predictions that are likely to be successful to improve the AI agent continuously over time.
Its first version was flawed, according to Craib. Data scientists were getting rewarded if they had better results than the others, resulting in competition. Similar to Wall Street, more or less like a zero-sum game. “They were fighting each other rather than fighting for the same goal”, said Craib to Wired.
In 2017, they have decided to turn Numerai into a more collaborative environment by building Numeraire, a token on top of Ethereum, and the cryptoeconomic model to incentivize the collaboration among data scientists, in order to create better models of prediction, therefore fighting for the same goal with different strategies.
How does it actually work?
Similar to Prisoner’s Dilemma described above, NMR gets destroyed if someone fails or “goes to prison”, so it is the best interest of all participants to avoid this. By not collaborating with each other, participants are likely to lose money.
Data scientists stake NMR along with their prediction on how well their models behave on usually one month window. Once the predictive model is live, data scientists get rewarded based on their confidence along the actual performance. If their predictions are poor, the staked tokens will get burned. The participants have incentive to invite participants that have the potential to improve the network, exposing the nature of a collaborative mechanism. Staking, rewards and punishments are mechanisms that push towards the development of a more collaborative environment.
So, does everyone win?
I believe it’s hard to answer this question. I didn’t find any conclusive answer, but the proposed mechanism is interesting. I usually say that digital transformation is not about changing the game, but how players play it. Numerai does propose an interesting collaborative mechanism at an extremely competitive market. If you’re interested, take a look at their whitepaper.
