Source: Deep Learning on Medium
Numerai is widely recognized as one of the most innovative hedge funds of the last few years sitting at the intersection of quantitive finance, artificial intelligence(AI) and cryptocurrencies. The fund hosts competitions among data scientists, AI experts, mathematicians, etc that create algorithms based on a specific thesis. The magic behind Numerai is a blockchain-based protocol that allow data scientists to create, submit and stake prediction theses in a completely decentralized way. Recently, Numerai decided to extend their platform to the rest of the financial community by unveiling Erasure, a decentralized prediction marketplace.
In general terms, Erasure is a decentralized protocol and data marketplace for financial predictions. The protocol provides the mechanics for data scientists to submit predictions based on a specific investment thesis and profit from its performance. Blockchain technologies and crypto-economic incentives ensure trustless honesty between hedge funds (buyers) and data scientists(sellers) in Erasure. Beyond some known successes from AI powerhouse SingularityNet, Erasure can be effectively considered one of the first practical applications of decentralized AI. The protocol removes many of the complexities of general-purpose decentralized AI platforms by enabling a very tangible and pragmatic scenario.
The Challenges with Decentralized AI
I’ve been avidly writing and speaking about decentralized AI for the last year. Despite being a strong believer in the future of decentralized AI, I will the first one to acknowledge that this new computing paradigm has seen very limited uptake within the AI community. In general, there are three major challenges that can be associated with the current generation of decentralized AI technologies.
· Computational: Decentralized runtimes are not optimized for storing large volumes of data or performing computations over large GPU clusters.
· Cultural: Without implicit trust, how can users assert the validity of predictions even if they perform well against historical data?
· Financial: For a decentralized AI model to work, it has to have the right financial incentives for users and data scientists to engage in the creation and operation of AI models.
A subtler challenge of the current generation of decentralized AI technologies is related to the fact that most of them are focused on creating general purpose platforms which is a very complex endeavor. While decentralized AI still struggles as a general-purpose AI paradigm, there are very specific scenarios that are ideal for a decentralized AI protocol. One of those scenarios happens to be financial market predictions.
The goal of Erasure is to provide a decentralized marketplace in which data scientists can upload predictions based on available data, stake those predictions using crypto tokens and earn rewards based on the performance of the prediction. While the first use cases are related to financial forecasts from Numerai, Erasure can be used for any time of prediction. Architecturally, Erasure combines several components that provide a foundation of decentralized interactions between buyers and sellers in a decentralized marketplace:
· Proof-Of-Existence Protocol: Erasure borrows some ideas from proof-of-existence protocols to keep immutable, timestamped records of all predictions submitted by a data scientist. Erasure uploads every prediction to IPFS, a highly scalable decentralized storage network, and records the hash of the IPFS address in the Ethereum blockchain.
· Prediction Protection Schema: Erasure keeps the data about recent predictions protected so that only the buyers can have access to it. The predictions are revealed to the rest of the network after an appropriate time elapses.
· Native Currency: Erasure uses the Numeraire token (NMR) as a native cryptocurrency for all the interactions between buyers and data scientists.
· Staking Mechanism: Erasure allows data scientists to stake NMR tokens as a way to express the confidence in their predictions. By staking their own tokens, data scientists are indirectly invested in the outcome of the prediction.
· A Discretionary Recourse Protocol: In addition to the staking mechanism, Erasure introduces a “griefing factor” that allow buyers to destroy part of the money paid for a prediction in case this one doesn’t perform accurately. For instance, a griefing factor of “1:10” means that for every $1 a buyer destroys of their own money, $10 of the seller’s stake will get destroyed.
The aforementioned components provide the basics of a very clever architecture that leverages blockchain and crypto-economics to provide the basics of a decentralized AI model. Numerai is certainly the most successful application implemented on Erasure but hopefully won’t be the only one. By opening the Erasure protocol, Numerai is taking the first steps towards enabling one of the first practical decentralized AI applications.