Algorithmia today launched a very interesting competition on Ethereum platform - an AI competition. The task is to design the best predictor for voters preference in the last US presidential election classified by longitude and latitude. The bounty for the creator of such a model is 3 ETH (~ $2,500 at the time of publication). The task has a form of a smart contract of a special type called DanKu - Algorithmia's own creation.
DanKu allows a task creator to code a smart contract with a problem that deals with machine learning, publish a hidden dataset made available once the model creator wishes to evaluate the performance of the designed model. Then, through the smart contract, the task creator establishes randomized training sets made publicly available. All of the computation to test the model are performed on Ethereum blockchain. Because it is a smart contract, both parties don't need to trust each other: once the conditions of the problem are satisfied, the task creator gets the designed model, and the model creator gets his reward.
Apart from the novelty of such use case, it is also a proof-of-concept of a unique service useful for the industry. Say, a company needs a machine learning solution to its problem, but doesn't have the resources to hire a data scientist. Then, with Algorithmia and its DanKu smart contract that company can create a smart contract and request a specific machine learning approach to solve the problem. This is the first time neural network inference is run on top of Ethereum.
If a machine knew how to define a contract automatically, and put it on some system that other systems could solve, there is a potential for machine learning systems solving problems autonomously at a certain level.
Co-founder and CEO, Algorithmia