The Ethereum blockchain is being used by engineers at Microsoft to “democratize” AI and machine learning, making these typically centralized and costly systems more accessible to everyone.
Smart contracts have the potential to profoundly impact how data is managed.
They are immutable and allow devs to reward users for participation and contributions. And, for a blockchain like Ethereum, thousands of decentralized nodes all over the world ensure that the code is always available, practically eliminating issues with downtime.
Smart contracts and AI
Justin Harris, a senior software development engineer at Microsoft, saw that Ethereum smart contracts have the potential to fundamentally change how machine learning and AI models are designed.
Access to well-designed machine learning algorithms can be problematic, according to Harris. These algorithms tend to be centralized, sold on a per-query basis, and trained using proprietary and expensive data, he wrote.
He envisions a new paradigm.
“One in which people will be able to easily and cost-effectively run machine learning models with tech they already have, such as browsers and applications on their phones and other devices.”
In the “spirit of democratizing AI,” Harris introduced a new open-source initiative from Microsoft: Decentralized & Collaborative AI on Blockchain.
The new paradigm:
In the new paradigm, Harris envision incentives and rewards for people interacting with—and improving—these machine learning algorithms. These algorithms would be free to use for evaluating predictions, which is ideal for things like building personal assistants or making systems that produce user recommendations.
Some potential ways to crowdsource such a system, Harris suggests, include game-ification—like what is seen on Reddit. Or, building in prediction markets similar to what Augur is pioneering. Or, even creating “self assessment,” where users pay a deposit and those who make good contributions are rewarded at the expense of bad contributions.
As a test, the Microsoft developers successfully set up a model which could classify the sentiment of a movie review (either as positive or negative). At the time, updating the model only cost $0.25 on the Ethereum blockchain in fees.
Overcoming blockchain’s limitations
That isn’t to say blockchain doesn’t have limitations. Transaction speeds (throughput) is a major bottleneck. Machine learning and artificial intelligence are notoriously hungry for computation.
Developers need to work around this by using models which are efficient to train, or by doing the hard data-crunching off-chain; they could even be integrated using oracles through a service like Chainlink.