Advanced Computing in the Age of AI | Thursday, March 28, 2024

Algorithmia Adopts Blockchain for ‘AI Economy’ 

Rawpixel.com/Shutterstock

Algorithmia, the AI marketplace, has come up with a way of buying, selling and executing machine learning models on a blockchain platform, ushering in what it says is an “AI economy.”

Separately, the Seattle-based platform vendor said Tuesday (Feb. 27) it is launching an AI and machine learning contract for developer bounties via a neural network running on the Ethereum blockchain. The contract is similar to earlier bounties used to improve the performance of existing algorithms. Algorithmia’s version would offer a reward of 3 ETH.

Pronounced “Ether,” the tradeable cryptocurrency also can be used by developers to pay for fees and services on the Ethereum network.

Company researchers released a paper this week detailing how machine learning models can be traded via blockchain, an advance the four-year-old startup claims could create a “path for machines to trade algorithms.” Details on the blockchain-based protocol for purchasing ML models called “DanKu” are available here.

Algorithmia claims its protocol is the first to run a neural network on a blockchain. “This means that the blockchain doesn’t function only as a public ledger; it can also run the most advanced software in a completely decentralized and trustless manner,” explained Besir Kurtulmus, a machine learning researcher at Algorithmia. (“Trustless” refers to blockchain’s ability to monitor transactions—debits and credits, for example—eliminating the need for a trusted third party.)

In seeking to add value to the current Ethereum blockchain, the company said the DanKu protocol would provide another layer of machine intelligence to applications. The underlying models will require training and testing, added Diego Oppenheimer, Algorithmia’s founder and CEO. The trustless protocol would allow users to “securely acquire new models to fit their business needs [while creating] an opportunity for machine learning engineers to efficiently build and deploy models,” Oppenheimer added.

The protocol works like this: a developer or a machine can request an AI or machine learning model to perform a specific task; the value of the resulting contract is held in cryptographic “escrow” until terms of the contract are fulfilled. A sample can then be downloaded to train models that are submitted back to the blockchain. At that point, the models are evaluated on the blockchain.

Source: Algorithmia

Once a suitable model is ready, a bounty reward is paid.

The startup’s bounty approach has been used by others to improve algorithms, most notably Google’s (NASDAQ: GOOGL)

$1 million bounty for improving a movie recommendation algorithm. Algorithmia’s variation is using blockchain to execute algorithms while offering incentives to improve them. The company said it plans to release future contracts for progressively more difficult problems.

The blockchain initiative builds on the application builder’s recent efforts to accelerate the development and deployment of machine learning models. Algorithmia’s “AI Layer” unveiled in November is designed to fill the gap in the current machine learning infrastructure that slows the deployment of models in production—what the company refers to as the “last mile” problem.

In its efforts to push machine-learning algorithms out of the lab and into production, the startup claims its portfolio of machine learning models is the largest AI marketplace of its kind.

Among the possible applications of the DanKu contracts running on blockchain are what Algorithmia calls “self-improving AI systems.” The approach would “enable AI agents to participate autonomously on both sides of this new market,” the company said in an email. For example, AI applications encountering new data sources could use the framework “to solicit new models that help it to understand.”

The company also foresees use cases where AI bots could search for contracts and use their own resources to train models for the marketplace.

“Blockchains are perfect for this kind of AI-to-AI economy,” the company asserted.

 

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).

EnterpriseAI