Vertigo3d | E+ | Getty Images
DAVOS, Switzerland – Blockchain could be used to prevent bias in the data on which artificial intelligence models are trained – and that could be a “killer use case” for the technology, executives told CNBC.
One of the concerns about AI models – the kind that underpin applications like ChatGPT – is that the data they are trained on could contain bias or misinformation. This means that the answers an AI system could give would contain these biases and false information.
Blockchain came onto the market in 2009 with the introduction of the cryptocurrency Bitcoin. In the context of Bitcoin, the technology is an immutable and tamper-proof public ledger of transactions. Companies have tried to use these principles in other blockchain applications, sometimes referred to as distributed ledger technology.
In the case of AI, training data can be stored on the blockchain. This allows the developers of the AI system to keep track of the data on which the model was trained.
Casper Labs, a business-focused blockchain company, partnered with IBM this month to develop such a system.
“The product we're building actually checkpoints the records and stores them on the blockchain so you have proof of how the AI is being trained,” Medha Parlika, chief technology officer and co-founder of Casper Labs, told CNBC during a panel discussion at the World Economic Forum in Davos this week.
“So if you're using the AI when it's learning and you notice that the AI starts to hallucinate, you can actually reset the AI. This allows you to undo some of the learning and revert to an earlier version of the AI.”
Hallucinations are generally understood to occur when an AI system outputs false information.
Blockchain is a technology that has been talked about for many years, and a variety of industries, from finance to healthcare, are looking for ways to use it.
However, Sheila Warren, CEO of the Crypto Council for Innovation, said a blockchain-based AI training ledger could be the “killer use case” for the technology.
“I actually think that the verification of an AI and some of the checks and balances… within an AI system will be blockchain-driven and blockchain-supported,” Warren told CNBC during the panel.