Executives told CNBC that preventing bias in the data used to train artificial intelligence models might be a “killer use case” for blockchain technology.
A worry with artificial intelligence (AI) models, such as those used in ChatGPT, is that the data they are trained on may contain biases or false information. This implies that an AI system’s responses could include these prejudices and inaccurate data.
The introduction of the cryptocurrency bitcoin in 2009 marked the arrival of blockchain technology. The technology behind bitcoin is a public ledger of transactions that is unchangeable and impenetrable to tampering. Companies have been attempting to apply these ideas to other blockchain applications, which are commonly referred to as distributed ledger technology.
Training data for AI can be stored on the blockchain. This will enable the AI system’s developers to monitor the data that the model has been trained on.
This month, IBM and Casper Labs, a blockchain company with an emphasis on business, collaborated to develop such a system.
During a panel discussion at the World Economic Forum in Davos this week, Medha Parlika, chief technology officer and co-founder of Casper Labs, informed CNBC that “the product that we are developing, the datasets are actually checkpointed and stored on the blockchain so you have a proof of how the AI is trained.”
“And so as you use the AI, if it’s learning and you find that the AI is starting to hallucinate, you can actually roll back the AI. And so you can undo some of the learning and go back to a previous version of the AI.”
In general, erroneous information provided by an AI system is referred to as a hallucination.
Blockchain technology has been discussed for a long time, and many different industries, from finance to healthcare, have been considering how to apply it.
But according to Sheila Warren, CEO of the Crypto Council for Innovation, a blockchain-based AI training data ledger might be the technology’s “killer use case.”
During the discussion, Warren told CNBC, “I actually do think that the verification of an AI and sort of the checks and balances… within an AI system, are going to be blockchain driven and blockchain backed.”