Blockchain: A Potential Solution to AI’s Trust Issue?
Recent trends show a growing concern about the reliability and impartiality of artificial intelligence (AI) algorithms. This has led some organizations to consider an unexpected ally: blockchain technology. Once primarily associated with digital currencies like bitcoin, blockchain’s ability to create a secure, shared digital ledger is now seen as a potential tool for enhancing transparency and accountability.
Blockchain’s core function is to maintain a tamper-proof record of transactions across a network, secured through cryptography. This ensures that every addition to the ledger, such as data used for training AI algorithms, is both transparent and immutable.
Companies like FICO and Casper Labs are exploring the use of blockchain to track AI algorithm development and training processes. This approach is increasingly relevant as AI becomes more prevalent in various sectors, and as regulatory bodies demand greater transparency and auditability of these algorithms.
Despite its potential, blockchain technology has yet to achieve widespread business adoption, even in applications like supply-chain tracking. However, the rapid expansion of AI applications could provide blockchain with a significant boost.
Scott Zoldi, FICO’s Chief Analytics Officer, explains that blockchain could facilitate the creation of detailed, trustworthy records of the data used in AI training, including who handled the data and the steps taken to validate it. While it won’t prevent biases or errors in algorithms, blockchain can offer a clear audit trail to understand these issues better.
On the other hand, blockchain doesn’t directly address AI’s “black-box” problem – the challenge of explaining why an AI model produces specific outcomes. What it does promise, according to Zoldi, is a more robust framework for investigating these outcomes.
Some industry leaders remain skeptical. For instance, Scott duFour, CIO of Fleetcor Technologies, views blockchain as potentially beneficial but not a standalone solution for AI governance. He emphasizes the need to integrate blockchain with existing tools that aid in interpreting AI model predictions.
FICO is currently using its blockchain-based tool internally, with plans to release it to customers later in the year. Meanwhile, Casper Labs, in collaboration with IBM, is developing a tool that offers “version control” for AI models. This tool, expected to integrate with IBM’s watsonx AI governance platform, will track data and parameters influencing models, allowing companies to revert to earlier versions if biases or inaccuracies are detected.
Blockchain adoption in AI governance faces challenges. Previous blockchain applications, like supply-chain tracking, struggled due to the existence of established, effective non-blockchain methods. However, as Mrinal Manohar, CEO of Casper Labs, points out, no such established methods currently exist for AI governance, positioning blockchain for potential success in this domain.
Despite the optimism, widespread adoption is not guaranteed. Avivah Litan, a Gartner analyst, believes the concept is ahead of its time, given the current market’s prioritization of governance and risk management.
As AI continues to evolve, the intersection of blockchain and AI could prove critical. Nicolás Ávila, CTO for North America at Globant, anticipates that both technologies will eventually address each other’s limitations, suggesting a symbiotic future where blockchain enhances AI’s trust issue and AI, in turn, resolves blockchain’s challenges.
