Photo from SonTech Research Team.
Companies are increasingly turning to blockchain, a technology once hailed as transformative but lacking business adoption, as a solution to address the challenges of ensuring the safety, fairness, and precision of artificial intelligence algorithms.
Blockchain, famous for its role in supporting cryptocurrencies like bitcoin, is a technology that enables the creation of a digital record of transactions, which can be shared across a network of computers. It employs cryptographic methods to permit each participant within the network to contribute to this record securely, all without the requirement of a central governing body. Once a transaction occurs, the blockchain maintains an unchangeable and permanent record of it.
In light of this, technology companies FICO, specializing in data analytics, and the startup Casper Labs, which focuses on blockchain, have announced their intention to utilize blockchain technology for monitoring the development and training of AI algorithms.
They emphasize that this application is both timely and critical, especially with the widespread adoption of AI in various industries, where companies are encountering challenges in placing complete trust in AI's results. Simultaneously, government regulators are intensifying their demands for companies to enhance the transparency and auditability of their algorithms.
Earlier business applications of blockchain, such as supply-chain monitoring, did not achieve significant market success. Whether the widespread adoption of AI can provide the boost that blockchain needs remains uncertain.
Scott Zoldi, the Chief Analytics Officer at FICO, mentioned that blockchain has the capability to monitor the precise details of an algorithm's training data, including who conducted the training, when it occurred, and what measures were employed to assess and validate the data. While companies engaged in building AI models typically attempt to trace this data trail, using blockchain would streamline the process, resulting in a shared, uniform, and reliable record. Blockchain cannot prevent algorithms from malfunctioning or displaying bias, but it can provide a transparent and traceable record that could reveal the reasons behind such issues, according to him.
Zoldi explained that blockchain can divide processes into smaller, manageable components, effectively functioning as mini-contracts. This allows for rigorous validation of each step at a very detailed level. The immutability of blockchain brings a degree of transparency and integrity to the process.
Blockchain does not provide a solution to the explainability or "black-box" issue in AI models. By itself, it doesn't uncover the reasons behind why an AI model generated a specific output. However, it does offer the potential for improved records that can assist in addressing that question, as stated by Zoldi.
Scott duFour, the Chief Information Officer of Fleetcor Technologies, expressed skepticism about using blockchain, likening it to a tool searching for a problem. While he acknowledges that blockchain could enhance trust in AI systems, he believes it should be used in conjunction with existing tools that assist designers in comprehending and interpreting predictions made by AI models.
Currently, FICO's tool is in use within the organization, with intentions to make it available to customers later this year.
Casper Labs, based in Switzerland, is collaborating with IBM to create a tool of its own. This tool will provide "version control," documenting the data and parameters that impact a specific model at a particular moment. If companies identify bias or inaccuracies in their models, they have the option to return to a previous version, according to Chief Executive Mrinal Manohar.
Addressing bias in algorithms remains a challenging and time-intensive endeavor, often hindered by the absence of suitable systems and tools within companies.
Casper's tool is currently undergoing beta testing and is anticipated to be available on the market in the third quarter of this year. It will be integrated with IBM's WatsonX AI governance platform, but Mrinal Manohar mentioned that it won't be limited solely to the WatsonX platform. AI governance platforms aim to provide guidance for the utilization of technology and to oversee and mitigate the associated risks.
Manohar pointed out that the application of blockchain in this context is less susceptible to the challenges experienced in earlier uses. For instance, in supply-chain tracking, companies had well-established methods for monitoring their products, making the transition to an expensive and intricate blockchain system a tough proposition. However, in the realm of AI governance, there are no pre-existing, well-established methods. This presents a unique opportunity for blockchain to potentially become an industry standard, according to Manohar.
Supply-chain initiatives faced challenges because they involved numerous vendors and stakeholders with varying levels of technological expertise required for tracking. However, in this context, fewer participants are required, as noted by Avivah Litan, Vice President and Distinguished Analyst at the research and consulting firm Gartner.
Nevertheless, this does not imply that the adoption of blockchain for AI governance is poised for rapid growth.
"It's a promising concept, but in my opinion, it's ahead of what the market currently demands," commented Litan. She pointed out that numerous companies have not traditionally given high priority to AI governance and risk management as they navigate their way through new projects. She rated the importance of AI governance for clients at around three on a scale of one to ten. However, she also mentioned that this importance is gradually increasing as the number of generative AI projects grows.
Manohar disagreed with the idea that the tool is ahead of the market and mentioned that corporate tech leaders who have tested the tool have provided positive feedback.
He argued that the reason companies are not currently giving priority to AI governance and risk management is not due to a lack of concern or belief in its future importance, but rather because many are uncertain about how to initiate these efforts.
Nicolás Ávila, the Chief Technology Officer for North America at software company Globant, expressed the need for these tools to demonstrate their effectiveness. Nevertheless, he does recognize significant potential in the combination of both technologies. In the end, he believes that AI will address the challenges of blockchain, and blockchain will address the challenges of AI.