FICO’s Blockchain Usage — Now Patented — Boosted Firm’s Accountability
Blockchain tech “ideal” to offer immutable record of the model development standard associated with its analytic model, company exec says
Immersion Imagery/Shutterstock modified by Blockworks
The analytics company behind one of the standard measures of consumer credit risk in the US has used blockchain technology in recent years to improve its accountability and compliance when developing models.
Now, the firm’s blockchain use for data and model governance is patented.
Founded in 1956, Montana-based FICO — formerly known as Fair, Isaac and Company — uses predictive analytics and data science to improve operational decisions. It holds more than 200 US and foreign patents.
The company is perhaps best known for its FICO Score, which reflects how consumers repay credit obligations. These scores are used by lenders, who use FICO Scores, among other factors, to assess credit risk of potential borrowers.
Scott Zoldi, the company’s chief analytics officer, told Blockworks in an interview that blockchain’s ability to enforce contracts — along with its immutability — makes it a natural tool for artificial intelligence (AI) and analytic governance and audit.
One of its latest patents covers the concept of a shared ledger that would “track end-to-end provenance of the development, operationalization and monitoring of machine learning models in an immutable manner,” the company said in a statement earlier this month.
FICO first applied for the patent in 2018 — believing the technology was ideal to offer an immutable record of the development standard for each of its analytic models, according to Zoldi.
The company’s standard defines data requirements, model architecture, allowable variables and ethics testing, for example, so that data scientists can more easily comply.
Keep reading for more excerpts from Blockworks’ interview with Zoldi.
Blockworks: When did FICO start using blockchain technology?
Zoldi: We started blockchain proofs of concept on select projects in 2018 and productionalized the capability in 2019. Since then, we have continued to enhance the user interface to optimize the experience for data scientists, the management team and approvers.
Blockworks: What exactly will the patent cover, and what is FICO planning to do with it?
Zoldi: The patented blockchain technology is used in all AI model development projects in my data science organization.
In this invention, each model requirement and development step is codified and added to an immutable blockchain. Specific scientists’ confirmations that the model meets those requirements, along with model testers and approvers, are added to the blockchain.
The blockchain provides auditability months and years after the model build, providing fine detail as to how the model was developed, what it’s sensitive to, and under what conditions the model should be used with caution or not at all.
Blockworks: How might FICO be looking to use blockchain differently than competitors?
Zoldi: Before blockchain we had model standards and enforced these through agile standups, sprint reviews and approval meetings. Blockchain now provides the automated tool to enforce our model development standard and reduce the risk of requirements, tests and approvals slipping through the cracks undone.
Project sign-off requires the data science team to demonstrate on the blockchain that the model has been completed, tested and meets acceptance criteria before the model is released.
This level of automation and enforcement is unique and allows FICO to achieve great efficiency in automating compliance with its responsible AI standards, compared to other companies that may have not invested in a model development standard or have less controls on enforcement.
Blockworks: Why is FICO bullish on blockchain? How can blockchain help businesses in ways other technologies can’t?
Zoldi: We are bullish in our use of blockchain because it provides mini contracts of tasks completed, tested and accepted per our model development standards for Responsible AI. Each scientist, tester and manager are accountable to meeting FICO’s standard, and that accountability is enforced and recorded to the blockchain.
Accountability promotes the right level of seriousness in developing AI. This is paramount in achieving and promoting confidence in the models we develop. Moreover, models need owners manuals that record all the details during development — these are invaluable when the models are used in production.
Applying model details to the blockchain gives us the granular information we need to responsibly monitor, support and maintain models deployed in production.
Blockchain’s nature of enforcing contracts, and the immutability of those contracts, makes it a natural tool for AI and analytic governance and audit. I highly encourage other businesses to consider similar use cases in their data science organizations.
This interview was edited for brevity and clarity.
Don’t miss the next big story – join our free daily newsletter.