How to Use Blockchain to Build Responsible AI: An Award-Winning Approach

“Innovation doesn’t happen overnight,” says the introduction to The Innovators 2023—Global Winners in Global Finance Magazine. In fact, innovation can take years to be recognized. This was the case with FICO’s now-patented use of blockchain technology for artificial intelligence and analytic model governance; the U.S. patent application was filed in 2018 and was granted in 2023.

As the principal inventor on this patent, I am extremely gratified that FICO’s novel application of blockchain technology has been recognized with two awards. Global Finance named it one of the World’s Best Financial Innovations 2023. And now this technology has won a Banking Tech Award for Tech of the Future. 

Innovation Doesn’t Gain Traction Overnight

Many inventors’ innovations were ahead of their time — contact lenses, solar cells, flushing toilets and electric cars, to name a very few. When I first began blogging about blockchain technology being used for AI model governance in 2017, I got a lot of side-eye and confused looks. Here’s how I described applying blockchain technology to AI, in my 2018 AI predictions blog:

Beyond its association with cryptocurrencies, blockchain technology will soon record “time chains of events,” as applied to contracts, interactions and occurrences. In these “time chains,” people and the items we interact with will have encrypted identities. The blockchain that is distributed will be the single source of truth, allowing audit trails of data usage in models, particularly in data permission rights… Beyond that, data event chains will create new opportunities for graph analytics and novel new AI algorithms to consume relationship data at scale.

In 2018 I refined these concepts, producing a patent application around using blockchain to ensure that all the decisions made about a machine learning (ML) model are being recorded and are auditable. These include the model’s variables, model design, training and test data utilized, selection of features, the ability to view the model’s raw latent features, and recording to the blockchain all scientists who built different portions of the variable sets, and participated in model creation and model testing. In 2023, the granted patent is described as:

“Blockchain for Data and Model Governance” – Covering the concept of a shared ledger that is leveraged to track end-to-end provenance of the development, operationalization, and monitoring of a machine learning models in an immutable manner. The solution and underlying methodology enforce the use of a corporate model development standard to ensure Responsible AI practices are abided by organizations.

Accountability Matters in Artificial Intelligence

As enabled by blockchain technology, the sum and total record of these decisions provides the visibility required to effectively govern models. Model governance and transparency are essential in building ethical AI technology which is auditable; as a data scientist and member of the global analytics community, creating ethical analytic technology is very important to me, particularly in my role of serving financial and enterprise customers.

The meteoric rise of ChatGPT and other large language model (LLM) tools similarly propelled artificial intelligence ethics (more pointedly, AI “hallucinations”) into the upper echelons of pop culture. In 2023, ordinary citizens gained first-hand knowledge of the perils of “black box” artificial intelligence for the first time; they were using it every day to do everything from cheat on homework to fuel body dysmorphia. All of a sudden, everyone wanted to know how, and why, these AI models work. But because their decisioning isn’t explained by creators such as OpenAI, Anthropic, Google, Meta and dozens of others, we don’t know.

I believe we are in the throes of an AI accountability crisis, which frankly will get worse before it gets better. It will only get better as businesses and consumers demand that AI models be more explainable. 

How Blockchain Can Reduce Negative Outcomes from Artificial Intelligence

More than a decade ago, long before blockchain and ChatGPT became buzzwords, I began implementing a structured approach to AI model development in my data science organization. In 2010 I instituted a development process centered on an analytic tracking document (ATD). This approach detailed model design, variable sets, scientists assigned, train and testing data, and success criteria, breaking down the entire development process into three or more agile sprints.

I recognized that a structured approach was required because I’d seen far too many negative outcomes from what had become the norm across much of the banking industry: a lack of validation and accountability. A decade ago, the typical lifespan of an analytic model looked like this:

• A data scientist builds a model, self-selecting the variables it contains. This led to scientists creating redundant variables, not using validated variable design and creating new errors in model code. In the worst cases, a data scientist might make decisions with variables that could introduce bias, model sensitivity, or target leaks.
• When the same data scientist leaves the organization, his or her directories are typically deleted. Often, there were a number of different directories and it was unclear what directory(ies) were responsible for the final model. The company wouldn’t have the source code for the model or might have just pieces of it. No one could definitively understand how the model was built, the data on which it was built, and the assumptions that factored into the model build.
• Ultimately this becomes a high-risk situation when the model is assumed to have been built properly and that it will behave well—but not really knowing either. This would often result in being unable to validate the model or understand under what conditions the model should be used. These realities drive unnecessary risk or a large number of models being discarded and rebuilt, often repeating the journey above.

Unfortunately, the companies releasing LLM AI systems do not exercise anywhere near the level of prudence or restraint of financial institutions. Driven by pressure to go to market faster than competitors, LLMs are quickly released “into the wild”—which has been compared to “open sourcing the Manhattan Project”—escalating their unpredictability and resulting in negative outcomes for the people and businesses using them.

A Blockchain to Codify Accountability

My patent describes how to codify AI, analytic and machine learning model development using blockchain technology to associate a chain of entities, work tasks and requirements with a model, including testing and validation checks. Blockchain replicates the approach I use to build models in my organization––essentially a contract between my scientists, managers and me that describes:

• What the model is
• The model’s objectives
• How we’ll build that model
• Areas that the model must improve upon, for example, a 30% improvement in card not present (CNP) fraud at a transaction level
• The degrees of freedom the scientists have to solve the problem, and those they don’t
• Re-use of trusted and validated variable and model code snippets
• Prescribing the model algorithms allowed
• Training and test data requirements
• Specific model testing and model validation checklists
• Specific assigned analytic scientists to build the variables, models, train them and those who will validate code, confirm results, perform testing of the model variables and model output
• Ethics and robustness tests prescribed
• Specific success criteria for the model and specific customer segments
• Specific analytic sprints, tasks and scientists assigned, and formal sprint reviews/approvals of requirements met.

As illustrated in Figure 1, all this gets broken into a set of requirements, roles and tasks which are put on the blockchain to be formally assigned, worked, validated and completed. Having individuals who are tracked against each of the requirements, the team then assesses a set of existing collateral, which are typically pieces of previous validated variable code and models. Some variables have been approved in the past, others will be adjusted, and still others will be new. The blockchain then records each time the variable is used in this model — for example, any code that was adopted from code stores, written new and changes that were made — who did it, which tests were done and the modeling manager who approved it, and my sign-off. 

 

In this way, blockchain enforces the proper use of the approved machine learning algorithms supported by the corporate model development standard. Moreover, it ensures that all ethics and robustness tests are completed, in addition to each requirement and success/acceptance criterion being met. No model is released without this process being completed, and it is all enforced using the blockchain.

Models with More Explainability, Less Bias and Fewer Hallucinations

In sum, overlaying the model development process on the blockchain gives the analytic model its own entity, life, structure and description. Model development becomes a structured process, at the end of which detailed documentation can be produced to ensure that all elements have gone through the proper review. These can be also revisited at any time in the future, and provides essential assets for use in model governance.  

 

This use of blockchain to orchestrate the corporate model development standard and agile model development process also can be used by parties outside the development organization. If an external party wanted to audit the way a critical model was built, their review could produce a statement such as, “all variables haven been reviewed and were approved by …” Likewise, if a revision or change in requirements made us want to understand all uses in production of variable #51817 from our large asset inventory of variables, we could easily query the blockchain to pinpoint any deployments in production models.

With blockchain to persist the analytic model development process, it becomes highly transparent and subsequent decisions auditable, a critical factor in delivering AI technology that is ethical and explainable. Explainability is essential in eradicating bias and “hallucinations” from the AI models used to make decisions that affect individuals’ financial lives—and, in the case of LLMs, their daily lives.

It's true that innovation doesn’t happen overnight. But the need for innovation to be applied sometimes does. I believe our society is at that moment today with widespread use of LLM artificial intelligence tools; more than ever, model governance is paramount in creating AI that is explainable, ethical and safe.

How FICO Can Help You Develop and Use Responsible AI

• Watch my LinkedIn Live conversations on Responsible AI
• Discover FICO’s AI, machine learning and advanced analytics solutions
• Download our AI Playbook: A Step-by-Step Guide for Achieving Responsible AI
• Review The State of Responsible AI in Financial Services