It’s Time To Set Industry Standards for AI
The AI hype cycle has shifted from real promise to real life. The intense complexities that accompany enterprise AI – like the data issues, sustainability and scalability, and of course, the ethics – have permeated boards, panels and analytics teams all trying to plow through.
AI technology has nearly unlimited potential to do good for businesses and consumers, but at the same time it can do untold harm. The outcome is up to us.
A while back, I collaborated with a global bank struggling to get its models into production. Issues largely centered on flawed processes with little to no model auditing, documentation or development standards.
As an impartial advisor, I shared a “state of the state” on AI, where we explored the nature and sensitivities of explainable AI and the foundation of robust AI development.
We discussed the need for a model development standard coupled with a blockchain for model development governance approvals and audit during the actual model build, instead of waiting until after it’s built to start asking questions.
One theme emerged from our discussions: the importance of developing a corporate standard across the artistry of analytic developers and methods. Think: houses built on the same block look odd if they’re constructed using different standards, or worse, they could be unsafe or unhabitable. Here again, a strictly adhered to and enforced building code is essential.
Change Existing Methods
Another international bank needed to reorganize its internal analytics function and engaged me in an advisory capacity. Various analytics teams were siloed under different business functions creating “analytic city states” within the organization that made it difficult to share and operationalize the best ideas and technologies.
Some teams were stronger than others, some were intolerant of ideas that weren’t their own, and others had near-religious zeal for their preferred analytic techniques and tools. This drove non-productive competition between teams that didn’t benefit the bank or its customers, and certainly didn’t support following a single corporate model development standard.
Working together, we reached a consensus: centralize everything under one chief analytic officer to enable the creation and enforcement of organization standards. You can safely build more houses when you don’t have to draft a new building code for every house. Likewise, you shouldn’t have to worry about rolling the dice as to which artist will be building your model.
Dealing with AI and Big Data Challenges
Last, I’ll never forget a memorable experience as an invited speaker at a bank in Brazil. The talk focused on AI and Big Data challenges in rapidly changing business environments and the appropriate use of supervised, semi-supervised and unsupervised models.
We drilled deep into the concept of explainable and ethical AI and its ability to be fair and trustworthy. This segued into how models can be wrong, and one must employ humble AI, and more importantly, empathy. Instead of promoting the myth and promise of AI as “superhuman,” we pulled back the curtain and talked about where the technology goes wrong, exposing its various warts, pitfalls, and traps.
Over lunch afterwards, the bank CEO committed his teams to implementing well-built ethical and responsible AI and using it to create new equitable opportunities that could elevate the people in his region who are disadvantaged and living in severe poverty.
There are truly huge transformative aspects of AI, but it’s our job as C-level analytic executives and data scientists to facilitate honest, frank discussions about how we enable that transformation safely to help ensure meaningful and ethical outcomes for everyone.
AI Industry Stanards Will Be Transformative
Our industry is at a pivotal moment.
Enterprise adoption of AI is still fairly limited with just 25 percent of analytics leaders reporting that their organizations have a fully unified, enterprise-standard approach to delivering AI projects, according to the Corinium study.
We can continue to work and experiment with this incredibly powerful technology in siloes, often without standard blueprints or an established building code and hope for the best outcomes.
Or we can come together and take the first step toward building a solid foundation of specific industry AI standards.
The time is right to start the conversation—won’t you join me?
About the Author
Scott Zoldi is Chief Analytics Officer at FICO, driving the company's innovation in artificial intelligence in its products. Zoldi has been responsible for authoring 110 analytic patents with 56 patents granted and 54 in process for the company. He is an industry leader in developing practical applications and standards for AI, Explainable AI, Ethical AI and Responsible AI. Zoldi serves on the Boards of Directors of Tech San Diego and Cyber Center of Excellence, and on the Cybersecurity Advisory of the California Technology Council. He received his Ph.D. degree in theoretical physics from Duke University.