Commercial & Citizen Interests
Capture
“Regulating AI requires a lumbering bureaucracy to take on notoriously secretive companies and rapidly evolving technologies.”
Regulatory capture refers to a situation in which agencies, policymakers, or regulators are influenced or co-opted by the industries or interests they are supposed to regulate, rather than acting in the public interest.
It is a form of corruption.
When actors in a particular industry, like, say Purdue Pharma L.P. or OpenAI, successfully influences actions of regulators charged with oversight, in order to serve their own commercial, ideological, or political interests, that's regulatory capture.
Explain Harms
Organizations with a mandate to develop products and/or services to benefit customers, employees, and shareholders should adopt a systems thinking approach to evaluating the benefits and harms of Large Language Models.
Just like the Sackler family inserted the words "is believed" into the OxyContin approval process, Product teams should be wary of vendors of AI products who engage in hand-waving, relying on "scientific evidence"...that is actually a letter to the managing editor of WIRED.
The R&D departments engaged in developing LLMs are still struggling with "model explainability": that is, they can't explain how or why the things work.
For ecosystems - like health care and financial services - that earned oversight, it was because the severity and duration of harms were bad or long enough that the federal or state government was forced to respond to the outcry of citizens. Even then, regulatory capture occurred along the way.
Product Governance
Product teams motivated to move the needle for customers and solve problems probably can't wait for a lumbering bureaucracy to set for standards and policy.
That's where data governance comes in.
For regulated ecosystems, the first contributors to a working group should likely involve Legal and Compliance.
Data has been the new oil, but it's a volatile material that came from somewhere, was generated by someone, and they might have opinions on it being used to help (or hurt) them.
The European Union is far ahead of the United States in treating data privacy as a right.
The re-training costs to remove information from data sets should be a consideration in the business case.
A Product organization that is practicing active data governance is at least making a good-faith effort to not get steam-rolled in the process.