IRB is Our Solution to California SB 1047

Governor Newsom has until Monday, September 30, 2024, to sign SB 1047.

(edit, 9/29/2024, Governor Newsom vetoed SB 1047).

Institutional Review Boards (IRB). You either love them or tolerate them.

OK, you are hopefully asking, what in the world? IRB and SB 1047.

SB 1047

California's Senate Bill 1047, known as the "Safe and Secure Innovation for Frontier Artificial Intelligence Models Act".

• Developers must certify under penalty of perjury that their AI models comply with safety protocols before training and making them available for public or commercial use.
• Ensuring that the model does not pose an unreasonable risk of critical harm to humans.
• Developers need to implement and maintain detailed safety and security protocols, capable of quickly shutting down the model if necessary, and must keep unredacted copies of these protocols for inspection.
• Developers must report AI safety incidents. Essentially, technologists must hold themselves responsible for what they build.

?IRB

Three letters that can invoke frustration or excitement with researchers and academics. As a technologist in higher ed, I remember always asking the question, does the IRB need to review this.

An Institutional Review Board (IRB) is an administrative body established to protect the rights and welfare of human research subjects. Here are the key points:

• IRBs review and monitor biomedical and behavioral research involving human subjects.
• Their primary purpose is to ensure the protection of the rights, safety, and well-being of human research participants.
• IRBs have the authority to approve, require modifications to, or disapprove research protocols.

IRBs exist so that human’s having the opportunity in their work to cause human tragedy, de-evolution or impact human flourishing are held accountable.

Some of the common rules considered are.

1. Involves more than minimal risk to human subjects?
2. Involves vulnerable populations, particularly:Children / minors under 18?Prisoners?Individuals with impaired decision-making capacity?Economically or educationally disadvantaged person
3. Involves sensitive topics or illegal behaviors?
4. Includes genetic testing or complex research designs requiring expertise of multiple board members?
5. Involves procedures that might cause physical harm or significant psychological / emotional distress?
6. Collects information that could seriously harm participants legally, socially, financially if their identity was revealed?

Why Not IRB for technology?

Recently, I asked some long time AI technologists why IRB wasn’t required for technology.

Several examples.

1. Self-driving cars. Putting a self-driving car on the streets. Aren’t humans involved and this technology poses more than minimal risk to human subjects?
2. Social media. Pushing out endless dopamine inducing technology involves more than minimal risk to human subjects. Particular children and minors under 18, right? And, both physical harm (suicides) and psychological and emotional distress have been inflicted, right?

What am I missing. If technologists were required or voluntarily held themselves to the standard of using IRBs, we wouldn’t need SB 1047.

We have people calling SB 1047 an “attack on AI innovation”. Are IRBs also an attack on innovation.

It may well go down in history as one of the most important decisions that an American politician ever made.

Chris Anderson, https://x.com/TEDchris/status/1837970805491204124

• Artists supporting SB 1047: https://artists4safeai.com/
• Academics and CEOs calling for a veto: https://vetosb1047.github.io/letter/
• Calling for fear mongering in relationship to China: https://www.sacbee.com/opinion/op-ed/article292197435.html
• Caltech calling for a veto: https://docs.google.com/forms/d/e/1FAIpQLScgA1GCo241Kfg-S5X2hMVAYivkqPGSnaC0VwvTy11uVJ3OLw/viewform?pli=1
• Youth Leaders supporting: https://safesecureai.org/youth-leaders

Let’s get SB 1047 passed. And add technologists to the already existing IRB structure.