California State Assembly Has Approved S.B. 1047, Which Can be Used Against Proprietary Software
Plus, sage old advice from Capt. Grace Hopper
THE American national archives have just been augmented by an epic old lecture or two of a scientist-celebrity (or national treasure) whose work interests us. This has been receiving some publicity, but what interests us most is how she talks of "cost of conversion" aka cost of proprietary, non-standard software. There's more through to the next 10 minutes or so. The subject came up in IRC several times this week (some people brought it up). "The Grace Hopper video is quite good," one person said. "The lecture was 1 year before GNU. FWIW the GNU Manifesto turns 40 next year. Something to put in the calendar."
Here is a lecture (in two parts): "Capt. Grace Hopper on Future Possibilities: Data, Hardware, Software, and People (Part One, 1982)" and "Capt. Grace Hopper on Future Possibilities: Data, Hardware, Software, and People (Part Two, 1982)".
This is rather timely because of growing recognition of the threat of proprietary software, even if most lawmakers allude to it using buzzwords like "hey hi" (AI). It means software that does something you neither understand nor control. Forget about the formal scientific definitions; those have long been abandoned by the media, which seems to be allocated a budget to hype up the bubble and feed the Ponzi scheme until the bubble "pops".
"The California State Assembly approved S.B. 1047 against "AI"," we got told. "Of course the devil is in the details, but if done right (that's a big *if*) then it could play a role in the advancement of FOSS and Open Data."
Revisit this old talk by Geer and especially see paragraphs starting with: "1. If you deliver your software with complete and buildable source code and a license that allows disabling any functionality or code the licensee decides, your liability is limited to a refund."
"Hey hi" typically means both the training set and the code are unavailable. Thus, the blackbox isn't even understood by its maker. That ought not be done.
As per press reports [1], "California state lawmakers attempted to introduce 65 bills touching on AI this legislative season" and [2] a "proposed law would require companies working on AI to test their technology before selling it for “catastrophic” risks". The law "would require big A.I. companies to test their systems for safety". [3]
The Verge says "Senator Scott Wiener, the bill’s main author, said SB 1047 is a highly reasonable bill" and we suggest reading "hey hi" as proprietary software because in practice that's just what they often refer to.
Ideally, the code will be Free software and models, if training upon some data, will make the data openly and freely available for audits (in order to prevent mischief). One way to make algorithms intentionally misbehave is to manipulate or bias the training set. So both are strictly needed. █
Related/contextual items from the news:
-
Musk voices support for California bill requiring safety tests on AI models
California state lawmakers attempted to introduce 65 bills touching on AI this legislative season, according to the state’s legislative database, including measures to ensure all algorithmic decisions are proven unbiased and protect the intellectual property of deceased individuals from exploitation by AI companies. Many of the bills are already dead.
-
California AI bill 1047, opposed by Pelosi, passes State Assembly
The proposed law would require companies working on AI to test their technology before selling it for “catastrophic” risks such as the ability to instruct users in how to conduct cyberattacks or build biological weapons. Under the proposed law, if companies fail to conduct the tests and their tech is used to harm people, they could be sued by California’s attorney general. The bill only applies to companies training very large and expensive AI models, and its author, Democratic state Sen. Scott Wiener, has insisted it will not impact smaller startups seeking to compete with Big Tech companies.
-
California Legislature Approves A.I. Safety Bill
The State Assembly approved the measure, known as S.B. 1047, which would require big A.I. companies to test their systems for safety before releasing them to the public. The bill would also give the state’s attorney general the power to sue A.I. makers for serious harms caused by their technologies, like death or property damage.
-
California State Assembly passes sweeping AI safety bill
Senator Scott Wiener, the bill’s main author, said SB 1047 is a highly reasonable bill that asks large AI labs to do what they’ve already committed to doing: test their large models for catastrophic safety risk. “We’ve worked hard all year, with open source advocates, Anthropic, and others, to refine and improve the bill. SB 1047 is well calibrated to what we know about forseeable AI risks, and it deserves to be enacted.”