The two biggest forces in two deeply intertwined tech ecosystems – large incumbents and startups – have taken a break from counting money and together they demand this from the federal government to stop even considering regulations that might affect their financial interests or, as they prefer to call it, innovation.
“Our two companies may not agree on everything, but it’s not about our differences,” writes this group with very different perspectives and interests: A16Z founders, partners Marc Andreessen and Ben Horowitz, and Microsoft CEO Satya Nadella and president/director legal affairs Brad Kowal. A very cross-sectional gathering, representing each big business and big money.
But they are supposedly taking care of little boys. That is, all the businesses that may be impacted by this latest try to abuse the regulations: SB 1047.
Imagine being charged a fee for improperly disclosing an open model! A16Z General Partner Anjney Midha he called it a “regressive tax” on startups and a “blatant regulatory capture” by Big Tech firms that, unlike Midha and his impoverished colleagues, could afford the lawyers needed to comply with the regulations.
Except that was all disinformation spread by Andreessen Horowitz and other wealthy interests who actually stood to suffer as supporters of billion-dollar enterprises. In fact, small models and startups would only be barely affected since the proposed law specifically protected them.
It’s strange that the identical form of targeted carve-out for “Little Tech” that Horowitz and Andreessen routinely advocate for was distorted and minimized by the lobbying campaign they and others waged against SB 1047. (In an interview with the bill’s sponsor , California State Senator Scott Wiener talked about this whole thing recently on Disrupt.)
This bill had its problems, but its opposition greatly exaggerated compliance costs and didn’t significantly substantiate claims that it will chill or burden startups.
It’s a part of a longtime pattern in which Big Tech – to which, despite their stance, Andreessen and Horowitz are closely related – operates on the state level, where it could possibly win (as with SB 1047), while asking for federal solutions that it knows will won’t ever come, or which can have no teeth because of partisan bickering and congressional ineptitude on technical issues.
This joint statement of “political opportunity” is the second a part of the sport: After torpedoing SB 1047, they will say they did it solely to support federal policy. Never mind that we’re still waiting for a federal privacy law that tech firms have been pushing for a decade while fighting state laws.
What policies do they support? “A different responsible market approach”, in other words: down with our money, Uncle Sam.
Regulations needs to be based on a “science-based and standards-based approach, recognizing regulatory frameworks that focus on the use and misuse of technology” and should “focus on the risk of bad actors exploiting artificial intelligence.” This signifies that we must always not introduce proactive regulation, but quite reactive penalties when criminals use unregulated products for criminal purposes. This approach has worked great in this whole FTX situation, so I understand why they support it.
“The regulation should only be implemented if the benefits outweigh the costs.” It would take 1000’s of words to clarify all of the ways this idea expressed in this context is funny. But they are principally suggesting that the fox needs to be included on the henhouse planning committee.
Regulators should “allow developers and startups the flexibility to choose AI models to use wherever they build solutions, and not tilt the playing field in favor of any one platform.” This suggests that there may be some agenda requiring permission to make use of one model or one other. Since this is just not the case, it’s a straw man.
Here is a lengthy quote that I have to quote in full:
The right to education: Copyright goals to advertise the progress of science and the applied arts by extending protection to publishers and authors to encourage them to make recent works and knowledge available to the general public, but not on the expense of society’s right to learn from those works. Copyright law mustn’t be co-opted to suggest that machines needs to be prevented from using data – the premise of artificial intelligence – to learn in the identical way as humans. Unprotected knowledge and facts, whether or not contained in protected subject material, should remain free and accessible.
To be clear, the clear statement here is that software operated by billion-dollar corporations has the “right” to access any data since it should give you the chance to learn from it “in the same way as humans.”
First of all, no. These systems are not like people; they generate data in their training data that mimics human activity. These are complex statistical projection programs with a natural language interface. They haven’t any more “right” to any document or fact than Excel.
Second, the concept that “facts” – by which they mean “intellectual property” – are the one thing these systems are interested in, and that some type of fact-gathering cabal is working to forestall them, is an artificial narrative we have seen before. Perplexity made the “facts belong to everyone” argument in its public response to a lawsuit alleging systematic content theft, and its CEO Aravind Srinivas repeated that mistake to me on stage at Disrupt, as in the event that they were being sued for knowing tidbits just like the Earth’s distance from the Moon.
While this is just not the place to totally discuss this particular straw man argument, let me simply indicate that while facts are indeed free agents, there are real costs to how they are created – say, through original reporting and scientific research. This is why copyright and patent systems exist: not to forestall the wide sharing and use of mental property, but to encourage its creation by ensuring that it could possibly be assigned real value.
Copyright law is much from perfect and is more likely to be abused as often as used. However, this is just not “co-opted to suggest that machines should be prevented from using data” – it’s used to be sure that bad actors don’t bypass the worth systems we’ve got built around mental property.
This is a fairly clear query: let’s allow the systems we own, operate and take advantage of to freely use the worthwhile work of others without compensation. To be fair, this part is “in the same way as people” because people design, run and implement these systems, and these people don’t desire to pay for something they do not have to, and they don’t desire to. I don’t desire regulations to alter that .
There are many other recommendations in this small policy document, which were little question covered in greater detail in the versions sent on to lawmakers and regulators through official lobbying channels.
Some of the ideas are undoubtedly good, if slightly selfish: “fund digital literacy programs that help people understand how to use artificial intelligence tools to create and access information.” Good! Of course, the authors invest heavily in these tools. Support “Open Data Commons – collections of accessible data managed in the public interest.” Great! “Examine procurement practices to enable more startups to sell technology to the government.” Excellent!
But these more general, positive recommendations are something the industry sees yearly: invest in public resources and speed up government processes. These tasty but irrelevant suggestions are merely tools for the more vital ones I described above.
Ben Horowitz, Brad Smith, Marc Andreessen and Satya Nadella want the federal government to step back from regulating this lucrative recent development, let industry resolve which regulations are value compromising, and invalidate copyright laws in a way that kind of acts as a blanket reprieve for illegal or unethical practices that many imagine have enabled the rapid development of artificial intelligence. These are principles that are vital to them, whether children are acquiring digital skills or not.