Steven Sinofsky: AI Doesn't Need New Rules Yet

Steven Sinofsky discusses why he believes the current push to regulate AI is premature and counterproductive. He argues that regulation is starting before anyone understands what they're actually regulating, and points to historical examples to support his thesis. With automobiles, it took 60 years before safety became a design criterion, leading to innovations like seatbelts and airbags. Similarly, the internet evolved through experimentation before regulation, and early precautionary regulation could have frozen it at AOL Instant Messenger or Web 1.0 rather than enabling YouTube, commerce, and other innovations. Sinofsky warns that the precautionary principle—regulating before problems occur—constrains the solution set and prevents beneficial innovation.

A key concern Sinofsky raises is regulatory capture, where companies lobby government to enact rules that benefit themselves while appearing to serve the public interest. He notes that AI companies have unusually asked Congress to regulate them, which he views skeptically as companies seeking to lock in advantages against competitors. He contrasts this with tech industry's historical stance of opposing regulation, and compares it to AT&T's arrangement where the government-backed monopoly promised universal phone access in exchange for regulatory control.

On open source specifically, Sinofsky argues there is no legitimate reason for AI companies to oppose open source models other than wanting to eliminate competition. He notes the irony that government-funded research is required to release software as open source, and that major tech companies like Microsoft benefited from open source communities despite building closed-source products. He views opposition to open source as "obnoxious" and anti-competitive.

Regarding existing legal frameworks, Sinofsky contends that nearly every problematic scenario people worry about—from non-consensual imagery to discriminatory lending—is already illegal. He suggests the first responsible step is ensuring existing laws properly apply to AI (similar to how EV safety rules had to account for batteries and front trunks), rather than creating entirely new regulations before understanding the technology.

On the U.S.-China competition, Sinofsky explains that both countries are using indirect "bank shots" rather than directly attacking AI. China is considering export controls on chips, while the U.S. considers restricting Chinese open source models. He argues that when companies like Anthropic advocate for curtailing open source to hurt Chinese competitors, they're actually advocating for rules that help themselves—conflating corporate interest with national interest. He views this as contrary to American tech industry traditions and values.