Speaker
Steven Sinofsky
Appearances over time
1 episodes
Episodes
1
Podcasts
Quotes & moments
It took roughly 60 years of automotive evolution before seatbelts and airbags became upstream design criteria — illustrating why premature regulation stifles innovation.
Just two years ago, AI company leaders went to Congress and literally pleaded to be regulated — a move Sinofsky called a 'crazy notion' that handed government the leverage it had been seeking over tech.
Sinofsky claims that every scenario critics identify as an AI risk — non-consensual nudity, discriminatory lending, dangerous drugs — already has a law against it.
If you receive a government grant for computer science research, you are required to release your software as open source — making government opposition to open-source AI internally contradictory.
Japan set up a huge national ministry in the 1980s to dominate the memory chip industry — and ultimately failed, showing that centralized government planning rarely wins technology races.
Detroit lobbied the U.S. government to stop Japanese car imports by claiming dumping — Japan responded by building factories in the U.S., but Detroit still lost and no longer makes cars.
Sinofsky noted that he had only been at Microsoft for about two years before the government began investigating the company in 1992, illustrating how quickly government scrutiny can arrive.
AT&T, a private company, became effectively the United States national phone company by promising universal telephone access to every address in America in exchange for a government-sanctioned monopoly.
Regulating before harm occurs doesn't prevent it — it constrains the available solutions, potentially freezing innovation at an immature stage, like locking the internet at AOL Instant Messenger.
Before writing new AI laws, governments should first audit whether the roughly 2 million laws already on the books apply to AI use cases — covering CSAM, spam, surveillance, and more.
One state passed a law saying AI can't be registered as an attorney — but this is already impossible under existing licensing law, illustrating how redundant new AI rules often are.
Sinofsky frames U.S.-China AI competition as an 'innovation leadership war' rather than a Cold War or trade war, noting mutual trade reliance complicates direct economic confrontation.
Every major technology — cars, planes, the internet — took decades before the right regulatory moment arrived. Regulating AI now, before we even understand what it is, risks locking in the wrong answers at the worst possible time.
Two years ago, AI company executives went to Congress with fear in their eyes, begging to be regulated. That wasn't safety-consciousness — it was regulatory capture. They handed government the opening it had been waiting for since it missed regulating the PC, the internet, and the mainframe.
New AI-specific laws are almost entirely redundant. Non-consensual nudity is already illegal. Discriminatory lending is already illegal. Dangerous drugs are already illegal. Even AI practicing medicine without a license is already impossible under existing licensing law. The only thing new regulation does is give companies a way to kneecap competitors.
Before writing a single new AI law, governments should audit the 2 million laws already on the books and ask: do they apply to AI? This is what happened with EVs — engineers had to verify that car safety rules designed for combustion engines still made sense for batteries. Do the same for AI, and you'll know what's actually missing.
A doctor who lets AI write a patient note and gets it wrong is fully liable — it's her medical license, not the AI's. Accountability doesn't disappear just because AI was involved. This clarity already exists in professional licensing law; we don't need new rules to establish it.
The U.S. and China aren't in a Cold War or even a classic trade war — they're too economically intertwined for that. What they're fighting is an innovation leadership war, and the only tools governments have are indirect: ban chips, apply tariffs, nationalize companies. These 'bank shots' rarely hit the target.
Detroit convinced the government that Japan was dumping cars below cost. Japan responded by building factories in the American South. Detroit still collapsed. The same playbook is being run today against Chinese AI companies — and historical precedent suggests it won't work either.
The government missed regulating the mainframe. It missed regulating the PC. It missed regulating the internet. AI regulation is its chance for a do-over — and AI company executives handed them the opening by showing up to Congress and begging for it. That's a once-in-a-generation political gift.
When Anthropic argues that restricting open source is the best way to hurt China, the actual translation is: restricting open source is the best way to help Anthropic by eliminating a competitor. Sinofsky calls it un-American and un-tech — an industry that was built on open academic research trying to pull up the ladder behind it.
The precautionary principle sounds sensible but leads to one outcome: you freeze the technology at whatever stage the government is most comfortable with. If applied to the internet in the 1990s, we'd all still be using Yahoo and AOL with no video, no audio, no commerce.
Analysis
What they talk about
- Government 42%
- Technology 33%
- History 17%
- Health & Fitness 8%
