David Sacks on Trump's executive order to create national AI standard

Thanks, Maria. Good to be here.

Well, that's right, Maria. The problem we have is that there's currently a thousand bills going through state legislatures right now. Over a hundred measures have passed. Some of them are contradictory and it's creating this confusing patchwork of 50 different state regulatory regimes. That's hard for innovators to follow, especially small startups. It's very hard for them to go to 50 different states and try to comply. So what the president has called for here is a single federal standard for AI regulation. We've declared that we're going to work with Congress on that. At the end of the day, we would like Congress to develop and pass that set of laws. This EO contains a statement of principles and it also creates a set of tools to push back on state regulations that are excessively onerous. But we're not saying we're going to push back on everything. For example, we want to protect child safety. We're not going to interfere with the rights of communities to say whether they want data centers or not. Those are local prerogatives. We're not pushing back on those things. But if there are state regulations that are excessive or onerous, then those are the things we want to push back on.

Sure. First of all, infrastructure is a completely separate issue. We're not touching that. Let me give you a couple of examples on the rule-making. Right now in New York, Kathy Hochul herself has said that the RAISE Act they just passed to regulate AI is too different from what's happening in other states and she's wondering whether they should just replace it with what was passed in California, which was SB 53. Even Democrat governors are realizing there's a problem and saying maybe we need to try to synchronize on one standard. That SB 53 requires tech companies to report various things to the government. You can imagine there could be 50 different definitions of what to report, who to report to, when to report. It's very confusing if every state goes their own way. So this is why it'd be a lot better to have one federal standard. Another example: In Colorado, they passed a law that prohibits what they call algorithmic discrimination. If an AI developer creates a model that has a disparate impact on a protected group, that could be illegal. Their list of protected groups is very long, including people who are not proficient in English. So presumably if your AI model says something bad about an illegal alien in Colorado, that could be illegal. We consider that interference with the First Amendment. We don't like the idea of ideology being inserted or required in AI models. We saw that during the Biden administration where they tried to require DEI be put in AI models. We don't like that. We want AI to remain ideologically unbiased. We want to pursue the truth, not an ideological agenda. That's an example of an excessive regulation we would look at pushing back on on First Amendment grounds.

Right. I think there's a First Amendment issue when states try to require AI developers to twist or adjust their models to produce deceptive outputs to meet a political agenda. That's the kind of thing we want to push back on. We don't want to see woke AI. Earlier this term, the president signed an executive order saying the federal government would not be procuring woke AI, defined as politically biased or highly ideological AI. We want to see the same thing at the state level.