Andreessen Horowitz is laying out a framework on Wednesday that it wants Congress to take up for artificial intelligence, including ideas on protecting kids and preserving a role for states despite the federal approach.
The nine pillars come as Congress has failed twice this year to preempt state regulation of AI. President Donald Trump also signed an order last week that could yank broadband funding from states with “onerous” AI laws.
The AI and venture industries have been eager to avoid varying state regulations. The failures in Congress, however, have highlighted the political need to go beyond preemption and put in place national policies that address concerns around AI.
“We wanted to be able to crystallize this into one living document that I think could help lawmakers,” said Collin McCune, head of government affairs at a16z.
Kids and teens. The framework says kids under 13 shouldn’t be able to use AI without their parents’ say-so, but the document warns against stopping 13 to 17 year olds from using all algorithmic tools.
Rather, if services know their users are in that latter age bracket, parents should use controls like “the ability to set privacy and content settings” or the power “to impose usage limits or blackout hours.” The teens should also get reminders that the AI isn’t human or a licensed professional.
AI tools should be required to have plans in place for when teens under 18 discuss wanting to harm or kill themselves. Those plans must include refusing to facilitate self-harm and referring users to crisis resources.
States and risks. The proposal from a16z suggests that, even with preemption, states should still be allowed to enforce their laws on harmful AI “in areas like consumer protection, civil rights, children’s safety and mental health.” That’s a wider lane for states than some Republicans hope to see.
The policy framework suggests AI models offer users visibility into who built the systems, their intended use and other information. States such as California and Colorado mandate further-reaching disclosures into areas like risk assessments.
Another pillar suggests testing to lessen AI’s “ability to evade human control” and to make sure AI doesn’t facilitate chemical, biological, radiological or nuclear attacks.
Other pillars of the framework involve supporting workers, using AI “to modernize government service delivery” and investing in infrastructure and research.
In discussing the framework, McCune, one of its authors, rejected the wide perception that a16z wanted preemption without standards. He noted the firm had pushed out some of the ideas now in the framework starting almost a year ago.
“The thing that I think everyone agrees on is we need bills,” McCune said.
