In last year’s polarizing “Techno-optimism Manifesto,” venture capitalist Marc Andreessen listed a number of enemies of technological progress. These included “technical ethics” and “trust and security” – a term used to describe internet content moderation efforts that he said were being used to subject humanity to a “mass demoralization campaign” against fresh technologies such as artificial intelligence.
Andreessen’s declaration was met with public and quiet criticism from those working in these fields, including at Meta, where Andreessen is a board member. Critics saw his text as a misrepresentation of their work to secure internet services.
On Wednesday, Andreessen offered some clarification: When it comes to his 9-year-old son’s online life, he’s a stickler for guardrails. “I want him to be able to sign up for online services and I want him to have a Disneyland experience,” the investor said in an on-stage conversation at a conference for the Human-Centered AI research institute at Stanford University. “I love free internet for everyone. One day he will also love free internet for all, but I want him to have walled gardens.”
Despite what his manifesto may have sounded, Andreessen added that he welcomes tech companies – and, by extension, their trust and safety teams – setting and enforcing rules about the type of content allowed on their services.
“Company after company has a lot of freedom to make decisions,” he said. “Disney enforces different codes of behavior at Disneyland than what happens on the streets of Orlando.” Andreessen alluded to the fact that tech companies could face government fines for allowing images of child sexual abuse and certain other types of content, so they couldn’t do without trust and safety teams.
So what type of content moderation does Andreessen consider to be the enemy of progress? He explained that he feared that two or three companies would dominate cyberspace and “connect” with the government in a way that would make certain restrictions universal, causing what he called “powerful social consequences,” without specifying what they were. may be. “If you go into an environment where there is pervasive censorship and pervasive control, you have a real problem,” Andreessen said.
The solution, he described, is to ensure competition in the tech industry and a diversity of approaches to content moderation, with some having greater restrictions on speech and action than others. “What happens on these platforms really matters,” he said. “What happens in these systems really matters. What happens in these companies really matters.”
Andreessen made no mention of X, the social media platform run by Elon Musk and formerly known as Twitter that his company Andreessen Horowitz invested in when Tesla’s CEO took the helm in delayed 2022. Musk soon fired most of the company’s trust and security employees, shut down Twitter’s AI Ethics Team, relaxed content rules and reinstated users who had previously been permanently banned.
These changes, combined with the investment and Andreessen’s manifesto, created the impression that the investor wanted few restrictions on free speech. His clarifying remarks were part of a conversation with Fei-Fei Li, co-director of HAI at Stanford University, titled “Removing Obstacles to a Robust Artificial Intelligence Innovation Ecosystem.”
During the session, Andreessen also repeated arguments made over the past year that slowing the development of artificial intelligence through regulation or other measures advocated by some AI safety advocates would repeat what he says is a misguided U.S. limit on nuclear investment decades ago.
Nuclear power would be a “silver bullet” to many of today’s concerns about greenhouse gas emissions from other sources of electricity, Andreessen said. Instead, the United States withdrew, and climate change was not contained as well as it could have been. “It’s definitely a negative risk aversion frame,” he said. “The discussion assumes that if there are potential harms, there should be regulations, controls, restrictions, breaks, stops, freezes.”
For similar reasons, Andreessen said he wants greater government investment in AI infrastructure and research, and more freedom to experiment with AI, such as by not restricting open-source AI models in the name of security. If he wants his son to have an AI experience at Disneyland, some policies developed by governments or trust and safety teams may also be necessary.