Sam Altman: The Two Major Risks of AI are Loss of Control and Power Concentration
OpenAI CEO Sam Altman expressed in a conversation with David Senra that his biggest concerns regarding AI are: AI becoming too powerful leading to humans losing control, and power being concentrated in one person, one company, or one model.
He emphasized that the correct path should be to enable people to deeply control the future, arguing that placing all trust in models and allowing them to make decisions for the world is anti-human.
Some in the AI field advocate for curing diseases and reducing costs in exchange for people giving up autonomy and influence, exacerbating inequality under the guise of safety. Altman called this a poor sales pitch.
Altman anticipates that the pace of AI capability advancement will outstrip society's and the economy's ability to absorb it, with human habits and institutional inertia slowing the transition. Meanwhile, human connections will become more valuable, and AI will drive a surge of small businesses.
OpenAI positions itself as a platform, providing direct interfaces and APIs, and strategically needs to actively eliminate good ideas to concentrate resources on great ideas, starting the company without customers, products, or established methods.
Discussions on AI safety are shifting from technical alignment to power structures, with funding and policy resources flowing towards open empowerment and platformization, benefiting a broad user base and small businesses, while putting pressure on those advocating for centralized control. Event-driven prioritization of human agency in AI governance is emerging.
Source: Public Information
ABAB AI Insight
Sam Altman has long been influenced by Peter Thiel and Paul Graham, emphasizing non-consensus bets and power law thinking. This framework has continued from his time at Y Combinator to OpenAI, focusing capital paths on computing power and research rather than short-term product dispersion.
Resources are concentrated on platform interfaces and iterative deployment, motivated by the desire to avoid excessive power concentration and keep humans in the loop, relying on early unscripted entrepreneurship to accumulate feedback from reality.
Analogous to the early unregulated expansion of the internet and the Google search entry point, AI is currently in the expansion phase from laboratory capabilities to social deployment, with the industry shifting from a single model competition to a platform ecosystem and agency balance.
Essentially, this represents a transfer of pricing power, with control shifting from model owners to users and developers, facilitated by open interfaces that lower the barriers to use and dilute the decision-making monopoly of a single entity.
ABAB News · Law of Cognition
- Empowerment takes precedence over concentrated safety
- Good ideas must be actively killed
- Human connection is a scarce resource in the AI era.