🖥 "Will AI Systems Truly Do What We Really Want?" — Dario Amodei
In a recent interview with Lex Fridman, Anthropic co-founder and CEO Dario Amodei talked about when artificial intelligence would become smarter than humans and take over scientific laboratories.
Highlights
➡️ In mathematics, physics, and biology, AI is already at a PhD level and might soon surpass the highest human professional standards.
➡️ Is it realistic for AI to become smarter than humans? It depends on the field. Some areas may have ceilings that are very close to what humans have done. If I look at an area like biology, there's a lot of room at the top for AIs to get smarter. Those ceilings may be very far away in other areas, like resolving conflicts between people.
➡️ Powerful AI (AGI) may appear as early as 2026 or 2027. But there could be a mild delay. For example, we could run out of data, we will not be able to scale computing power, or the models will stop improving, and we will have to invent a new architecture.
➡️ In the early stages, AGI will be like a graduate student. You assign him a project, and he will read all the literature, write the code, order the equipment, conduct the experiment, perhaps along with a human lab assistant. Ultimately, the AI will be smarter than the professors supervising it and start managing the research itself.
➡️ I do not rule out that powerful AI can get out of control or go "to the dark side." Are the AI systems going to do what we really want them to do? One of the main risks is a catastrophic misuse of the models in domains like cyber, bio, radiological, and nuclear, things that could harm or even kill thousands, even millions of people if they go wrong. And so every time we have a new model, we test it for its capability of doing tasks autonomously.
➡️ Barriers to progress have always been there. But eventually, change happens, and usually thanks to a few people. I think this will be the case with the development of AI. Barriers will begin to collapse gradually, and then they fall all at once.
❕ Anthropic is the developer of chatbot Claude and one of OpenAI's main competitors. Amazon and Google are among the investors.
📱 You can watch the full interview here.
More on the topic:
💩 Dario Amodei: how AI could transform the world for the better
#interview #Claude @hiaimediaen