Three Perspectives on AI

Perspectives on AI diverge based on how seriously one takes AI, AGI, and ASI.

• AI camp: Acknowledges what AI can actually do today.
• AGI camp: Believes AI will be able to do much more in the future.
• ASI camp: Believes that within a natural lifespan, AI will be able to do almost everything better than humans.

The author leans toward the ASI camp, accepting all three, and explains that many frontier AI labs and their employees share this view.

Most people have not even fully experienced the first stage. They use ChatGPT only for simple tasks and then generalize past failures as current AI limitations, or dismiss AI as useless without knowing about recent features like coding agents.

Even with current AI, tasks previously handled manually can be done more easily, and work or questions that were previously not attempted due to cost can now be tried. The argument is that a transformation comparable to the internet has already begun as the marginal cost of making a single request to AI approaches zero.

Policymakers and economists generally accept only the first AI camp and even underestimate AI's impact. They judge AI as uncompetitive based on current prices without considering the possibility that usage costs will drop significantly soon, or assume that tasks AI currently performs poorly will not improve in the future.

The reason the AI debate remains stuck on issues that are already resolved is that many people fail to accept the first camp. After acknowledging AI's current capabilities, one should look into improving usage methods and so-called 'unhobbling', and then move on to the AGI camp, which considers the capabilities of future AI.