An AI system that outsmarts every human in every task.
For
- No reason to think that machines can treach human brain (Representationalism)
- Current systems are already quite close to human level intellect
- Even if we dont have the capabilities right now, we are pouring so much funding in AI R&D
- AI is far better at non-human suited tasks (Can reason about Quantum Mechanics)
- Recursive Self Improvement → Singularity
- Even if we cant get ASI, HLAI can still have qualitative advantage over humans:
- LLMs are knowledge bases
- AI can do cognitive tasks much faster than we can
- AI can outnumber us and easy to make copies of itself
Against
- LLM systems still miss something (No breakthroughs, continual learning)
- AGI requires fundamentally different ideas (not blind scaling)
- As science and tech improves, ideas are harder to find: No New Ideas