Do we understand intelligence enough to formalize it in mathematical or computer science terms? We don’t, because otherwise there’d be no need for AI benchmarking. In fact, the concept of intelligence is remarkably fuzzy.
- How well does the correlation between higher intelligence and positive life outcomes hold across the intelligence distribution? How strong is the evidence for the correlation holding beyond two standard deviations above the mean (IQ > 130)?
- Is there an upper limit to intelligence? How far are humans from it?
- Is intelligence a meaningful concept outside of the range observed in humans?
- Are there problems for which intelligence isn’t helpful?
- Is there a level of intelligence that can be achieved by AI but that would be incomprehensible to any human? This is what Scott Alexander asks on Astral Codex Ten
- If intelligence is related to pattern recognition, why isn’t it more trainable in humans?
- When taking an intelligence test, there’s a time limit. The reason is that we believe the ability to think quickly is one of the hallmarks of intelligence. But what exactly is the relationship between speed and intelligence? Which problems can be solved with limited intelligence if given unlimited time, and for which problems is time not enough?
- Assume a problem that can’t be parallelized. What’s the relationship between the number of people (or more generally, agents) working on it and the time to solution? This is assuming that all people or agents are similarly intelligent. Does the answer depend on the type of problem?
- What are the evolutionary tradeoffs with intelligence? Is it just the metabolic cost of a big brain? Is the fact that intelligence isn’t universally selected for in animals a sign of its limited usefulness?
- Why has the Flynn effect seemingly stalled?
More questions about: Cave art | Domestication | Appearance
Nehaveigur 