What is intelligence, really?

Before we can talk about human intelligence, AI, AGI or ASI, we have to agree on what intelligence actually is. Most arguments about AI go nowhere because two people are using the word differently.

01My definition

Intelligence is the ability to conceive a problem and solve it. The more complex the problem, the higher the intelligence.

Two things follow from that.

Everyone has gaps. Someone brilliant in one field can be ordinary in another. A great surgeon can be hopeless with money. Be strong in a few fields and you're a polymath. Be genuinely intelligent in every field? Nobody is. That person doesn't exist.
The best intelligence crosses over. The rarest kind takes what it knows in one field and uses it to crack a problem in another. That's the person who seems oddly comfortable with problems they have no training for. It's also the person who solves the old problem in their own field that stayed unsolved because the gap in knowledge was too wide for anyone else to bridge.

Hold on to that second point. It's the dimension where today's AI is weakest, and it's the line between AI and AGI in the next articles.

02How the scientists see it

My definition isn't the only one. In 2007, researchers Shane Legg and Marcus Hutter collected more than 70 published definitions of intelligence. Four views matter most:

Four views of intelligence

One general abilityIn 1904 Charles Spearman noticed that people who do well on one kind of test tend to do well on others. He called the shared factor g. IQ tests try to measure it.
Many separate abilitiesIn 1983 Howard Gardner argued for several intelligences: language, logic, music, space, body, people, self. Popular, and still debated.
Reaching goals anywhereLegg and Hutter boiled their 70+ definitions down to one line: intelligence measures an agent's ability to achieve goals in a wide range of environments.
Learning new things fastIn 2019 François Chollet argued that real intelligence is how efficiently you pick up new skills, not how many skills you already have.

My definition sits closest to the last two. Solving problems is reaching goals. Solving new problems, especially by borrowing from another field, is Chollet's learning efficiency in action.

03What intelligence is not

Not knowledge. An encyclopedia knows a lot and solves nothing.
Not speed. A calculator multiplies faster than you and still can't conceive a single problem.
Not consciousness. Whether something feels is a separate question. I come back to it in Can AI feel?
Not wisdom. Being good at solving problems says nothing about whether you picked good problems.

That last one is the seed of everything that follows. A mind that can solve almost any problem is powerful whatever it decides to work on. So what it decides to work on becomes the most important question there is.

04The short version

Intelligence: the ability to conceive and solve problems. Harder problems, higher intelligence.
Everyone has intelligence gaps. Nobody is intelligent in every field.
The rarest skill is moving knowledge from one field into another.

05Sources

Legg, S. and Hutter, M. (2007). A Collection of Definitions of Intelligence. arXiv:0706.3639.
Legg, S. and Hutter, M. (2007). Universal Intelligence: A Definition of Machine Intelligence. Minds and Machines 17.
Spearman, C. (1904). "General Intelligence," Objectively Determined and Measured. American Journal of Psychology 15.
Gardner, H. (1983). Frames of Mind: The Theory of Multiple Intelligences.
Chollet, F. (2019). On the Measure of Intelligence. arXiv:1911.01547.

All posts

Appointment setting3

Copywriting3

AI services2

Deep Think AI13

Get the next markup by email.

One real piece of copy per issue, rewritten line by line: a cold email, a web page or a newsletter.