In Iain M. Banks’ Culture books, the Culture Minds are AIs possessing an intelligence far surpassing our own. Many people have speculated that we might one day meet ultra-intelligent aliens, or that our own descendants might be genetically-engineered to have very high IQs. So what would it be like to be an ultra-intelligent entity and is ultra-intelligence really going to happen?

The Measurement of Intelligence

We measure intelligence by a battery of IQ tests and then we describe how intelligent a person is by stating their IQ: this may sound kind of circular to you. In fact the concept of IQ is a statistical one. Historically psychometricians gave representative groups of people batteries of tests covering many different aptitudes: visual, verbal, arithmetic, spatial imagery, reading, vocabulary, memory, general knowledge and so on. They then looked at the matrix of correlations between results to see whether there was an underlying general intelligence factor (the ‘g’ factor) underlying success across these diverse tasks (the mathematics is called factor analysis). It turns out that there is a common factor, and this conclusion proved robust over many different kinds of task.

Specific tests were then devised where success would correlate very highly with this ‘g’ factor – and thus were born ‘intelligence tests’ and the concept of IQ. Now, very few people would care about this if IQ had no practical consequences: however, many, many studies have shown that IQ is a strong predictor of success in areas where a commonsense view is that high intelligence is required. Examples include corporate and military leadership, professional occupations (medical, legal, finance, consultancy, engineering), science and research.

Empirical research, especially twin studies has shown that IQ is strongly hereditable. This means that intelligent parents tend to have intelligent children. Some people originally thought that there might be just a few genes coding for high or low IQ but it now appears that there are hundreds. As IQ is distributed in populations normally (i.e. in a bell-shaped curve) this is just what the central-limit theorem would lead you expect [1]. IQs are normalised with mean μ = 100 and standard deviation σ = 15 for Western European populations.

What is Intelligence?

So what are IQ tests actually measuring? We are all familiar with them and people have even written AI programs to take IQ tests with success, so it’s hardly a mystery. In fact intelligence as measured by IQ tests measures three things.

1. Cognitive Speed
Speed in acquiring, processing and responding to information. This corresponds to doing a sufficient number of questions in the time allotted.

2. Capacity for Abstraction
This is the creative ability to find the right abstraction, the right way to ‘look at the problem’ so as to make it solvable. In a test, this might correspond to realising that you have to count the dots in one corner and the lines in the other corner of each box and see how they are different in the different boxes in front of you to solve the problem.

3. Capacity for Reasoning
This is the ability to reason abstractly to the correct conclusions. In a test, this might correspond to looking at the three boxes on the page, identifying the underlying rules (from the Abstraction step) and then applying them to find the correct fourth box from the list of options underneath.

It should be clear that many practical problems in our complex civilization look like this. It’s equally clear that such skills as common sense, ability to persuade others and a general empathy and likeability all leverage intelligence but are not reducible to it. This explains why our brightest people (probably mathematicians and string theorists in IQ terms) do not actually rule the world.

The Evolution of Intelligence

Almost all the creatures which manage to survive very well on earth are not very intelligent (houseflies anyone?). Nature does not like general purpose intelligence. The human brain consumes 20% of the energy available to the body although it’s only 2% by weight: if there was a way to recode what the human brain is doing into a series of low-level reflexes in a smaller brain then that option would have been selected for.

Big, smart brains support a generalist strategy, optimal only when an ecological niche keeps changing sufficiently rapidly that the evolution of specialist hardwired reflexes can’t keep up. In human evolution, highly variable environmental conditions in the Pleistocene, human migrations and the development of more complex societies through agriculture in the Holocene have all played their part. Creating complex and dynamic societies is one of the key drivers of intelligence: consider the intellectual demands of being a senior executive in a major corporation or a developer of cloud-computing architectures.

Meeting the Ultra-intelligent

Suppose we imagine an entity with an IQ massively higher than anything on earth today: an ‘ultra-intelligence’ (UI). Perhaps it’s an AI system we will eventually get around to designing; perhaps it’s someone with extremely bright parents who got very lucky in meiosis and had a helpful mutation or three; or perhaps we finally meet the super-smart aliens.

We would expect a suitably-educated UI to have at their disposal a very large set of well-interconnected abstract concepts and be able to use them to frame situations and draw consequences very, very quickly. In dealing with the UI you would feel that it understood whatever you were talking about far better than you. As you started a sentence, you would see a flicker of interest in its eyes which would turn to boredom within a second as it predicted exactly what you were going to say. The UI, in dealing with us, would spend a lot of its time bored out of its mind.

If you were at a UI meeting or party you would be humiliated. Conversations would start, develop and complete before you had figured out what was going on. Allusion would follow allusion and you would always be a couple of steps behind. You would strongly identify with any dog that happened to be present at one of your own parties.

Evolving to Ultra-Intelligence

Is the human race on its way to evolving ultra-intelligence through natural selection? For that to happen two factors would have to co-occur.

Firstly, there has to be a mismatch between the present distribution of intelligence in the human population and the ecological niche in which we find ourselves. To put it bluntly, those individuals endowed with greater IQs should be more successful in life than the less smart. In an advanced capitalist country it seems clear that life success, measured by status of career, is indeed correlated with higher IQ: examine the chart below where the vertical lines within the horizontal bars show the 25th, 50th and 75th IQ percentiles (the bars extend from 10% to 90%) for various occupations.

The second factor in evolving ever-greater intelligence is that the consequence of greater life-success should translate into greater reproductive success. Again, to put it bluntly, those of lower IQ would have to be disproportionately failing to reproduce, causing those genes to vanish over time from the population, replaced by genes coding for higher intelligence. Given that modern states do not leave adults or their children to die, even if they are not that successful in life, this second driver for greater IQ over time is probably not present today.

However, artificial selection could substitute if the technical and social barriers to genetic engineering for intelligence were to be overcome. In the longer term, this may be the mechanism by which the human race becomes ultra-intelligent.

But perhaps there is a limit to how smart most people need to be even in a super-advanced technological civilization? Not one person in ten thousand understands those principles of General Relativity which underlie the Global Positioning System. But the practical understanding of these theories is incorporated into your smartphone and you certainly don’t have to be a genius to use it (merely quite smart).

Some of Iain M. Banks’ Culture Minds found the mundane realities of life, even in such an advanced society as The Culture, to be quite tedious. They preferred to spend their time in the virtual reality of ‘Infinite Fun Space’ where there were no bounds to interesting complexity. That may be the future of human (and alien) ultra-intelligence: a retreat into ever-so-much-more-interesting virtual realities with some smart military and engineering types left in mundane reality to keep the computing substrates safe and build the new ones [2].

It may not be safe to dumb-down quite yet.

Notes

[1] If a large number of separate events conspire together to produce an outcome, we expect to get a binomial/normal distribution. For example, if you throw 100 coins over and over again and plot how many heads you get each time, the results will cluster around 50 but will fall away either side in a bell-shaped curve. Each gene which has a hand in determining IQ is like one of those coins.
[2] As so often happens, Greg Egan got there first.

Further Reading

1. “The 10,000 Year Explosion: How Civilization Accelerated Human Evolution”, G. Cochran & H. Harpending, Basic Books, 2009.
2. “The Limits of Intelligence”, Douglas Fox, Scientific American, July 2011. This featured SciAM article speculates that humans might already be close to the biological limits of brain size and complexity, thus limiting the potential for further increases in intelligence. Some of the barriers, such as the noise-limited narrowness of axons, seem hard to fix in the current neural architecture; others, such as the energy cost of more neurons, seem softer. Adjunct brain-modules using advanced technologies might also work.