Sunday, December 27, 2015

What Exactly Is Intelligence?

Intelligence is a word thrown around everywhere, without all that much definition or explication. In general, things can be defined by an operational test. For intelligence, one could come up with some test to give to proposed intelligent creatures, and if they achieved some score, on what is called a metric, they would be classed as intelligent. One you could define intelligence as a continuum variable, instead of a binary choice, as say it is the score that some creature gets on a standard test. Quibbles about what test to use can be endless, futile and meaningless, and again, there can be some pretty stupid choices made for intelligence tests.

One nonsensical test was called out in a previous blog. The memory test. As if being able to remember labels for things requires intelligence. If this was used, any computer would be a genius. The kind of questions people who should not be allowed to use the word intelligence come up with sound like this:
1) What birds fly from Canada to Venezuela each year?
2) Who was the sixteenth secretary of state for California?
3) Where is the source of the Nile river?
4) Name fifteen named asteroids?
5) Which author wrote ‘The Old Man and the Sea’?
6) Where is Tasmania?
7) Name the dynasties who ruled China?
8) How far is Alpha Centauri alpha from Earth?
9) What is the largest city on the African continent in population?
10) Which element conducts electricity the best if purified?
These questions might merit a gasp from someone who does not think much about what intelligence should be and how to measure it. Each of them requires nothing more than a database look-up. Real intelligence testing would require problem solving, and depending on the variant of the definition, might require more media than words alone.

Metrics measure intelligence or anything else, and decide if something deserves a label or alternately, attach a number to the thing being tested. This creature has an IQ of 100 is an example of the quantitiative aspect of testing. Metrics can be used for black boxes, and there may be some use for them. However, if you want to build a good metric and a good test for measuring it, you should understand what it is that you are measuring. Creating something because it sounds good may impress the inexperienced, but good metrics come from a deep understanding of the processes that go on when the testing is done, and when the attribute being measured is being used.

To be more explicit, if you don’t know how intelligence works or what it does, you cannot come up with a good test for it. You can certainly come up with a hundred tests for it, and quibble forever about which one to use. But until intelligence is described not by a metric or a test, but in terms of a process, it cannot be well measured.

Simple things don’t require much knowledge of the process. If you have a thing you found in a kitchen, and you want to test to see if it is a toaster, you can put some bread into it, plug it in and turn it on, and see if toast comes out. That’s because you have defined the output, toast, as bread which was heated on both sides until it desiccates a bit. What do you do when you find some creature in an alien starship crashed here on Earth, and you want to find out if it is intelligent? You don’t. That isn’t a good question.

What is the purpose of labeling some alien creature, which might be the pilot or might be his pet, as intelligent? It could be useful in the media. It could be useful in communicating with it, unless it was the alien equivalent of a parrot. What you need to do is determine what you want to do with the alien, and test for that. It is the same for a child here on Earth. You need to determine what it is you would like the child to be able to do, and then test for that. Give it problems to solve.

Back to the details of intelligence. Intelligence, as a process, is the processing and interpretation of data, and certainly not developing a one-to-one correspondence between a description and a label. It is the ability to reduce data to a model, to extract trends from data, to see analogs and relationships between different parts of the data, to see structure in the data, to create structure out of the data so it is useful, to find differences between near identical sets of data, to filter out meaning from data, to connect multiple separated blocks of data together to create something coherent, to see where blocks of data have natural divisions or natural trends, to interpret it in a one, two, three or more dimensional way, and many other things. Intelligence has the ability to determine from a collection of data in a context, just what to do to it, choosing from among the previous list or from a larger subset.

Intelligence at higher levels means more sophisticated processing of blocks of data. Testing for this can be done at an elemental level, like finding some pattern in a large body of data, or on a more complex level, like determining the best way to describe a data set or to model it in as few elemental constructs as is possible in order to capture the main variation in it. Intelligence is certainly quantitative, and not much can be done without quantitative skills. Pattern finding does not require arithmetic type skills, but pattern finding is still a quantitative exercise, but of a different type. The word quantitative often is used to connote arithmetic aspects, when it should instead be used to connote anything more than verbal associations. Describing the structure of something is quantitative, even if the structure can be described in simple and discrete terms.

Thus, to measure intelligence in a proper way, for an alien or a human, requires coming up with a list of the ways in which data can be processed, and then deciding on how to measure the subject’s ability to do it. Recognizing the method to be used is often included as part of the challenge. There is no well-known intelligent intelligence test. Perhaps it is worth doing at this stage in Earth’s history.

No comments:

Post a Comment