When considering the idea of a heuristic,
an AI algorithm that makes a decision after limited search,
it is productive (and also often funny) to consider one of the most fundamental heuristics in human and animal life:
the heuristic for mate choice.
When you choose your mate, you do not do an optimal, exhaustive search.
You use a heuristic.
Can't try out 1 billion partners.
Can't wait 80 years.
Have to make decision in limited time
after limited experience.
Mate choice heuristic
in birds:
Mating dances.
Usually the male dances and the female selects. This is inherently funny.
So the question is why this evolved.
Assuming the concept of dancing makes any sense, consider the female's choice algorithm.
She
cannot watch dances forever. She must make a decision.
Simple heuristic: n and f
Simple heuristic:
"Commit to first solution you find".
Obviously a poor search strategy
(you could meet the best first, but unlikely).
Simple heuristic:
"Date (or observe) for some time
but don't commit. Then switch mode.
Commit to next really good partner that comes along."
How long to wait before switch mode may vary by species.
"Until last week of mating season" for some species.
"Until I make partner" for some humans.
n = no. of samples before switch to "commit to next" mode.
f = tolerable fitness to stop with.
n and f could vary.
f probably evolves over time
- need to date a few n in order to have an accurate f.
For normal humans
it might be something like
n=4 and f = 80 percent.
With "normal" values of n and f, this heuristic
avoids committing too early.
But won't sample forever, avoiding good solutions until it is too late.
Will this heuristic guarantee finding best partner you
could possibly meet? No.
Won't meet the best possible partner on the planet because n is small finite.
Might even lose the best partner you do meet
if you met them before n.
Discussion
In the above mate choice algorithm, what would the following look like?
n=0 and f = 0 percent
n=1000 and f = 99 percent.
n=0 and f = 99 percent.
n=1000 and f = 0 percent.
Reading
"Emergent Patterns of Mate Choice in Human Populations".
Jorge Simao and Peter M. Todd,
Artificial Life 9(4): 403-417 (2003).
The search algorithm logic is a heuristic.
And f, the evaluation function, is also a heuristic.
Often we find it hard to judge f, as much fiction is based on.
Mr. Darcy
explains to Elizabeth that she is in at least the top 100 million people for him worldwide.
Maybe even the top 10 million.
Image from here.
(Mr. Darcy did not actually say this. I am making a joke.
But his first marriage proposal was
in fact very rude.)
Dating apps
You would think computers could really help with mate search.
But
dating apps and sites
have run into all sorts of problems when they encounter the facts of evolutionary biology.
Too much choice in mate choice (the Internet, the big city)
can be even worse than too little choice (most human life before the modern era).
"How A Math Genius Hacked OkCupid To Find True Love".
Wired, Jan 2014.
UCLA student devises algorithm to generate dating profiles that get more responses.
Algorithm fails because it is focused on getting responses
not on being compatible.
He ends up going on 87 dates with incompatible people.
Computer simulation
"Why Men Get So Few Matches on Dating Apps".
A fascinating computer simulation
from Memeable Data,
13 July 2023.
Evolution explains why dating apps are often a heuristic gone wrong,
that will not work.
Here is the basic logic:
Dating apps run into the problem of evolutionary biology:
It is low risk for males to "like" or meet stranger females.
It is high risk (or at least higher risk) for females to "like" or meet stranger males.
The app cannot change these facts of evolutionary biology, which affect all behaviour.
The computer simulation then demonstrates why:
Average males get few or no matches. Even though many females would like them.
Average females get too many matches.
Too many to talk to.
Too many weirdos.
And too many matches with a small group of high-competition males
who the females are unlikely to meet.
The app steers them away from males that would work for them.
It ends up not working for either sex.
The computer simulation
shows how hard it is for the app not to have these problems.
Asymmetry gets magnified:
One thing that happens is that any asymmetry (built-in by biology) gets magnified not damped down.
Average males get less likes, so they change behaviour and start "liking" more females.
But then females are bombarded, and start reacting to likes with less and less enthusiasm.
The initially flawed algorithm provokes behaviour that makes it even worse.
Ancient Brain
