"Proposition 3: memes — not programs — co-evolve with human programmers"

Looking for human-like intelligence in algorithms gets the level of abstraction wrong.

If we want to see software systems behaving like the cunning, brutal, creative creatures of the animal kingdom (like us) then we should look at computer programs as memeplexes.

And we should be looking at the behavior of individual memes.

In this view looking at functional modules is obviously wrong: any functional implementation is only a phenotype (an “instance”) of a particular meme. The meme itself may be expressed in any number of concrete code patterns. In any Turing-complete language there are technically an infinite number of ways to express the same functional pattern. But behind all those phenotypes we may identify a single meme.

To study the phenotypes and miss the meme is basically what structuralist approaches have been trying to do for the last 40 years. That does not work.