Evolution "Doesn't Need" Mutation. Here's Proof. - Blaise Agüera y Arcas [video]

8 minutes into the video yet again a physicist who has no idea what a Turing machine is, he claims a Turing machine is like a laptop that cannot self replicate because its tape is symbolic and a von Neumann machine is more like a 3D printer since the memory grid corresponds to a space of atoms.

This is so insane. Every physicist and every other AI speaker needs to please take an actual class or two in 1) computability, and 2) complexity. As in actually read through Sipser or other equivalent textbook and do the homework problems and pass an exam.

To see public speakers openly abuse Alan Turing's own theories/arguments is so offensive and wrong. They would not do this in their own field be it physics or genetics.

For posterity, here's a critical take by YT user @DrinkColdWater on the claims made in the presentation:

LOL that’s not random.

What he’s calling “random code generation” is basically random sampling of inputs fed into a hand-designed, rule heavy machine that already contains the ingredients for replication. That’s not “magic from noise,” that’s a biased dynamical system doing exactly what its rules make likely.

The hype he try to make of it is insane...

The generator isn’t random. The only random part is “which two tapes did we pick this round?” After that, everything is a fixed interpreter with fixed semantics. That’s deterministic structure, not randomness.

And he baked in copying. If your “physics” includes primitives that can move/copy symbols and you allow self-modification, then replication isn’t a miracle it’s an attractor you engineered the system to have. You didn’t “discover life,” you built a toy world where copying is one of the easiest stable behaviors.

There’s also selection even if he acts like there isn’t. Finite soup + overwriting means anything that copies itself (or its buddies) becomes overrepresented. That’s implicit selection. Once selection exists, the system stops looking random very quickly.

So no, he didn’t show “randomness becomes programs.” He showed that random encounters + a carefully chosen computational substrate + implicit selection will concentrate self-replicating patterns. That’s cool as a demo, but it’s not the metaphysical mic drop he sells it as.

Also, “no mutation” doesn’t mean “no novelty.” He still has randomness via random encounters and a huge recombination operator (concatenate + execute + split). So the “wow, novelty without mutation” pitch is misleading he removed one novelty channel while keeping another massive one.

And compressibility is not complexity. His entropy by zipping plot mostly proves “lots of repeated stuff showed up.” That’s redundancy. Redundancy can coexist with function, but it’s not automatically “increasing complexity.” A soup dominated by a few copied strings becomes very compressible even if it’s dumb.

The “phase transition” doesn’t prove abiogenesis. Sharp transitions are common in threshold systems with feedback (epidemics, percolation, branching processes). If you set up replication + mixing + selection, you should expect a tipping point when copy capable fragments become common enough.

Honestly, if anything this suggests the opposite moral: getting “life-like complexity” out of “non-life” requires an insanely specific set of built-in constraints and mechanisms. Without those carefully engineered rules, the default outcome is just drift toward junk, not a spontaneous climb toward organized function.

Another user writes This talk feels like an example of believing in your own hubris a bit too much, and I think this nails why I'm instinctively repelled by this guy. He comes across as egocentric and smug, and his jokes are 12-yo level. To be fair, he says that much himself.

Also, when someone promises to show you "magic", expect that the rabbit was there all the time, it didn't just materialize in the hat.