[Retric]

You can make a prediction about what not how. Set things up and E Coli will develop antibiotic immunity, repeat the experiment 100 billion times and you might see 1+ billion different solutions. Insert fragments of a solution and repeat the experiment and you increase the odds to get the same solution, but the original solution space was still vast.

Consider, we say cancer mutates to disable specific genes, but not how those genes are disabled because it's outcomes not methods that are so common.

[AllegedAlec]

> repeat the experiment 100 billion times and you might see 1+ billion different solutions

You won't, though. There has been some excellent research on this done by amongst others Burmeister and Meyer in the Lenski group. Evolution is 'channelled' into certain solutions by its previous history and the process of co-evolving with other species.

[Retric]

That simply means some solutions are more probable than others, but there is a very long tail. Or at least as extrapolated from computer simulations.

[AllegedAlec]

Sure. It's still a game of chance. However, evolution is not random in the purest sense. Evolution itself skews the distribution of the kind of mutations you see.

For example: on of the papers by the Lenski labs show that the lambda phage evolves a certain defense mechanism very reliably in duplicated studies, despite it requiring several very specific mutations.

[Retric]

Random does not mean a uniform distribution. Toss 2 six sided dice and 2's and 12's are uncommon. But it's still a random process.

Also, saying you often get some set of mutations may be likely if subsets of those mutations provide advantages or there is some process that makes those mutations more likely.

[AllegedAlec]

> Random does not mean a uniform distribution. Toss 2 six sided dice and 2's and 12's are uncommon. But it's still a random process.

True, but dice don't adjust their own shape so that they're more likely to stop at certain values.

> Also, saying you often get some set of mutations may be likely if subsets of those mutations provide advantages or there is some process that makes those mutations more likely.

Yeah, but doesn't take away from the fact that evolution is canalised into certain options (most of the time) by its own evolutionary history.

[Retric]

I am not disagreeing with that. I am simply saying even if 95% of the time you see a small set of options the other 5% get's increasingly bizarre. Start grouping them and the last 0.001% case has just about anything.

So, sure you can specify odds, but again that's very different than saying something is not random and thus predictable.

PS: I tend to focus on black swan events because they tend to be more important. If you focus on the most likely outcome that becomes less important over time. M1 - M10,000 might each be happy paths, but M1 - M10,000 is is extremely unlikely to have all of them be happy paths.