From what I can see, active information is the probability of getting a beneficial mutation. Per the usual, the major pitfalls are not knowing all possible beneficial mutations for a given selective pressure and the potential for committing the Sharpshooter fallacy.
The cit+ mutation in the Lenski experiment could be good example to work from. We simply don’t know all of the possible mutations that could have conferred the cit+ phenotype. There are probably many different promoters and recombination events that could have worked in this situation, not to mention the possibility of beneficial substitution mutations within the original promoter. I would also be curious to see how the E. coli SOS response could affect the results in this specific case.
The studies leading to the discovery of the SOS response are especially cogent to the overall thrust of Bartlett’s paper. Some of the earliest arguments for a possible mechanism of adaptive mutations was based on higher than expected lac+ revertants in E. coli. As it turned out, when starvation caused DNA damage this turned on several genes, including genes responsible for increasing the substitution and recombination mutation rates. The E. coli weren’t specifically creating mutations that fixed a broken lac gene, but were instead increasing their overall random mutation rate.
My understanding is that “active information” is the difference between the probability of getting a beneficial mutation completely randomly vs the probability of getting that same beneficial mutation in the presence of cellular corrective machinery and bias towards particular types of mutation.
I can give you an example of such “active information” in a genetic algorithm, if you like.
I suspect that ultimately Bartlett’s ideas are irrelevant, because even if he does manage to show that there is “active information” in genomes or cellular processes that lead to more beneficial mutations that would be expected by chance, that will do absolutely nothing to show that that “active information” isn’t the result of 4 billion years of evolution.
All of the “information” arguments from creationists are nonsense.
Information doesn’t actually exist. That is to say, it is not a natural kind in any reasonable sense. Information is a human abstraction, invented by humans for their own theorizing about the world.
If you see agency involved in the use of information, then that is correct. But it is human agency.
Evolution gave us biological creatures and causal processes. It did not give us information. That you see information in biology, is because we humans invented the idea of information and found it a useful way (useful for us) to study natural processes.
So yes, information comes from intelligent design – the intelligence is human, and the design is in designing ways for us to talk about the natural world.
And the same thing happens in somatic hypermutation in the antibody response. The side effect is that many lymphomas are started by its lack of specificity producing mutations in genes expressed at high levels at that point in B-cell maturation.
Having read the definition of active information in the paper now, I really don’t think it’s bad at all. It seems like a sensible definition to me, and I can even think of examples of how it could be measured in limited situations. Given the definition, it is also demonstrable that active information can itself evolve(and I could give easy to understand examples).
The key point going forward will be any putative claim or argument that active information present in some system could not have evolved. As such, the real problem is not with the definition of active information itself, but with the claim in the abstract that “Active information can be very useful in differentiating between mutational adaptations which are based on internally-coded information and those which are the results of happenstance.”
-because it’s not clear how one would distinguish active information that evolved from active information that did not.
The Active Information (AI) is equivalent to a Likelihood Ratio test. The implication is that most statistical tests are also “teleonomically aligned”. The implication for your everyday T-test are suddenly quite profound!
Continuing on …
An obvious flaw here: Bartlett is evaluating the data (the result of search) and not the algorithm. The information in a particular algorithm, say Newton-Raphson for example, is always the same no matter what function it is applied to.
I’m sure he can do the math, but his interpretation of that math is highly questionable.
Well the “active information tutorial” on page 2 seems pretty straightforward to me. The real question when it comes to actual biology is then going to be what exactly counts as a “blind random search”, what Bartlett calls IΩ.
Supposing we just talk about random substitution constrained to a very small locus, we could imagine that we measure the fitness effects of all single-nucleotide substitutions in that locus. If there is a biochemical bias in the process of substitution (which of course we know there is), is that then going to count as part of the “search under analysis”(IS), or is that the “pure random search”(IΩ)?
If it’s part of the “search under analysis”(IS), then the absence of substitution bias must be what is meant by a “pure random search”(IΩ). That should then imply, if I am to make sense of the definition, equiprobable substitutions. That all transitions and transversions are equally probable. Thus we could determine if any bias in the substitution process contributes active information to evolution of the locus, if such a bias increases the probability of beneficial mutation.
Supposing it does, we also know biochemical reasons for why this bias exists, and that these biases themselves can be changed by mutationally altering either the enzymes(or their regulation) that cause or deal with them, either directly or indirectly, and thus we know that active information putatively present in the system and responsible for causing substitution bias, is itself an evolvable attribute.
The SOS response is a global increase in mutation rate, true. But there are some other stress responses in bacteria that lead to increased mutation rates that are not strictly global but are “focused” on highly expressed genes and in some other ways. These are not “guided” mutations, and it is IMO misleading to describe nonuniform mutations patterns as “non-random.” But mere upregulation of global mutation rate is just the start of mutation regulation. Cf. Susan Rosenberg’s work and nice review below.
The more I think about this, the less sense it makes. A deterministic search algorithm will find the target or it will not. The only random factor is the starting point in the search space, and possibly the search space itself if you want to choose one randomly. We can add searches that allow randomness (GA, simulated annealing) which will create a region of search space where the probability of finding a target is strictly between 0.0 and 1.0, but this too will depend on the search space.
I don’t think this is a search problem at all, but rather an optimization problem. The algorithm climbs the hill to the top and stops. Algorithms with randomness might be able to avoid local maximums and find better solutions than deterministic ones. Probability of finding the target is simply the wrong question - It’s a matter of algorithm efficiency, and the best solution that can be found in the available time. The optimal solution isn’t even relevant, it only need to be “good enough”.
Would a GA example suffice, or do you want a biological one? I mentally worked out the “active information” in a ‘weasel’ variant yesterday. (Which is more than Bartlett has done).
I don’t have one of those, unfortunately. I’m not sure it’s even possible. The mutation probabilities vary depending on the nature of the cellular processes that perform the copying and error-correction, so there’s no way to establish a meaningful “pure random search” probability (IΩ).
Bartlett’s mathematical definition is well defined - the difference in effectiveness of two search algorithms based on the ratio of their probability of success. It’s when he tries to apply this to biology that it becomes muddled, and he basically starts making up claims such as “Measuring active information is measuring the information that the genome (as it presently stands) has about likely beneficial future configurations” or “What active information measures is the alignment of the genome itself to the problem of finding viable genetic solutions to selection pressures”, neither of which follows from the definition.