Evolutionary foresight goes mainstream

The mainstream seems to be finally waking up to some of the idea I expressed in 2008 in my "evolution sees" essay. In particular, Richard Watson, Eörs Szathmáry and others have recently been arguing that evolution can learn. Not just with trial-and-error learning, but with real connectionist learning. Here are some of their papers:

• What can ecosystems learn? Expanding evolutionary ecology with learning theory
• How Can Evolution Learn?
• Evolutionary connectionism: algorithmic principles underlying the evolution of biological organiszation in evo-devo, evo-eco and evolutionary transitions

Though their conclusions seem similar to mine, their argument looks a bit different. They suggest that connectionist learning in ecological populations might be important. If that happens without animal nervous systems, that would be new and interesting. Even without this idea of theirs, my argument for the evolutionary significance of connectionism still stands.

Why cultural evolution is faster

So far most comparisons of the rate of cultural evolution - relative to the evolution of DNA-based creatures - have failed to control for the shorter generation times of memes compared to their hosts. A failure to do this leaves open the possibility that cultural evolution is faster than mammal evolution simply because memes have shorter generation times than mammals do. This is a problem with Charles Perreault's 2012 article on the topic, for example, as I pointed out at the time.

John Wilkins laid down the gauntlet on this issue in 2009, here:

the speed of cultural evolution is pretty much the same as the speed of biological evolution. The problem is that the “rate” is not absolute. Speed in evolution is always relative to generations, not to years. I feel that cultural evolution will tend to be roughly the same relative rate as biological, if only because error rates tend to be at or about the same general level in transmission processes.

However, intuitively it seems as though cultural evolution has led to nuclear power and moon landings in a short period of time - while organic evolution stumbled around for billions of years without doing these things. It seems as though empirically, cultural evolution really is faster, contra-Wilkins.

There's no shortage of theoretical candidates that hope to explain why cultural evolution is faster. In cultural evolution, mutations take place in brains, rather than cells. Variation can arise as a result of intelligent design, using interpolation and extrapolation. Evaluation can be performed under simulation. If you consider brains as containing Darwinian or quasi-Darwinian processes then the generation time there is even faster than cultural generation times - faster even than bacterial generation times.

One of my favorite candidates for a factor that explains the increased speed of cultural evolution involves the distinction between supervised learning and reinforcement learning. In supervised learning, mistakes are corrected by a supervisor, that presents the correct solution. A common alternative to this is reinforcement learning - where actions are scored by a reward function or a utility function. Mistakes are scored, but not corrected.

In organic evolution, organisms are mostly scored - in terms of reproductive success. However, bad quality actions are mostly not corrected. Supervised learning is associated with having a brain. Once you have a nervous system you can often learn using supervised learning. An important function of the brain is predicting consequences of actions - and the most common case of supervised learning involves predicting future sensory inputs. Here, after predictions relating to future sensory inputs are made, sensory inputs are actually received. The predicted and actual perceptions can be compared and supervised learning techniques can be used to correct any mistakes. Though there's no actual supervising agent, this qualifies as supervised learning - since the correct output is presented to the learner on every timestep.

Where does culture come in? Culture makes individual learning add up over the lifetimes of multiple individuals. Without culture, learned information is lost when the individual dies and their brain is eaten by worms. In principle, learned information can still get into the germ line via the Baldwin effect - but that is still a pretty slow process. With cultural transmission, information acquired using supervised learning can directly be passed down the generations and it can accumulate over time.

I won't lay out the case here for supervised learning being faster and better than reinforcement learning - that's part of machine learning folklore.

At first glance, this seems like an argument for psychological evolution being faster than DNA evolution. After all, brains have been around for 500 million years or so. They aren't a new phenomenon. So, how can supervised learning explain how it is only cultural evolution that seems faster? The idea is that psychological evolution is also faster, but that until the origin of culture, it couldn't really go anywhere. Without culture, information in brains is obliterated in every generation and can't really accumulate. As I mentioned, it can still affect evolutionary rates - via the Baldwin effect. No doubt the Baldwin effect does speed up evolution - but it does not do so as much as cultural transmission does.

I think these observations help to address some criticisms of cultural evolution. For example, here's Massimo:

The conclusion that biological and cultural evolution are different also nicely accounts for the fact that cultural evolution is so much more dynamic (it happens much faster) and unpredictable than its biological counterpart. If we think of both as instances of Darwinism that difference becomes more puzzling.

To start with, this is a bit of a straw man, practically nobody is saying that cultural evolution and biological evolution are exactly the same (excepting maybe Ben Cullen).

Anyway, according to the idea described on this page, one of the most significant differences is explained by learning theory - it is supervised learning. The impact of this on evolution dates back 500 million years or so - to the origin of brains. However before culture got started, its influence on the evolutionary process was somewhat muted - due to the regular destruction of information in brains in every generation.

Supervised learning represents a type of feedback from outcomes to the production of new variants to test. In the context of evolutionary theory, this is somewhat similar to Lamarckian inheritance: the inheritance of acquired characteristics. It is however important to note that this is not really a new effect - it dates back 500 million years. Nor can one say that such evolution is not "Darwinian" - since Darwin was an enthusiast for such feedbacks. Indeed, he promoted a theory of "gemmules" which featured feedback from somatic cells to the germ line. What we can say is that cultural evolution is not Weismannian. It features non-selective feedback from the phenotype to the genotype. However, we must also note that evolution in the organic realm is not really Weismannian either. Weismannian evolution - in which phenotypes only affect genotypes via selection and mate choice - was a product of Weismannian's choice of traits. He cut the tails off rats - and observed that this 'acquired characteristic' was not inherited. If Weismann had looked at rat fleas of rat pox instead he might have drawn the opposite conclusion - that acquired characteristics were inherited.

In computer science, lack of access to supervised learning helps to explain the limitations of genetic algorithms and genetic programming - and helps to explain why memetic algorithms are needed.

Darwin meets Turing

A cryptic title - but this post is about applying models of universal computation to universal Darwinism.

Most versions of universal Darwinism agree that evolutionary theory applies to brains and thinking. This idea was pioneered by B. F. Skinner and D. T. Campbell and promoted by W. H. Calvin, G. Cziko and G. Eldeman among others. Evolutionary theory explains all goodness of fit and all knowledge gain.

If evolution explains the operation of brains, it ought also to explain the operation of computers - since both are general purpose input-transformation-output learning systems. We have some nice, simple models of computation. Can universal computation illuminate Universal Darwinism? In this post we will find out.

We will use the NAND gate + interconnect model of parallel computation and see how it relates to evolutionary models. Copying is a primitive operation in Darwinism: in NAND land it corresponds to signal branching. Selection is another primitive operation in Darwinism: in NAND land, it corresponds to signal termination. That just leaves the NAND gate itself. The NAND gate takes two inputs and produces one output. There are two ways of looking at the NAND operation from a Darwinian perspective. One is as a conditional selection operation. A NAND gate obliterates or inverts one of its inputs depending on the value of the other one. The other way is as a merging or joining operation between two signals. That completes the relationship between these two models.

What did we learn from this exercise? Merging or joining operations turned out to be fundamental. Mutation was not fundamental. It turns out that you can model mutations using copying, selection and merging - if necessary.

Intuitively, the products of evolution include brains - so it is not surprising that some models of evolution are capable of computing partial recursive functions.

However, a universal model has some negative aspects. There's a sense in which universal models are capable of producing any output - and notoriously, models which predict everything are not very useful. We can take some consolation in the idea that there can be all kinds of differences between different universal systems - they differ in speed, degree of parallelism, memory to compute ratio, relative component costs, brittleness, support for synchronous operation - and so on.

One thing I learned from building this model is that my usual reply to critics who allege some Darwinian models lack predictive power is not completely satisfactory. I usually say that constraining the scope of the mutation operator is enough to limit the resulting predictions. However, if there's a recombination operator, that can also lead to universality - and produce a model that is compatible with a lot of observations. It looks as though mutation and recombination both need limiting.

This post presents a model on the level of the bit. Another way of building evolutionary models of computational processes is to rise above the level of the bit. Conventionally, most mutation and recombination takes place between genes - rather than bits - and genes are conventionally quite a bit bigger than bits. This path produces a range of interesting models which have already been well explored in some detail by genetic algorithm and genetic programming enthusiasts.

Is evolution a form of learning?

Some say that learning is a form of evolution - e.g. see my Keeping Darwin in mind essay. However others say that evolution is a form of learning. For example, here's Leslie Valiant (from Probably Approximately Correct) in a section titled "Evolution as a form of learning":

To see evolution as a form of learning we view the genome in evolution as corresponding to the hypothesis in learning. The performance of the genome corresponds to its expected closeness to ideal behaviour, where the expectation is taken over the distribution of experiences the world offers.

What is going on here? Is evolution a form of learning? Or is learning a form of evolution?

I think that this is a fairly easy question: learning is a form of evolution, but not all evolution is a form of learning. Some evolution is more like forgetting than learning. It represents a loss of adaptive fit. If you forget everything you ever knew, that's still a part of evolution, but it is hard to see it as a form of learning. So: evolution is the larger category, while learning is a subset of it.

Perhaps those who claim that evolution is a form of learning should start by making it clear that it is only adaptive evolution they are talking about. Then they would have a reasonable point.

Pemetic engineering

In order to develop engineers, evolution used:

• Organic inheritance (genes);
• Individual learning (pemes);
• Social learning (memes);

The engineers proceeded to develop memetic engineering and genetic engineering. However engineers also engineer individually-learned ideas, in a process that is rarely mentioned and doesn't have such a common name.

Here I'll refer to it as pemetic engineering. This is named after pemes (private memes), my preferred meme-like term for individually-learned ideas.

The topic has some overlap with rational thinking. It involves applying engineering to individual learning, and your own thoughts.

In this post, I'll make two basic points:

• The order the engineers approached these topics is the reverse of the order in which they were originally developed - i.e. memetic engineering preceded pemetic engineering which preceded genetic engineering.

• Pemetic engineering is interesting stuff.

The universality of language means that most ideas can be socially transmitted. However there's still a large role for individual learning. Riding a bike, for example, is maybe about 10% socially-learned and 90% individually-learned. Individual-learning is an important form of learning - and it represents a big and important topic.

Pemetic engineering involves about applying engineering techniques to individually-learned material. It's a fairly personal thing - and not very social. If something can be passed on to others, it's probably more a case of memetic engineering.

Although individually-learned ideas, by definition, have not yet been socially-transmitted, some of them are protomemes - and do go on to be transmitted socially - i.e. pemes can become memes.

Though pemes are not normally socially transmitted, pemetic engineering techniques can be. There's considerable overlap between pemetic engineering techniques and those used in memetic engineering.

Cultural variation visualized on a phylogenetic tree

The following diagram shows a two dimensional representation of an asexual species with the ability to learn exploring a fitness landscape.

Time is vertical, green bars represent DNA-based hosts, red bars represent ideas being copied inside their hosts' brains. Social learning is not represented in the diagram - for the sake of simplicity - so the red learned variants are destroyed when their associated hosts die, but the reader can imagine what social-transmission would look like, if it was illustrated on this diagram.

The learned variation alters the environment the hosts are selected in - and so influences their evolution. There is no attempt to illustrate this in the diagram, though.

The basic point in this post is to illustrate the similarity between this diagram, and plant roots:

The diagrams look similar because both represent evolutionary tress. In both cases, the ancestors are neared to the root than the descendants.

Indeed, the resemblance is closer than it may at first appear - since the plant roots are hosts to mycorrhizal fungi. So, in both cases, there's a host and small, rapidly-reproducing symbionts. The symbionts explore the surrounding space, and help to determine the path that their host takes through that space.

In my book on memetics, I make a somewhat-related comparison, comparing brains with tree root nodules.

Lemes

The incorporation of individual learning into memetics appears to precipitate a crisis. Individual learning and social learning are very similar, interact deeply and coevolve. Placing a dividing line between individual and social learning is pretty unnatural and difficult - because of the scale of the interactions between these two areas.

What makes more sense scientifically is a unified theory of idea and memory evolution - which handles how ideas and memories evolve both within minds and between minds.

I go into this issue in previous posts:

• The case for private culture;
• The case for private memes;
• Individual learning in memetics;

In the context of the existing memetics paradigm, there are various ways of dealing with this:

• Enlarge the definition of a meme to include individual learning;
• Enlarge the definition of culture to include individual learning;
• Create new units to deal with all learned information.
• Do nothing - try and muddle through with the existing concepts.

I discussed the possibility of enlarging the definition of culture in The case for private culture.

This post looks at the possibility of creating new units to deal with learned information.

My proposal is "lemes" - an abbreviation for "learned genes". Here's a diagram showing three possible units:

Where do memes fit into this diagrem? In two possible ways. The first is conventional, the second is radical.

New terminology: Lemes

Meme terminology promoted

The advantage of using "lemes" is that it doesn't involve the confusing redefinition of any existing common terms.

The problem with the term "leme" is that it seems relatively unlikely to get anywhere. Scientists might benefit from the three separate names for units of learned information - but most people are only going to bother with one term - and the term "meme" has pretty-much already won the battle for that role.

If "lemes" failed to take off, the problem would remain, and the solution would have been pretty ineffective.

I think the umbrella category that covers all learning is the most important, and that it should probably get the best term - which today appears to be "meme". So, I don't think the idea of "lemes" described on this page would be very practical to implement.

While we are entertaining the idea of expanding memetics to include all learning, perhaps we should consider the possibility of expanding it further - to include the evolution of the brain.

Not all evolutionary changes in the brain are to do with learning - there are also changes due to developmental processes, for example.

I think this type of expansion of the idea would probably be taking things too far, though.

The case for private culture

There are many defintions of "culture". Most agree that culture is a socially-transmitted phenomenon. This post will criticise those definitions - and argue that we can make better use of this word.

Firstly, the etymology of the word "culture" is based on an agricultural metaphor. It refers to "cultivation of the mind". It makes no mention of social transmission. I propose that individual learning may also be used to "cultivate the mind".

Then there's the issue of what is important about culture:

Brains allow organisms to use their brains to adapt to the environment. Organisms use both social and individual learning to help them do that. However, social learning additionally allows groups of organisms to maintain knowledge pools for long periods of time - allowing for cumulative cultural evolution. This possibility seems to be the main phenomenon that makes culture of interest. However, now there are other means of creating persistent databases - by creating backups. The possibility of making backups creates potentially-immortal learning systems which can preserve large databases without the need for separate individuals to act as peers. Such systems could exhibit cumulative cultural evolution using individual learning only.

Also, we currenly lack a good term for "learned material". "Culture" is material that is learned socially. The term "knowledge" excludes learned behaviour patterns. The term "skill" excludes learned knowledge. The term "engrams" only refers to how learned material is stored. "Learning" seems pretty overloaded. Overall, there seems to be no good general term for "things that have been learned".

These considerations suggest that it would be better - scientifically speaking - if the term "culture" was defined so that it includes individual learning.

I don't think is is tremendously important to have a term for social learning that excludes individual learning - since social learning and individual learning are such inextricably interconnected phenomena - however, if such a term is needed, "transmitted culture" could be used.

Of course, "culture" is a common word with considerable inertia. It seems appropriate to hesitate before attempting to assign it a counter-intuitive scientific meaning. Redefining common English terms is a reliable way to self-identify as a crank. However, expanding the term "culture" so would resolve some boundary problems with both cultural evolution and memetics. I think that we should bear this option in mind.

References

• The case for private memes;

The case for private memes

Memes are commonly associated with social communication and learning.

However, socially-learned information coevolves with information acquired within the brain by individual learning. Socially-learned ideas compete with individually-learned ideas for attention and for the same neural real estate. All learned information evolves - not just socially-learned information. The Darwinian basis of individual learning was appreciated by Skinner. The case that learned information evolves along Darwinian lines has been made before - for example by Gary Cziko in Without Miracles.

Social learning depends on the same mechanisms as individual learning to operate. Also, many socially-learned operations incorporate individual learning into their transmission process. For example learning to ride a bicycle is probably about 10% social learning and 90% individual learning.

Cultural evolution may be modelled as a process in which individuals learn from others, combine the results with information they already have, add to it with individual learning, and then pass the results on. Individual learning is a fundamental part of this process.

Ideas inside brains are not represented to treated very differently if they are learned from another individual, or produced locally. For example, if Alice sees Bob using a slip-knot and then reverse-engineers the knot, the resulting knowledge is very similar to if Alice invents the slip-knot by herself.

I think consideration of where the optimal scientific boundaries lie strongly suggests that an evolutionary science of ideas should embrace both social and individual learning. Attempting to place a division between social and individual learning is a dubious way to divide the scientific area - since these two fields are so closely interdependent.

In the past, I've described such ideas in terms of "intercranial memes" and "protomemes".

This post will discuss more radical solutions.

Solutions

One possibility would be to create new terminology to deal with Darwinian approaches to individual learning. Skinner has already provided us with the term "extinction" - but the terminology of individual learning could be much more thoroughly Darwinised.

Another obvious possibility would be just to expand memetics to include individual learning. Memetics already has some momentum. Cloning it to handle individual learning seems inferior to just expanding memetics to cope with it. That would mean that protomemes became full memes.

This would pleasingly unite Dawkins' "memes" with Semon's "mnemes".

When giving his etymology of the term "meme", Dawkins originally offered the "consolation" that:

it could alternatively be thought of as being related to `memory'

However, a "memory-gene" only really makes sense if it covers individual learning too.

As I said in my post on meme etymology:

Semon's influence can live on by making memes into "memory-genes" - making use of the "consolation" that Dawkins offered us.

Meme's eye view

If genes are what cells inherit from their parents, memes should be what cultural creatures get from their parents. However, just as parasites reproduce inside their hosts, memes typically reproduce inside minds - as well as between them - and so the parent of a meme may well be another meme inside the same mind. With this perspective, it seems rather odd to only consider memes that are descended from memes in the previous generation of hosts. It should be the previous generation of memes that are under consideration.

Disadvantages

This solution proposed here does have some disadvantages:

Expanding memetics would conflicts with the 35-year old history of memes and memetics. It would probably divide the memetics community - which could do without such upheavals. It would cause considerable further confusion. Also, the attempt might fail.

Critics would probably delight in some memes retreating further into the human mind - where investigating them empirically would become more difficult. They would also probably delight in some memes getting more tangled together. Already memetics regularly faces the criticism that memes aren't as discrete as they should be if they are to be "like" genes. At the moment memes neatly divide into discrete chunks during environmental transmission - but protomemes being memes would mean that that was no longer true. These criticisms are pretty daft - but memeticists should probably hesitate before willingly providing their critics with ammunition.

The move might also divide memetics from the adademic field of cultural evolution. At the momemnt, it is quite convenient for me to be able to say that memes have scientific validity - since they are almost exactly what Boyd and Richerson defined their "cultural variants" to be in 1985. If memetics expanded, that claim would become inaccurate.

I don't plan to resolve these political disputes here at this stage.

This post exists to point at the field of learning (composed of both social-learning and individual-learning) and to say: that area is what we most urgently need an evolutionary science of - and that area should probably have the best available terminology.

For more about this issue, please see: The case for private culture.

Individual learning in memetics

Individual learning is important to memetics. In this post, we will consider why that is.

The following diagram illustrates the various ways in which information is preserved in evolution.

On the left we see the cellullar realm. On the right is inheritance via brains. Environmental persistence is at the bottom.

"Dual inheritance" theories deal mostly with information represented in DNA and information that is socially-transmitted. These are the biggest two persistent channels.

On the right hand side, "individual learning" is included. That refers to the subject that B. F. Skinner studied. However, individual learning is a bit of an odd-one-out in this diagram. The reason for that is that individually-learned information does not normally persist for very long - since it dies with the individual. In humans it sometimes turns into socially-transmitted information - thus persisting beyond the death of its host.

Now we get to the point of this post. Individual learning is critical to understanding social learning.

For example, learning to ride a bike is about 10% social learning and 90% individual learning.

When ideas compete for attention inside minds, their origin (from self or other) is not critical - all ideas compete for mental real estate in a similar way.

Individual learning builds on top of socially learned information, and the results are then transmitted to the next generation via social learning again.

These areas of social learning and individual learning are highly interwoven and interdependent.

It is impossible to study the evolution of socially learned behaviours in isolation. Because of their interdependence, the more scientifically significant domain is based on the union of social and individual learning - the study of the evolution of learning systems.

Individual learning too is best modelled using an evolutionary foundation - as was observed by Skinner himself. However, the Darwinian science of thought competition is even further behind than the Darwinian science of meme competition.

I think part of the problem here is the word "culture". The term carves social learning off as though it is an independent thing - when in fact that is a misleading way of thinking about it.

The deep dependence of social learning on individual learning raises the issue of whether we should grant protomemes and intercranial memes full meme status - in defiance of years of tradition.

References

• Tyler, Tim (2011) Does culture include artifacts?
• Tyler, Tim (2011) The case for private memes;
• Tyler, Tim (2011) The case for private culture;
• Tyler, Tim (2011) Intracranial memetics and intercranial memetics;
• Tyler, Tim (2011) Protomemes.