Sunday, 3 January 2016

Brown and Heyes: Social learning and the other cooperation problem

This is the latest in my series of blog posts summarising the talks and responses that took place at the meeting 'Inheritance and cooperation'.

Unfortunately, some idiot forgot to press the 'rec' button on this one (an idiot called 'Clarke'). So i cannot make any audio available I'm afraid : (

Dr Rachael Brown is a Lecturer at Macquarie University in Sydney who has written about learning, its ability to act as an inheritance mechanism and its effects upon evolutionary processes. Her talk, 'Generating benefit: Social learning and the “other” cooperation problem', explored the idea of treating social learning as a cultural inheritance mechanism, and one that occurs, furthermore, in non-human animals.  

First of all, Brown drew an important distinction between social learning and social transmission. 'Social learning' has a technical definition in psychology, as a cognitive process which occurs as a consequence of observing, or interacting with, another agent. It can be intra or inter-species, giving rise to situations where one species scrounges of information gathered by another. 

Social transmission, on the other hand, occurs when behaviour manifested by one individual exerts a lasting causal influence on the rate at which another individual acquires or performs the same behaviour. In other words, you get social transmission when social learning has a lasting effect. There are various cognitive mechanisms which can support social transmission, including imitation (copying), stimulus enhancement (when attention is drawn, by another agent, to a particular object or part of the environment) and other socially biased input mechanisms.

The traditional view is that non-humans do transmit behaviours socially, but with insufficiently high fidelity to make any difference to evolution. In particular, socially transmitted animal traits do not generally persist long enough so that cumulative improvements can be added.  Only humans are sophisticated enough to use teaching and explicit copying, which can give rise to cumulative cultural evolution.

However, Brown countered this traditional view by drawing our attention to several pieces of evidence for multi-generational inheritance of traits by social transmission in non-human animals. For example, bluehead wrasse fishes learn the location of traditional mating sites using mechanisms of local enhancement and sensitisation that are robust over numerous generations (Laland and Janik 2006). Aisner & Terkel 1992 used cross-fostering experiments to prove that Black rats use stimulus enhancement to socially learn a special method for stripping the scales from pine cones. New Caledonian crows use ecological scaffolding to learn to manufacture, and use, stick-and-leaf tools in foraging. Kenward et al 2005 argue that attentional biases have been canalised to increase the fidelity of this learning. Finally, Lucy Aplin, Ben Sheldon and team's work with Great Tits in Oxford demonstrates that the birds can socially transmit arbitrary solutions to foraging problems (Aplin et al 2015).

Brown followed this survey with an exploration of the ways in which social learning might be able to generate benefit - provide fitness advantages - to groups of animals. 'Why live in a group?' she asked. Solitary individuals, after all, avoid many problems that come with group living. For example, the costs of competing with your group mates for space, food and mates. The elevated risks of parasites and disease. The perils, as Paul Rainey once memorably put it to me, of 'shitting in each others' yard'. Furthermore, social learning is cognitively costly. You have to pay attention to your group mates, at the very minimum. Why bother?

One function that has been suggested for social learning is as a source of social information. Group members can get information from the group, without having to pay the costs of finding it out themselves. For example, they can find out about foraging sources and dangers. Bees use their waggle dance to tell each other about the location of nectar. Animals in groups essentially have many eyes - so they can find food faster in patchy environments and find novel adaptations faster. Is this enough to make up for the fact that they also have many mouths to feed?

One of Brown’s core insights is that thinking about social transmission as a form of inheritance opens up the possibility that group living has diachronic as well as synchronic benefits -  benefits that are accumulated over the lifetime of the organism, but also over the lifetimes of many generations of organisms. In other words, if the information transmitted can endure over time, then it can accumulate indefinitely. The mouths will be quickly outnumbered by eyes if previous generations of eyes are part of the sum.

Brown supposes that there will cognitive limitations, in the case of non-human animals, that might prevent social transmitters from integrating bits of accumulated information in a compositional manner. So the Caledonian crow tools might not get more complex over time, but nonetheless they can acquire significant advantages over solitary living if they can accumulate many simple tools, or pieces of knowledge about food sources, over time. 

There are costs associated with social transmission, however, not least the risks that maladaptive traits or false information will be perpetuated. This is thought to be worse in highly changeable environments, where information quickly becomes outdated. A 2009 study by Kevin Laland and colleagues found that this cost was ameliorated if the population included a mixture of social and asocial learners, or a mix of neophilic and neophobic personalities. Another way to reduce risks is to utilise some sort of learning strategy, which includes rules about who and when to copy (Laland & Hoppott 2013).

Another interesting point made by Brown was that the ‘free rider problem’ might be transformed in groups with social learning, because the cost of being excluded from the group includes losing access to all the information the group possesses, which may be high. If information is vital to survival, then the cost of being kicked out of the group is really high and cheating just wouldn’t be worthwhile.

Brown finished with some tantalising speculations about extending the idea of the group as an information store to other inheritance mechanisms.  For example, in microbial aggregates it has been hypothesized that lateral gene transfer enables cells to draw upon a common store of genetic information, which can be used in defending against antibiotic treatments. Presumably, this idea works best for mechanisms of horizontal transmission, although it is also commonplace to talk about gene flow within sexual species as constituting a sort of genetic information exchange. 

Professor Cecilia Heyes FBA is a Psychologist at All Souls College, Oxford, who studies the evolution of cognition, especially social cognition in humans. She is developing an account of what makes humans unique, in which we are characterised by an unlimited capacity for learning how to learn.

Heyes praised Brown’s focus on persistence as essential for the accumulation of diachronic advantages to social transmission. She noted that the term ‘fidelity’ obscures an important distinction between the persistence of a cultural trait -the extent to which its effect endures - and its exactness - the extent to which precise, detailed copied are made.  A population can only accumulate knowledge over many generations if that knowledge is long-lasting. Brown is right, furthermore, to identify important group advantages that could rest on persistence, even if not accompanied by exactness.

But Heyes is skeptical about the evidence for social learning in non-humans. She argued that if you look carefully at Rendall et al's 2010 paper, their data doesn't support the claim that social learning receives the highest score. She suspects that social learning, without persistent effects, is not as helpful as many have claimed. She believes Aplin's Great Tit work provides the strongest evidence we have that primates are not special among non-human animals, when it comes to social transmission. However, she isn't persuaded by this study that social transmission isn't incredibly rare in non-humans, because the experiment used such an unusually salient and distinctive pair of behavioural variants.

The great tits observed a model moving a brightly coloured door to the right using the red side, or to the left using the blue side. Heyes sold us a compelling story about what-it's-probably-like-to-be a Great Tit taking Aplin's test. Psychologists, apparently, have masses of data showing that most creatures enjoy the salient and the familiar. When one bird observes another pecking the correct door, it is visually very obvious which way their body moves, as their tail bobs up and the beak stretches towards the large coloured logo. Any bird performing the same movement is implicitly rewarded by that precious feeling of familiarity, and pecking the opposite door just isn't as satisfying. The combination of bright, moving colours and body movement will have made the action highly attention grabbing, even from some distance, and easy to associate with the payoff, the sight of the model feeding on the seed beyond the door once it had been moved. This would reduce the likelihood that the observers would try the other behavioural variant when they approached the feeder for the first time, and reduce the likelihood that, should they try the other variant, they would find it as rewarding as the one they had observed. In psychology jargon this is because there would be little 'generalisation of conditioned reinforcement' from the observed variant to the alternative variant.

Aplin and colleagues certainly found persistence - 85% of the 75 birds who tried the alternative stuck with their socially learned preference. But it's possible that such persistence will only occur when the salience and distinctiveness of the different options is unusually high - much higher than it will be in nature. In which case we cannot generalise much from the result.

Finally, Heyes drew upon Godfrey-Smith's work to argue that social transmission is not really a form of inheritance, for three reasons:

Firstly, social transmission, unlike biological evolution, includes a process in which socially learned traits are modified by practical intelligence. Secondly, social transmission doesn’t give rise to proper parent-offspring lineages, because behaviours have too many cultural ‘parents’. One consequence of this is that there is wide scope for biases to creep in, as in conformist bias when an individual copies from some parents in preference to others, according to which behaviour is exhibited by a majority. Lastly, one of the important properties of a high powered selection process is the creation of new opportunities in which a trait can be tinkered with - improved.  But social transmission does not result in the birth of new individuals, it merely spreads the traits to existing individuals, so it cannot increase the overall number of platforms on which improvements can be tried out.

In the case of Aplin's birds, there may be no problem about practical intelligence. But there certainly was conformist bias, rather than fair inheritance from each cultural parent. And the observers were not free to undertake further tinkering of the behaviour they learned. They were constrained by the apparatus to get the food in one of just two ways.

In sum, Heyes admired Brown’s attempt to bridge the gap between animal social learning and human culture, but she isn't convinced that even the best study around -Aplin's study - provides a firm foundation for such a bridge.

I think Brown and Heyes are right to push for greater disentanglement of exactness and persistence in transmission. The two are often run together in signalling theory, under the assumption that exactness supports persistence, because signals that deteriorate ultimately get lost amongst noise. But this neglects all manner of more complicated scenarios. Suppose the signal is copied with such extreme exactness that the receiver is too exhausted to transmit it. Or suppose that a signal can achieve massive persistence in virtue of being highly evolvable, so that each copying event is rather inexact. Evolvability requires a particular level of exactness – not too much, not too little, in addition to persistence.

How plausible is it that non-human populations could sustain information over long enough time scales? Would it impose a large burden on individuals’ memories? Perhaps there could be some sort of division of labour between specialists in different sorts of knowledge? Like the people at the end of Ray Bradbury’s Farenheit 451, who set out to preserve human culture against the book burners by memorising one book each.That seems to impose weighty costs of a different sort though - organising who is memorising which bit of data efficiently. There might be room for personality and other sources of variation to help out here. Presumably one of humanity's key innovations was to start storing information in our local environment so as to ease the burden on individual memory. The construction of tools is one way of storing information, with writing obviously coming later. 

Final thought then, perhaps it is worth questioning how much cultural evolution would be supportable by human social learning if it occurred in the absence of techniques for storing information outside of our heads?

No comments: