Information use in social interactions – thoughts from/after an awesome workshop

IMG_1157
Mont-Collon – with glacier!

I’ve just returned from a workshop in Switzerland that I was really excited to be a part of: “The costs and benefits of information acquisition and use in social interactions”, organised by Michael Taborsky and Leif Engqvist from the University of Bern. We had 8 excellent speakers, with plenty of time between for group discussions (and four-course Swiss meals) looking out over the imposing north face of Mont Collon (3,637m!). I was really keen to be involved in this workshop as it really tied together my previous work (focusing on adaptive functions of certain types of social interactions in a cooperative species) with what I’m looking at now and moving more towards: how individuals use information about their environment and social groups. Michael’s work on how individuals respond to social information in cichlid groups has been a big influence on how I’ve thought about cooperation, so I was also very motivated to meet and hear more from him.

The focus of the workshop was on social information, but given the broad interests of the speakers, we heard and talked about a variety of topics around this theme, from cognition’s role in group-living, to plasticity in information use. I’ll summarise what I took from the talks, and give a brief overview of the key messages or questions that we took away from the workshop as a whole.

Michael started by asking what information individuals use when making decisions, and introduced Lloyd Morgan’s idea of parsimony in this context: i.e. behaviour should be interpreted as having used the most simple rules, and the lowest order of information. How true is this in the case of animal decision-making? He gave us some lovely examples from his own work on Norway rats highlighting the more complex information rats use when deciding whether to help a partner (by pulling a lever that gives the partner a food item), and the importance of context, something we came back to again and again over the course of the workshop. For example, rats pull the lever quicker to help out a partner if that partner had previously given them a high quality food item, relative to partners that had provided the focal individual with a lower quality item (Dolivo & Taborsky 2015).

Jens Krause (Humboldt University, Berlin) then spoke about decision making in the context of collective behaviour and movement. He’s been doing some neat experiments with robotic fish to test how an individual’s behaviour within a group can alter the behaviour of others (Landgraf et al. 2013).

ROBOFISH! image from http://images.sciencedaily.com
ROBOFISH! image from http://images.sciencedaily.com

If one individual moves in a direction that’s counterintuitive, for example, will other individuals forego their own personal information to follow them on the basis of social information? He also introduced a new concept to me: the “quorum threshold”, the proportion of a group that “decides” for an individual, i.e. an individual’s likelihood of choosing an option is based on whether X% of a group are already committed to that option (leave patch, for example) (Sumpter & Pratt 2009).

Below is a TED talk that Jens did at Ghent on group movement and swarm intelligence…

Sasha Dall from the University of Exeter asked a question that then came up in almost every subsequent discussion – what do we mean when we talk about information? (Dall et al. 2005.) He argues that the details we pick up about our environment is NOT information, but ‘data’ – information is what is derived from that data that reduces our uncertainty about the world – and that we should discriminate between content vs mechanism. For example, the news may provide information but is not inherently informative – it is a ‘channel’. (Interestingly, it seems that the problem of distinguishing data from information may be partly due to the fact that we don’t have a distinct word for what Sasha terms ‘data’ – Michael pointed out in group discussion that in German, the word informationen is analogous to ‘data’ in this sense, and is different from informatione which in this case is analagous to meaningful information).

Straying a little into my particular interest in information use – in the context of plasticity – Dave Stephens from the University of Minnesota spoke about the model he’s developed to derive predictions about how individuals should respond to personal information and other signals from the environment (McLinn & Stephens 2006).

Figure taken from McLinn & Stephens 2006
Figure taken from McLinn & Stephens 2006

For example, if an animal’s experience with choice A is good, when would it pay to ignore that experience and trust a signal about which choice is better? (See figure). Dave has tested this model empirically with blue jays and Drosophila. Using the latter, he’s shown that some types of environmental change (i.e. when individuals must use experience to inform decisions, rather than rely on innate preferences) can promote the evolution of learning by allowing flies to lay eggs on media of two different flavours with fitness consequences (only eggs on one flavour allowed to survive), and using quinine on one medium as a signal that could be learned (Dunlap & Stephens 2009). Fascinating stuff! A lovely example of combining theory and experimental work that really got me inspired…

Sarah Brosnan from Georgia State works with captive primates to test theories about reciprocity, decision-making and inequity. Particularly striking in her talk were examples of individuals reacting to receiving a poorer reward than a partner for completing the same task, showing clearly that animals have a sense of fairness…

… something that was also shown beautifully by Keith Jensen from Manchester University, who has been looking at what he calls “Other-regarding concerns”; a potentially uniquely human psychological mechanism allowing us to care about the welfare of others? (Jensen et al. 2014). He’s done some experiments that really blew me away, testing (separately) how children (aged 3 and 5) and chimps react to witnessing the food of one individual being stolen by another. Would the focal child/chimp punish the thief even if they didn’t lose anything from this interaction themselves? Chimps didn’t (Riedl et al. 2012) – but kids overwhelmingly did (Riedl et al. 2015)! Could this concern for social norms and fairness (even when fairness doesn’t involve the first party) be what makes human so uniquely prosocial?

Moving back to theory, we also heard about some models of cooperation that Leif Engqvist is developing at Bern (as yet unpublished I think, so I won’t go into details), and from Jeff Stevens who joined us on Skype from Nebraska. Jeff has been doing some neat game theory-type experiments with human subjects to test the cognitive constraints on cooperative models like direct reciprocity (whereby you help an individual that has previously helped you) and has found that errors are common when people are asked to recall the behaviour (cooperation or defection) of previous partners, and increase with the number of interacting partners (Stevens et al. 2011). People were more likely to remember rare-type behaviours (i.e. when defecting is rare, remembered these partners better over varying retention intervals), possibly a mechanism by which cheating is discouraged and cooperation is maintained (Volstorf et al. 2011)?

Lastly, we had a brilliant talk by Thomas Bugnyar from the University of Vienna on his work looking at how learning behaviour in corvids influences and is influenced by social structure. fab6f962a7I am a huge fan of Thomas’ work, as it really integrates the field of cognition with evolution by looking not just at what corvids are capable of learning and retaining, but how this might drive evolution of traits, and social systems! He’s shown through some beautifully simple experiments that ravens pass on information about things they’ve observed horizontally AND vertically (i.e. transgenerational transfer of info) and that they are attuned to third-party relations such as change of dominance within and outside their social group (Massen et al. 2014). Here’s a lovely Q&A with Thomas in Current Biology: he’s one of them most enthusiastic and interested people I’ve met! He’s also organising next year’s European Congress for Behavioural Biology in Vienna, which will be an excellent conference (great speaker line up already!).

These brilliant talks gave us lots to talk about in our group discussion sessions. Here are the main themes that struck me as important in these few days – I’d love to hear anyone’s thoughts on them!

  • again and again we came back to the question of – what do we mean by “information”? At times it seemed that the argument was semantic, but the more we went round in circles the more it seemed important to think about this question carefully and clarify our definitions. I’d come up against this problem before without really taking it further when discussing with my students – if a signal is dishonest does it provide information? I actually heard about the idea of Shannon entropy for the first time – worth reading about if thinking about social systems, information, and its reproductive value in this context. This seemed a useful summary of these issues (from Information Theory Workshop, 2004. IEEE, DOI:10.1109/ITW.2004.1405273
  • should individuals always use the highest order of information available? We developed this question/idea by conceiving a situation in which there were no costs of information acquisition – would there be any circumstances then under which some information could be ignored? There was a lot of discussion here about social context and how that might influence the  benefits of obtaining more information relative to the costs of time delay or information retention.
  • when might it be adaptive to ignore information? This question fed into the previous in getting us thinking about social context and costs/benefits… but was also interesting to consider in terms of Dave’s flag model which predicts that certainty of experience might over-ride the usefulness of potentially unreliable signals.

Finally – the CUSO (Conférence universitaire de Suisse occidentale – a consortium of Swiss universities including Neuchatel, Lausanne and Bern) organises workshops like these every year. I would HIGHLY recommend them – this was an excellent way of meeting other young scientists and established researchers in a friendly, discursive setting.

We finished off with a hike along the jaw-droppingly beautiful valley – perfect!

IMG_1171 IMG_1181-2

References:

Dall et al. 2005. Information and its use by animals in evolutionary ecology. TREE 20(4): 187-193.

Dolivo, V & Taborsky M. 2015. Norway rats reciprocate help according to the quality of help they received. 

Dunlap, A. S. & D. W. Stephens. 2009. Components of change in the evolution of learning and non-learning. Proceedings of the Royal Society (B) 276: 3201-3208.

Jensen et al. 2014. The emergence of human pro sociality: aligning with others through feelings, concerns, and norms. Frontiers in Psychology. Vol.5; article 822.

Landgraf et al 2013. Advances in Swarm IntelligenceLecture Notes in Computer Science Volume 7928, 2013, pp 1-10
Massen et al. 2014. Ravens notice dominance reversals among conspecifics within and outside their social group. Nature Comms. 5, Article: 3679 doi:10.1038/ncomms4679
McLinn, C.M. and Stephens, D.W. 2006. What makes information valuable: signal reliability and environmental uncertainty. Animal Behaviour 71:999-1252.
Riedl et al. 2012. No third-party punishment in chimpanzees. PNAS 2012 109 (37) 1482414829
Riedl et al. 2015. Restorative Justice in Children. Current Biology 25 (13), 1731-1735
Stevens et al. 2011. Forgetting constrains the emergence of cooperative decision strategies. Frontiers in Psychology. Vol 1, Article 235.
Sumpter & Pratt 2009. Quorum responses and consensus decision making. Phil. Trans. R. Soc. B (2009) 364, 743–753
Volstorf et al. 2011. The Good, the Bad, and the Rare: Memory for Partners in Social Interactions. PLoS ONE. DOI: 10.1371/journal.pone.0018945

Allonursing – what are the benefits?

I wrote a while back to ask – is “cooperation” really cooperation if it imposes no cost on the giver? Two things are usually assumed about cooperative behaviours: that they incur a cost to the participating individual, and that they provide a benefit to the receiver. Back then I’d recently shown that subordinate female meerkats that allonursed (provided milk to offspring that weren’t their own) didn’t, as expected, lose a significant amount of weight while doing so relative to other subordinate female helpers in the group, suggesting that the long term costs of allonursing weren’t high at all – the opposite of what I’d expected to find!

But what about the benefits of this behaviour?  This is what I tried to find out next in an analysis that has recently been published in the Journal of Animal Ecology. For allonursing to really be a cooperative behaviour in meerkats, it should provide benefits to the individuals actually receiving the extra milk (pups), and possibly also their mothers, who might be able to reduce their own input because their babies are getting food from another source.

As a reminder, in meerkats, only one dominant female breeds while other females help her to rear her offspring without necessarily having any of their own. In this system, unlike in communally breeding species where females all nurse each others’ young, the costs and benefits should be easier to figure out because only one “type” of female provides milk (subordinate aloneness) and one “type” of female receives the benefits (dominant females whose pups are allonursed)…

… like this!

Capture
In communally breeding species (a) mothers are simultaneously donors and recipients of allonursing, meaning that while mothers and offspring may benefit from offspring receiving allonursing, quantification of these benefits is likely to be confounded by costs (energetic costs to mothers of providing milk to non-offspring, costs to offspring of maternal resources being diverted to non-siblings). In singular cooperative breeders (b) only one female in the social group produces offspring, which are raised by subordinate helpers (denoted “SUB”) not concurrently rearing young of their own, so help is not reciprocal. Resultantly, the costs of providing allonursing are not confounded by the benefits of receiving it and vice versa. Dashed lines enclose individuals likely to incur costs through allonursing; dotted lines enclose individuals likely to benefit from allonursing. Small circles denote offspring.

So what benefits might we expect to see pups and mothers gain from the lactational attentions of a subordinate female?

PUPS: For baby meerkats, and many other mammals, milk is the primary source of nutrition in early life, and so we’d expect that (as for all of us – sigh) getting more nutrition would mean gaining more mass (Koenig 1997). And gaining mass at this stage of a mini-kat’s life is good news, as heavier pups are more likely to survive to grow up as potentially reproducing adults in the social group (Russell et al. 2002)…

© 2007 Ron Reznick www,digital-images.net [#Beginning of Shooting Data Section] Nikon D2X Focal Length: 200mm Optimize Image:  Color Mode: Mode II (Adobe RGB) Long Exposure NR: Off High ISO NR: Off 2007/08/20 11:03:57.2 Exposure Mode: Aperture Priority White Balance: Color Temp. (5300 K)  Tone Comp.: Less Contrast RAW (12-bit) Metering Mode: Multi-Pattern AF Mode: AF-C Hue Adjustment: 0° Image Size: Large (4288 x 2848) 1/350 sec - F/4 Flash Sync Mode: Not Attached Saturation: Normal Color Exposure Comp.: 0 EV Sharpening: Normal Lens: VR 200mm F/2 G Sensitivity: ISO 200 Image Comment: (c)RonReznick_www.digital-images.net [#End of Shooting Data Section]
For meerkat babies, chunky is funky © 2007 Ron Reznick
http://www.digital-images.net

MOTHERS: For meerkat mothers, the benefits are perhaps a little less clear. They don’t actually receive any food themselves – but, it’s possible that by their offspring receiving food from another individual, it frees up their time to forage more and nurse less, or reduces the energetic demands on them in terms of milk production. If this was the case, we’d expect that mothers with allonursing help should be in better condition when the pups are ready to forage for themselves than mothers who have to carry the burden of lactation alone. Being in better condition also has implications for other aspects of maternal life – such as reducing the time to the next pregnancy.

To try and figure out how allonursing influences measures of pup condition, and maternal condition, and how these things might all be interlinked (along with environmental and social factors in a great web of ecological complexity), I constructed (with the help of the wonderful Katie McGhee) a structural equation model.

Simplified structural equation model, arrows depicting expected relationships
Simplified structural equation model, arrows depicting expected relationships

Above is a simplified version of what we created – which was basically a web of relationships, which we specified based on knowledge of the meerkat system (in other words, we know that rainfall = heavier mums = larger litters, so we could specify those “paths” in the model). We then specified paths between allonursing and the things we expected it to influence, and tested the fit of the full model – and hooray, it fit the data! Then, to determine how important allonursing is in these patterns, we took it and its paths out of the model, and re-tested the fit, with the predictions that if it’s important in the patterns specified in the model, a model without allonursing in it should not fit the data. But – surprise! – the model without allonursing fit the data just fine — telling us that the influence that allonursing has on pup mass (g), survival, and maternal condition and lactation duration were not important.

[I really would recommend this model structure. With messy ecological data — and honestly, isn’t that everyone-who-doesn’t-work-on-experimental-systems data? — this really gives you a sensible framework within which to test predictions not just about linear relationships between two variables, but how a variable influences multiple patterns. This paper was a must-read for me – and there’s a whole book available online about the use of SEM in natural systems. Have a look! I’m happy to answer basic questions about how/why to use it.]

But back to the meerkats – and the strange lack of benefit that allonursing appears to provide (or not!). There are a couple of explanations for why we didn’t find the obvious benefits of allonursing that we were expecting to.

  • we were looking at the wrong things – for example, Roulin & Heeb (1999) suggest that pups might benefit from nursing from a number of females as it might allow them to get different types of immune compounds in the milk they receive from multiple sources, resulting in a stronger immune profile that might have a better chance of fighting off diseases. Unfortunately we haven’t yet found a way of sampling milk from the meerkats so haven’t been able to test whether different females differ in what immune compounds are present in their milk.
  • the evolution of allonursing might not be driven by the benefits to recipients, but because the costs are low for the allonurses themselves (MacLeod & Lukas 2014). If you don’t have much money, a fiver seems like a lot, right? But, when you have £100 and it’s going to disappear in a few days, giving away a fiver doesn’t seem like that big a deal. That could be the scenario that subordinate female meerkats find themselves in, especially if they have recently given birth or lost pups to infanticide, and have excess milk… Being charitable with their resources also might benefit them in the long run via…
  • social benefits to allonurses. Were we looking for benefits in the wrong place? It’s possible that subordinate females allonurse more to curry favour with the dominant female than to actually provide milk to hungry pups – a clever move, especially if sucking up to the dominant female means you’re less likely to receive aggression or be chucked out of the group (this is something I’m moving on to look at in the future!)

What we really need to figure out, in order to answer these questions, is how much milk subordinate females are actually passing on to offspring when they allonurse. As I mentioned, we haven’t yet found a way of measuring this in wild meerkats – a project for the future!

References:

Koenig, B. (1997) Cooperative care of young in mammals. Naturwissenschaften, 84, 95–104.

MacLeod, K.J. & Lukas, D. (2014) Revisiting non-offspring nursing: allonursing evolves when the costs are low. Biology Letters, 10, 20140378.

Roulin, A. & Heeb, P. (1999) The immunological function of allosuckling. Ecology Letters, 2, 319–324.

Russell, A.F., Clutton-Brock, T.H., Brotherton, P.N.M., Sharpe, L.L., Mcilrath, G.M.F., Dalerum, D. et al. (2002) Factors affecting pup growth and survival in co-operatively breeding meerkats Suricata suricatta. Journal of Animal Ecology, 71, 700–709.