Friday 20 July 2012

Cracking the Tough Nut of Chimp Tool Use

A paper just out in PLoSOne reports that chimpanzees, given some experience and enough of a weight difference, prefer to use heavier hammer stones when cracking hard nuts. This is apparently quite exciting: this is the first study to isolate weight as a property relevant to the task of cracking open a nut.

This caught my attention because weight is not actually the only key property that determines nut cracking success. A heavy hammer is great, but it will eventually become too heavy to lift, and for a given size stone there may very well be an optimum weight (similar to how people choose very specific combinations of size and weight when asked to throw objects to a maximum distance; Zhu & Bingham, 2011). In the current experiment, when the only difference between the objects was weight, the chimpanzees often went to the heaviest stone because it took the fewest strikes and least time to crack the nut. But is this just an artefact of the current experiment? And if so, can an ecological approach find ways to find out just how chimps choose their tools?

The experiment
There were three experiments, all with the same 6 chimpanzees. One key feature to remember, therefore, is that over the course of the study the chimps all had a lot of practice at this task - this is very interesting and reflected in the data.

Experiment 1 used roughly cube shaped hammer stones (6cm x 8cm x 6cm) that weighed 300g, 600g or 1200g. The weight was evenly distributed throughout the objects. The chimps were asked to choose a stone and crack a nut, and were allowed to switch stones if they wanted to. 
Two of the chimps exhibited a preference for a hammer; Loi preferred the heavy one and Zamba preferred the light one. There was no preference on average, however. All the hammers could successfully crack a nut: when used, however, the heaviest hammer took fewer strikes and less time than the others to crack the nut. There was very little switching around between hammers, however. 

The cuboid was awkward to hold (hint one that weight is not the only key factor). To crack a nut the chimps had to align a flat side with the anvil stone and could therefore misalign the hammer, wasting a strike. In the next two experiments the hammer stones were replaced with 7cm spheres of the same, evenly distributed weights.

Experiment 2 showed an overall group preference for the heaviest object, although no individual chimp showed this preference significantly. Experiment 3 increased the weight range (to 200g, 800g and 1400g) and found a stronger preference for the heaviest hammer; 4 of the 6 chimps showed the effect and would often switch away from the lighter hammers if they picked one up first.

Over time, with practice and a wide range of weights, 4 of the chimps learned to use weight to identify the most efficient hammer stone from the set available. 

What to do next?
There are many things to try next, if you are prepared to think about this task ecologically and measure some more interesting things.

First, weight was clearly not the only relevant property. In Experiment 1, the number of strikes and time to crack the nut did not decrease as a linear function of weight - the middle weight was problematic.
Number of hits and time to crack the nut as a function of hammer weight
Changing to the spherical object cleaned this up - so clearly size and shape matter. Part of makes a good hammer stone is how graspable it is by the chimps. So the first thing to do is to vary the maximum object extent (Mon-Williams & Bingham, 2011) and to do so relative to hand span. You could also add handles, etc. You can also vary size and weight independently within a shape - do chimps learn preferences for specific combinations, as humans do in throwing? What about other properties of the hammer stone? Is it rigid and strong enough? (There's evidence of apes using different stones for different nuts in the wild, for example).

All of this manipulates the affordances of the hammer, specifically how moveable it is (Shockley, Carello & Turvey, 2004) and how easy it is to wield as a hammer. Exactly which properties of the hammer the chimps are tuning into is an empirical question, and the current study has limited their search space artificially. By letting them explore a set of hammers with a controlled but wide set of potential properties, we could identify what the chimps are actually perceiving.

Once we know what they are perceiving, we can ask how are they perceiving it - this means studying information. 

This, unfortunately, is complicated in haptic perception. Research has demonstrated that when we wield objects, the dynamic property we perceive is not mass or size per se, but the resistance of the object to being moved - it's inertia. The description of the objects inertia in 3D space is called the inertia tensor and the perception of object properties via this tensor is called dynamic touch (Turvey, 1996).

The problem is that, while we're pretty sure the inertia tensor is the dynamic property being perceived, we do not know how this is specified in the kinematics (motion) of the haptic system (I mentioned this problem when discussing Withagen's work on non-specifying variables). It's a real obstacle; however, there are still plenty of experiments you can run to show that it is, indeed, the inertia that is the key affordance property, and not the weight. 

For example, the weight in the hammers in the current study was evenly distributed throughout the objects. This means that the inertia tensor is symmetric (the object resists being moved equally in all directions; think about a tennis ball). To see if behaviour follows the inertia tensor and not simply mass, you need to decouple these properties by changing the distribution of mass throughout the object (think about a tennis racquet). You then let the chimps interact with the objects and see what happens (well, it's a little more complicated than that but this is the basic idea).

There are also ways to present visual information about inertia; by setting the object in motion, there is visual information that reflects the underlying dynamic of the mass distribution, for example. Again, non-trivial to do with apes, but the principles are there to be adapted.

The current study measured number of strikes and time required to crack the nut. One thing I noticed in the study is that the strike frequency decreased with mass; the chimps were wielding the objects differently. I would love to see more detailed kinematic analyses of the actions, relative to the inertial properties of the objects. If the hammer stone was marked with distinct colours or if the apes were happy to have passive markers placed on their wrists, you could digitise high speed video fairly easily to obtain these data.

I would also love to quantify the apes' exploration of the objects. There wasn't a huge amount, but there were some switches between objects, especially in the later experiments, as the chimps picked up and discarded the lighter hammers (they picked them up in the first place because the objects were designed to be visually indistinguishable). What did they do when they lifted the objects? Did they heft the hammers? How much time did it take?

This paper assumes that weight is the key property in hammering open a nut.  However, what the affordance property is that's shaping behaviour is always an empirical question, as is the identity of the information that allows the affordance to be perceived. The relevant experimental tools exist from human studies, and I would love the opportunity to work with these authors to adapt these protocols to work with chimps. I think that an ecological task analysis and the related experimental techniques have a lot to add to this exploration of tool use in our closest relatives.

Mon-Williams, M. & Bingham, G.P. (2011). Discovering affordances that determine the spatial structure of reach-to-grasp movements. Experimental Brain Research, 211(1), 145-160.  Download

Shockley, K., C. Carello, and M. T. Turvey (2004). Metamers in the haptic perception of heaviness and moveableness. Perception and Psychophysics, 66, 731 742. 

Schrauf, C, Call, J, Fuwa, K, & Hirata, S (2012). Do Chimpanzees Use Weight to Select Hammer Tools? PLoSOne, 7 (7) DOI: 10.1371/journal.pone.0041044

Turvey, MT (1996). Dynamic touch. American Psychologist, 51(11), 1134-1152.

Zhu, Q. & Bingham, G.P. (2011). Human readiness to throw: the size- weight illusion is not an illusion when picking the best objects to throw. Evolution and Human Behavior, 32(4), 288-293. 


  1. If you've not seen it already, this might be of interest to you as well (all about chimps lifting!)

    1. '30 never before published studies', huh - not worrying at all :)

      This is apparently the latest hurrah in an argument about theory of mind and whether apes can represent things like weight. The paper I'm talking about is partly a response to this stuff, hence the fascination with apes knowing about weight. Weird :)

    2. Hmmm... This is the context I expected (as mentioned on twitter).... Don't underestimate Povinelli. He is no slouch on the publishing front, and even his unpublished studies are ridiculously good. Though I don't always agree with him, IMHO he is the smartest person doing ape cognition at the moment.

      Also, my expectation would that though Povinelli wouldn't use the word "affordance", his studies would have weight as a crucial component of the affordance.

    3. Weight is clearly part of the deal. But a) it's unlikely to be the whole thing and b) even if it is, the information for weight can't be weight, because weight is dynamic and information is kinematic.

      The question is: can you do affordance research with chimps? I'm assuming it would be hard, but not impossible and well worth the effort, especially if people are so excited about whether chimps 'know' about things like weight.

  2. If we can't do affordance research with animals, then we are screwed. We should be able to use fish and mice, nevertheless chimps. Though of course it mirror's the decline in the larger field of psychology, the disappearance of comparative work in Eco Psych is problematic.

    I don't know about all of Povinelli's weight stuff, but I know some of it. He has done lots of experiments in which the (relative) weight of objects should have been detectable visually, using object motions and interactions, as well as allowing the chimps to see other people interacting with the objects. Certainly not all of his experiments would reveal specifying invariants, but I'll bet at least a few of the 30 would satisfy you in those regards. Regardless, in these types of tasks young children pick up on the relative weight of objects pretty quickly. If I recall, chimps showed no evidence of selecting objects due to weight and no physiological signs of anticipating object weight (in muscle tension before contact, etc.) even across many, many trials with highly distinct objects. There were more than enough trials that one would expect the chimps to master the tasks, even if just through operant conditioning.

    Again, I don't always agree with Povinelli, but he is right in his overarching argument that other animals have "minds" worth trying to understand in their own right, i.e., absent the pretense that they are degenerate humans. His early work demonstrated quite convincingly that chimps don't understand looking (the looking activities of social partners), and now he is trying to show that they don't understand weight. He phrases it in terms of representation, but I would bet he would be just as happy to phrase it all in terms of perception and affordances; that issues isn't really his beef.

    As for how excited we should be about this either way... I don't have a major opinion. Comparative psychologists are particularly good at getting riled up by narrow issues with questionable payoff.

    Also, you are completely correct that the authors of the paper above didn't seem to do anything to show how weight affected break-ability. This is an amazing oversight. Given that the rock was just as small part of the chimp-body-arm-rock system, the rock's weight might have made trivial contribution. I can break nuts perfectly well with the hammer my 5 year old plays with and with a 20-pound sledge hammer. That is a pretty wide range.

  3. My only worry about studying affordances in chimps is that I don't know how hard it is to get chimps to do the kinds of careful experiments you'd need to do. Obviously I think they're perceiving affordances.

    I'm intrigued by the weight stuff, then - they can clearly perceive weight, but don't respond to prompts to us it that a human child would respond to. Of course, this may just mean that you aren't asking the chimps the question in the right way - perhaps they aren't interested in the kinds of prompting humans are interested in. It's like scolding a dog like you'd scold a child - it's not the content of the words doing the work on the dog.

  4. It depends on what you mean by "perceive weight". Once the chimps are holding the object, they behave correctly relative to its weight, but... to slip into cognitive language... the question is whether they "attribute" that quality to the object beyond responding-correctly-to-immediate-kinesthetic-feedback, i.e., weight does not seem to be an important part of their "representation" of an object. (More Eco-friendly language available upon request.)

    I glanced through the book a bit (through Google Books). For example, chimps will sort otherwise identical objects based on size, color, shape, etc. pretty easily (a few dozen trials to criterion might suggest a slow chimp), but it is really hard to get them to sort by weight (the quickest chimp did it in 390 trials, and 3 chimps took over 1000 trials to reach criterion!). There was also a cool experiment where they dropped optically identical objects of different weights, to create acoustic information regarding weight, then they had the chimps pick them up to put on a table. The height of the initial pick-up did not differ by weight. This stayed true even if the chimp heard the impact of a different weight object (dropped out of sight). Note here, perception of weight is being measured during movements incidental to a task the chimp is well motivated to do (namely, put an object on table to get reward). My bet is that this would be an ideal way to do affordance research with animals.

    Regarding the latter task (p. 90, I think): "In summary, despite their previous experiences on these weight-lifting tasks (as well as the weight-sorting tasks described in Chapter 4), our apes did not appear to use the dramatically different sounds generated by the two objects to program vertical lift forces. This is particularly striking in the face of their previous experiences. After all, both in formal testing and their extensive, everyday interactions with objects in their living compound, our apes had vast experience with objects of different weights generating different sounds when dropped."

    At any rate, skimming through the book, it is painfully cognitive. The experiments could easily be transformed into Eco-Psych terms (with some then being kept and others rejected). Alas, I am not sure Povinelli's rationale can be so transformed.

    Skimming the literature beyond the book... Apparently there is some more recent evidence suggesting that chimps are sensitive to weight in some tasks... but the only ones I can find involve direct interaction with objects where weight makes a clear difference. Hence, I am not sure how this new work relates to Povinelli's assertions.

    1. They can clearly perceive weight related affordances - they can, indeed, pick things up, they can, indeed, move themselves around. The questions remains, what, precisely are they perceiving and how?

      Interesting results, anyway - it would make a fascinating backdrop to try and empirically identify the thing that works!

    2. Hmmm... I am writing long, and not elegant, replies... attempt at rephrase...

      Chimps do not seem to respond to non-kinesthetic information specifying weight-related-affordances.

      Note also that letting them pick up the object is a bit of a cheat. It is like saying that people perceive the door way is walk-throughable, but only by touching it with the far parts of their body, and only when they are half-way through. Or that they can perceive that a gap is cross-able, but only in the middle of the jump. Also, once the object is in hand one can correct based on many non-weight factors, such as the optic movements of the arm (i.e., one can use more force if the object is not high enough, less force if it is too high, and ignore weight completely).

      Does that make more sense?

      I know you probably don't care this much... but is occurring to me that it might be very interesting to try to transform this work into more Eco-Psych friendly language.

    3. That makes more sense (although of course all this is up for empirical grabs - still need to figure out what the ape is actually doing).

      I'm actually very interested in this - I think getting ecological about animal stuff would be great fun and very productive.