In the previous meeting, we distinguished claims about a creature’s mental states and capacities from the behavior and other observable responses that may give us evidence for those claims. We also emphasized that “having a mind” or “being intelligent” bundles together many different questions.
The same observed behavior may fit several different explanations of what is going on mentally.
So it isn’t enough that an animal behaves as we would expect if it had some interesting mental capacity. We also need to ask what other explanations fit the behavior, and what further evidence might help us choose between these explanations.
In our reading from Dr Dolittle’s Delusion, Anderson’s first major example is Clever Hans, a horse who appeared able to answer questions about arithmetic, dates, German words, and other subjects by tapping his hoof the appropriate number of times.
Hans’s performances initially looked like evidence for surprisingly sophisticated intellectual abilities. They were especially impressive because Hans could perform for people other than his trainer. So the obvious hypothesis that the trainer was deliberately signaling the answers did not explain what was happening.
But Oskar Pfungst varied the situation of the test. Two findings were especially revealing:
Pfungst concluded that Hans was responding to very subtle, unintended bodily cues from the people around him. He would begin tapping in response to one set of cues and stop when the person who knew the answer unconsciously reacted as Hans reached the expected number.
So compare these hypotheses:
Hans’s original performances fit both hypotheses. The new experimental situations separated them.
Hans illustrates a general strategy. Suppose two hypotheses, H1 and H2, both predict some observed behavior B. Repeatedly observing B may not help us decide between them. What we want instead is to vary the situation until H1 and H2 make different predictions, and then see what happens.
Evidence of this kind discriminates between competing hypotheses. Pfungst’s tests did exactly that: the arithmetic and cue-reading hypotheses predicted different results when the questioner did not know the answer, or when Hans could not see someone who did.
Anderson draws two important lessons.
First, experimenters can cue an animal without meaning to. Good controls therefore have to exclude more than fraud and deliberate coaching.
Second, an animal may be solving a different problem from the one the experimenters think they have set. Hans really had learned something impressive. He was extraordinarily sensitive to human bodily cues that the humans themselves barely noticed. But that was not the capacity that the original experimenters thought they were testing.
So the lesson is not that Hans “wasn’t intelligent after all.” It is that his performance depended on a different capacity from the one originally attributed to him. This is another reason to ask about particular capacities rather than broad labels like “intelligent.”
Anderson next discusses a traditional principle called Morgan’s Canon. Its original formulation says, roughly, that we should not interpret an animal’s behavior using a “higher” psychological faculty if the behavior can be explained using a “lower” one.
There are obvious problems with that wording. What makes one psychological capacity “higher” or “lower”? Why suppose that animals always use the least sophisticated capacity?
If the same behavior is explained by both a psychologically more complex hypothesis and a less demanding alternative, that behavior alone does not establish the more complex interpretation.
Always choose the simplest-looking psychological explanation, because it is probably true.
The point is instead about what the evidence has shown. If H1 and H2 both explain the observations so far, those observations do not yet justify saying that H1 has been established.
There are two especially important consequences.
If the evidence does not establish that an animal has capacity M, that does not show that it lacks M. Compare:
These are very different. Further evidence may eventually support the conclusion that M is present after all.
Morgan’s Canon does not require us to remain with the psychologically less demanding interpretation forever. It requires us to have evidence for moving beyond alternatives that would explain the same behavior.
A useful strategy is:
Pfungst did exactly this: the arithmetic and cue-reading explanations made different predictions when the questioner did not know the answer.
Anderson’s next example is subtler. A nesting piping plover may perform a “broken-wing display” when a predator approaches. The bird appears injured and moves away from the nest. A predator may follow the apparently easy prey, allowing the plover eventually to fly away, after the predator has been drawn away from its eggs or chicks.
It is natural to describe the bird as pretending to be injured in order to deceive the predator. But that description builds in several claims about what is controlling its behavior. We can distinguish progressively richer explanations:
Each successive explanation attributes more psychological structure. We should ask which distinctions the evidence actually supports.
Carolyn Ristau’s observations provide evidence that the plover’s behavior is flexible:
These observations count against both of the first two interpretations. The display is not merely a simple response automatically elicited by a predator, and it does not look like a stereotyped sequence that simply runs once triggered. The plover adjusts what it does depending on what kind of threat it faces and whether its earlier behavior is succeeding, which supports a more goal-directed interpretation.
Intention A: Get the predator to follow me away from the nest.
Intention B: Get the predator to falsely believe that I am injured, so that it will follow me away from the nest.
B attributes something A does not: that the plover represents the predator as having a particular representation or belief about it. Ristau’s evidence supports the idea that the plover flexibly tries to draw the intruder away, but Anderson argues that it does not yet require this further belief-manipulating interpretation.
“We have not established intentional deception.”
“We have established that the plover does not understand or manipulate the predator’s beliefs.”
The more complex intentional-deception interpretation introduces what researchers call theory of mind: capacities for representing other creatures in terms of their mental perspectives or states.
Here “theory” does not mean a consciously formulated scientific theory. A creature has a theory of mind, in the relevant sense, when it represents others in terms of their mental perspectives or states — for example, what they know, believe, want, or intend. Researchers also investigate sensitivity to what another creature can see — though animals might track others’ lines of sight without representing their mental states in a richer sense. So sensitivity to what another animal can see is evidence relevant to theory of mind, but doesn’t by itself establish that creatures attribute beliefs, knowledge, or other mental states to other animals.
Attributing theory of mind to a creature is more demanding than saying it responds intelligently to other animals’ behavior.
For example, a creature might learn, “When that other animal does X, I should do Y,” without representing “The other animal does X because it believes or wants such-and-such.”
Similarly, successfully causing another animal to behave in a useful way does not by itself show that one understands the other animal’s mind.
This is one reason intentional deception is philosophically interesting. On the richer interpretation, the deceiver does not merely produce behavior that happens to mislead someone. The deceiver aims to influence the other creature’s behavior by influencing what the other creature represents or believes.
Anderson’s “watchbird” example makes the same point. Some birds give false alarm calls that cause competitors to scatter, leaving food behind. The call therefore misleads the other birds. But it remains a further question whether the caller deliberately manipulates what they believe, rather than merely using a behavior that has proved effective at making them leave.
The Clever Hans case also helps reinforce a distinction from our earlier meetings.
We can ask:
Pfungst gives us a better causal/mechanistic explanation, and that changes what Hans’s performance is good evidence for. But it does not tell us what arithmetic understanding itself consists in.
As we continue, we’ll ask about many different capacities, including:
These are not automatically rungs on a single ladder from “less mind” to “more mind.” A creature may have some without others; what looks like a unified package in normal adult humans may contain components that can come apart.
A surprising performance is interesting, but philosophically we’ll ask:
These questions will recur when we turn from animals to humans and artificial systems.
Clever Hans warns against moving too quickly from impressive behavior to a complex psychological interpretation. The plover shows the reverse: further evidence can genuinely rule out simpler explanations.
Distinguish the competing hypotheses and ask what the evidence actually supports.
That is the central question we’ll keep using as we examine other animals: Which interpretation of their behavior is best supported, and why?