September 19, 2009
By Antti Kauppinen
This is the first in a series of posts about recent work in experimental philosophy. I will be examining some persistent general issues with the different experimental approaches by way of looking at particular papers in some detail. I’ll begin with ‘Two Conceptions of Subjective Experience’ by Justin Sytsma and Edouard Machery. The problem that the study highlights is that everyday language is often vague, ambiguous, or just spoken loosely, so that we can’t draw conclusions about people’s concepts just by looking at what they say in response to prompts. We first need to tease out just what people mean, and this can’t be done in a survey that doesn’t allow for a back-and-forth between the researcher and the subject. This would be a problem even if experimentalists solved all the other problems raised by myself and others.
Here’s how the abstract of the paper begins:
Do philosophers and ordinary people conceive of subjective experience in the same way? In this article, we argue that they do not and that the philosophical concept of phenomenal consciousness does not coincide with the folk conception. We first offer experimental support for the hypothesis that philosophers and ordinary people conceive of subjective experience in markedly different ways.
There’s more in the paper, but since my focus is going to be on the alleged experimental support, I will not discuss the rest of the paper.
Sytsma and Machery’s (S&M) thesis is that while philosophers conceive of perceptual experience and bodily sensations as belonging in the same category of subjective experience, unified by the fact that there something that it’s like to be in the relevant states, the folk don’t do so. Their evidence for this comes from the following online study. Both philosophers and non-philosophers were presented with the following prompts about a robot (and corresponding ones about a normal undergraduate):
Jimmy is a relatively simple robot built at a state university. He has a video camera for eyes, wheels for moving about, and two grasping arms with touch sensors that he can move objects with. As part of a psychological experiment, he was put in a room that was empty except for
one blue box, one red box, and one green box (the boxes were identical in all respects except color). An instruction was then transmitted to Jimmy. It
read: “Put the red box in front of the door.” Jimmy did this with no noticeable difficulty. Did Jimmy see red?
Jimmy is a relatively simple robot built at a state university. He has a video camera for eyes, wheels for moving about, and two grasping arms with touch sensors that he can move objects with. As part of a psychological experiment, he was put in a room that was empty except for one blue box, one red box, and one green box (the boxes were identical in all respects except color). An instruction was then transmitted to
Jimmy. It read: “Put the red box in front of the door.” When Jimmy grasped the red box, however, it gave him a strong electric shock. He let go of the box and moved away from it. He did not try to move the box again. Did Jimmy feel pain when he was shocked?
When it came to the normal human case, there was no significant difference between philosophers and the folk: most agreed that the person saw red and felt pain. But in the case of the robot, most philosophers answered that the robot neither saw red nor felt pain, while most non-philosophers answered negatively to the pain question but positively to the seeing red question. S&M conclude:
On average, the folk (but not philosophers) are willing to ascribe the perceptual state of seeing red to a simple robot. Given the illustrative centrality of the example of the redness of red to the philosophical concept of phenomenal consciousness, our results indicate that philosophers’ concept of phenomenal consciousness is not how the folk understand subjective experience. (p. 13)
Now, there is a pretty obvious response to this: maybe ‘seeing’ is ambiguous between differential responsiveness to the visible features of the environment and such sensitivity + phenomenal experience, or, better yet, the first of these concepts gets triggered in loose talk. It’s not that the folk and philosophers have a different concept: rather, the ordinary term can express either, and the folk latch on to the former in the context of the study. Consequently, there is no one-to-one correspondence between language and thought, and we can’t draw any conclusions about the latter by looking at the former before we’ve established one by resolving ambiguities, taming vagueness, or tightening loose talk.
I raised the issue in a Q&A a couple of years ago (as did a number of commenters on the Experimental Philosophy blog when the study was posted there recently), and it turned out, unsurprisingly, that this was not a new concern to the authors. Given that the response is so frequent, S&M address it in the paper. As they nicely put it, according to it ‘seeing red’ has both informational and phenomenal readings, and the folk select the former, unlike the philosophers. So what, according to them, is wrong with this natural thought?
Their first response is that the distinction between the two readings is ‘ad hoc’. This is a strange thing to say. Roughly speaking, a distinction is ad hoc if the only use it has is to respond to a criticism. But the fact that ‘seeing’ and other sensing verbs have an informational reading should be obvious anyway, and it would be shocking if nobody had pointed it out before in the literature. After all, we talk like this all the time: the baby alarm ‘hears’ the baby cry, and alerts the parents, the missile ‘senses’ the heat pattern and redirects itself, and so on. Ambiguity is probably too strong a term to describe this – as I’ll suggest, there’s some reason to think that this is just loose talk, a convenient shorthand that we (the folk) recognize as strictly speaking false. Compare this to the loose use of ‘knows’ – we use ‘know’ for accidentally true belief all the time, but are willing to withdraw the claim if some literalist busybody pushes us. Surely (please tell me) nobody would parade a survey showing this as a discovery about the folk concept of knowledge and its divergence from the philosophers’ one.
What triggers one reading rather than another is a separate issue. I don’t think the burden of proof on why the folk pick out one rather than another reading is on the ambiguity/loose talk theorist, who might not be particularly interested in theorizing about people’s actual use of words. You wouldn’t demand that of other ambiguity claims, would you? But if I had to guess, I’d say most people are being charitable and choosing the least demanding reading – “Ummm…. if the robot didn’t see the box was red, it couldn’t choose the right one”. If S&M were serious about testing for this, they would run a study using prompts in which the ambiguity or looseness is naturally resolved one way or another, and compare the results. I wonder what simply adding “literally/strictly speaking see red” would do? (If that made a difference, it would support the loose talk interpretation.) Or “did Jimmy a) exhibit sensitivity to different wavelengths of light/visual information, b) exhibit sensitivity to redness, c) see red (tick as many boxes as you think apply)”. Sure, this is not trivial – but it’s not that hard either (waiving for the moment general problems with surveys as a tool for getting at people’s concepts).
Second, S&M claim that the pattern of findings about the folk doesn’t support the ambiguity interpretation, because in that case, they claim, “we would expect that a reasonable proportion of the folk would answer negatively to the question “Did Jimmy see red?”—as philosophers do—while the
remainder would answer positively” (p. 16). Now, why would we expect this? Suppose the folk note that the robot is sensitive to colour and charitably pick the informational reading when answering the question. What pattern of responses would you expect, when everybody is in the same context? You’d expect the vast majority to pick the same sense. And that’s just what we find. Only in a different context where the phenomenal reading was made salient would you expect ‘a reasonable proportion’ of the folk answering negatively. Maybe tell a story about a blind person, who is able to detect red surfaces with the help of a super-duper scanner. Would the folk think this person was seeing red? I doubt it. Then a completely parallel story with Jimmy. My hunch is that you’d get a reasonable proportion of not seeing red – especially if alternative responses were available.
Finally, S&M claim that the comments the minority who say the robot doesn’t see red don’t suggest they make use of a phenomenal reading. But again, the most natural reading of the comments is that the subjects are indeed groping toward the phenomenal reading. S&M relate that they say things like “I’m not sure how color is understood by robots and computers”; “seeing is a human attribute”; “seeing is something which animals do.” Now, if they weren’t restricted by the survey method, they could have asked them “What do you mean by ‘a human attribute’? What’s the relevant difference between humans and machines?”, and so on. Alas, the dogma of experimental philosophy, I’m afraid, says that this is contaminating the data, or something. So we don’t know what they were getting at. But it’s pretty reasonable, I think, to assume on the basis of such responses that the people who didn’t think the robot literally saw red figured that you need to be a creature that has conscious experiences.
So, in short, the three responses S&M offer to the ambiguity or loose talk criticism are extremely weak – if anything, the evidence they appeal to supports that interpretation. Is it a surprise that people with a philosophical training gravitate toward a narrow, literal interpretation of a term, while people without such training prefer a loose usage that works well for practical purposes? Not in the least. Why not, then, accept this simple and general hypothesis about the entirely predictable set of data, rather than reach for a story that attributes massive error to generations of really, really smart members of the folk whose curiosity about these matters led them to reflect on them professionally? And why not properly test for the obvious competing hypothesis, when at least the first steps would be very easy to take, even within the constraints of a survey methodology?