Symposium on Pendaran Roberts, Keith Allen, and Kelly-Ann Schmidtke’s “Folk Intuitions about the Causal Theory of Perception”

Welcome to our fourth Ergo symposium, featuring Pendaran Roberts (University of Warwick), Keith Allen (University of York), and Kelly-Ann Schmidtke’s (University of Warwick) “Folk Intuitions about the Causal Theory of Perception” with commentaries by Eugen Fischer (University of East Anglia) and John Schwenkler (Florida State University). I’d like to thank each of the participants for all their hard work.

According to the ‘causal theory of perception’, it is a conceptual truth about perception that when you veridically perceive an object (e.g., see it), the object that you perceive causes your perception of it (Grice 1961; Strawson 1992; Lewis 1980). A traditional argument for the causal theory is that it successfully predicts and explains our patterns of intuitive judgments about when you do and do not qualify as seeing an actual object in front of you. For example, the causal theorist invites us to consider various hypothetical scenarios in which our perceptual experience intuitively matches the way the world is, but in which the causal processes underlying our experiences are somehow interfered with—e.g., through clever manipulation of your brain or environment. Intuitively, the causal theorist submits, you do not count as seeing what’s directly in front of you in these scenarios, even if your experience matches what the object is like. The best explanation for this, the causal theorist proposes, is that the causal condition required to apply the concept of sight is not satisfied in the scenario we have envisioned.

In their article, Roberts et al. empirically investigate whether the causal theory of perception embodies a conceptual truth about perception, as its proponents allege it does. To test this question, Roberts et al. performed an experiment in which philosophically untrained participants were presented with multiple vignettes describing hypothetical scenarios involving visual experience and actual material objects in the environment, and were asked to rate the degree to which they agree that, in the hypothetical scenario, they count as seeing the actual object. Significantly, among the vignettes describing causally deviant experiences, some described scenarios in which the source of causal interference is due to a blocker—i.e., an object that blocks the subject’s line of sight of an object, such as a mirror or a virtual reality head set—whereas other vignettes described scenarios in which something else interferes—e.g., ingestion of a hallucinogen or direct neural stimulation by neuroscientists. The main aim of Roberts et al.’s experiment was to determine whether participants’ intuitions about seeing are sensitive to causal deviance, as a causal theory of perception predicts, or more specifically to blockers.

Roberts et al. found that although participants tended to disagree strongly that they would be seeing a thing if their line of sight were blocked by something else, there was a lack of consensus among participants about whether having their experiences interfered with in some other way (e.g., by hallucinogens or direct brain stimulation) would prevent them from seeing what is actually there. More specifically, in cases of non-blocker causal deviance, while a large number of subjects strongly disagreed, a sizable number strongly agreed that they see the actual object. Roberts et al. interpret their finding as evidence that folk intuition about perception involves a no-blocker condition on veridical perception, rather than a more general causal condition. Consequently, they deny that the causal theory embodies a conceptual truth about perception. They conclude by arguing that their position is consistent with various views about conceptual truth, and defending their conclusion against alternative interpretations of their findings and potential objections to their experimental methodology.

You can find the target article, commentaries, and Roberts et al.’s response below.

Target article:

Roberts, Allen, & Schmidtke “Folk Intuitions about the Causal Theory of Perception”


Eugen Fischer “Experimental Assessment of the Causal Theory of Perception”

John Schwenkler “How Do the Folk Think of Seeing?”

Authors’ Response:

Roberts, Allen, & Schmidtke “Folk Intuitions about the Causal Theory of Perception: Reply to Schwenkler and Fischer”


  1. Aaron Henry

    Hi Pendaran, Keith, and Kelly,

    I wanted to begin by saying that I found the main finding of your paper–that subjects appear to hold a ‘no blocker’ condition on seeing, rather than the more general ‘no causal deviance’ condition as traditionally assumed–to be very surprising, as well as fascinating .

    My question concerns one of the alternative explanations for your findings that you discuss, according to which participants strongly agree about the disabling conditions for perception (e.g., that blockers prevent sight), but disagree about its enabling conditions. According to this objection, it’s possible that at least some participants regard the scientist (in the case of “Clock”) or the hallucinogen (in the case of “Snake”) as somehow causally enabling veridical perception of the actual object in the environment. You respond to that objection by pointing out that the question emphasized a lack of counterfactual dependence: in both cases by including the phrase “As it happens, …”, and in one case by explicitly stating that there is a lack of counterfactual dependence (“the scientist would make it look to you as if there were a clock even if there were not one.”). I wasn’t fully convinced by that response. For one thing, an absence of counterfactual dependence does not entail absence of causation. Even accounts that attempt to explain causation in terms of counterfactual dependence (e.g., the different versions of Lewis’s view) must acknowledge situations in which A causes B even though B does not counterfactually depend on A—e.g., in various sorts of pre-emption cases. With that in mind, it might be worth considering whether some participants are thinking of cases like Clock and Snake as cases of causation without counterfactual dependence. For example, perhaps the participants who classify these as instances of seeing are thinking “in the actual course of events, it is the thing out there in the world that causes my visual experience of it (hence, given the lack of blockers, I count as seeing it). But if it hadn’t, the neuroscientist or the hallucinogen alone would have been sufficient to cause me to have the experience anyway, in which case I wouldn’t count as seeing it.” That would be to liken what’s going on here to a case of pre-emption or maybe overdetermination, though there are other possible scenarios in which causation can operate without counterfactual dependence that would be worth thinking about.

  2. Pendaran Roberts

    Hey Aaron,

    Thanks for the comment. I think that your concern is a natural one for the attentive reader. We did try to test the concern (see Section III of the response). What we did is that we used modified clock and snake cases, where we attempted to make it clear that no causal link exists between the object and the subject (see that part of the response). It’s not clear to me that the cases we designed fully account for the concern in question, but they at least partially account for it. I still personally worry that perhaps participants are thinking that the scientist/drug are interacting with the objects (the clock and the snake) in some way or another, and so the clock/snake still have some causal role to play in the subjects’ experiences. However, it is not immediately obvious to me what to do to adjust the cases to avoid concerns like this. Any ideas?

    • Aaron Henry

      Thanks for your response, Pendaran. I had actually forgotten about the new data you discuss in your response to Eugen’s objection when posting my comment. It goes some way toward allaying the worry: the modifications that you introduced to the original scenarios make it frankly harder to imagine how someone might be assuming any causal connection between the object and the experience—unless maybe the scientist in Snake is being regarded as a causal intermediary, who, in the actual causal sequence but not in the counterfactual causal sequence, is producing a visual experience as of a sake in the subject, because there is a snake there. My critical remark was primarily directed at the assumption that causation requires counterfactual dependence, which is neither true as a general principle, nor obviously safe to assume in these weird scenarios.

      • Pendaran Roberts

        Hi Aaron,

        I think this line of concern is a good one. It’s a very natural concern I think for a philosopher or someone of like mind to have, hence we ran that small experiment included in the response.

        Regarding the modified cases, perhaps some participants are understanding the scientist as being a causal intermediary. However, it’s harder to see how they could think that the drug is a causal intermediary.

        Basically, we think that more research is required to figure out what’s going on. It’s not obvious to us what the answer is. I’ll speculate some just for fun and perhaps to provoke interesting discussion.

        It might just be that some participants are working with a weaker sense of ‘see,’ where seeing just requires a no blockers condition. Other participants might be working with the stronger sense more familiar to philosophers. If this is the case, perhaps there is no real issue here for the causal condition for perception in the stronger sense, but we can talk of perception in a weaker sense too. And perhaps the fact that both these senses exist is something that philosophers have failed to pick up on or fully understand? That could be an interesting result.

        Alternatively, perhaps conceptual truths are not all equally easy to determine by looking at thought experiments. Some conceptual truths are harder than others. Philosophers are better at examining thought experiments than non-philosophers, and the folk who think seeing occurs in the non-blocker cases are mistaken. This is an instance of the expertise defense, for which a lot of literature exists. Williamson is its famous proponent.

        I think that in many cases the results of experiments that show unusual or unexpected results can be explained by aspects of how our language works that are not relevant to the question at hand. In a sense, this is an instance of the expertise defense. Participants are not wrong exactly, but they don’t know what we’re asking. They aren’t interpreting things correctly. The complexities of language make it hard to get non-philosophers to pay attention to the right aspects of the cases and to understand the cases as a philosopher would. Do I think that something like this is going on in our study? Perhaps, but it is not obvious to me that it is. Attempts to clarify things (the modified cases in the response) did not stop significant numbers of participants agreeing that seeing occurs.

        There is perhaps another way that the expertise defense could work here. Some conceptual truths are just harder, simpliciter. They require more time and reflection. It’s like saying that some things are just harder to see than others. A chameleon in a tree is harder to see than a monkey. People without experience spotting chameleons in trees are likely to do worse at this task than chameleon spotting experts. This isn’t because they don’t understand what the task is or what they’re supposed to be doing. It’s not an understanding problem. It’s a perception problem. Likewise, perhaps what we have here is an intuition problem. Some truths are harder to intuit than others. Participants are not misunderstanding the cases, but their intuitions misfire anyway.

        I guess at the moment I kind of like this last option.

  3. Wes

    If I’m following this correctly, you find that aside from the positive control case Tomato, participants mostly significantly deny or are at chance for seeing across the experiment, and Clock did not come out different from the negative control Elephant. So, is this much evidence against the causal condition? Additionally, it is troubling that you removed half the participants in your study due to confusion with your materials and that the pairs are so unmatched. For example, in some stims you say “looking at” and in others “looks to you as if” or prime people with phrases like “unbeknownst to you” or that the object is “exactly” or “really” there. Are you worried what you are seeing here amounts to wording effects?

  4. Pendaran Roberts

    Hello Wes,

    Thanks for the comment. I’ll respond to one half and then the other.

    “If I’m following this correctly, you find that aside from the positive control case Tomato, participants mostly significantly deny or are at chance for seeing across the experiment, and Clock did not come out different from the negative control Elephant. So, is this much evidence against the causal condition?”

    Perhaps, perhaps not. It depends on what you think it takes for something to be a conceptual truth. We have a section in the discussion of the paper entitled ‘Conceptual truth,’ which might help to answer your question. The problem for the causal condition being a conceptual truth is that for the clock and snake cases many strongly hold that seeing occurs.

    I think that the best way to save the causal condition being a conceptual truth is to hold that the non-blocker cases are conceptually harder than the blocker cases, but it is not immediately clear to me how to go about proving this (or even what it means exactly, beyond a rudimentary, intuitive sense of the idea that I have).

    “Additionally, it is troubling that you removed half the participants in your study due to confusion with your materials and that the pairs are so unmatched. For example, in some stims you say “looking at” and in others “looks to you as if” or prime people with phrases like “unbeknownst to you” or that the object is “exactly” or “really” there. Are you worried what you are seeing here amounts to wording effects?”

    We wanted to bias our sample towards participants who understood the cases. We had three comprehension checks for each case, and participants looked at 6 cases (in a random order). If participants got just 1 of these 18 comprehension items incorrect, they were kicked out of the study. In other words, we required them to get 100% of 18 items. So, it is not that surprising to us that we had to kick out roughly half of the sample. We had very high requirements!

    Regarding the second half of your comment, it is possible that there were wording effects. However, I do not personally think that these effects explain the results. Notice that we reproduced similar results in the experiment in Section III of the response, and the cases here are modified from the original cases. I suspect that the effect we are finding is real. I encourage more work to be done on this subject. I find our results puzzling but not easily explained away.

  5. Wes

    Hi Pendaran,

    Thanks for your response. “The problem for the causal condition being a conceptual truth is that for the clock and snake cases many strongly hold that seeing occurs”. Your argument seems to rely on seeing 21%/15% at 10. That is interesting of course, but you can’t just cherry pick responses without looking more fully at these data.

    On the issue of “comprehension”, it’s very difficult to say what “bias” this extremely large task demand may have introduced into your study. You can’t assume your requirements had the effect you intended. Were the results different when the entire sample was included in the analysis?

    On unmatched pairs in your stimuli, I did not see any details of other experiments in Section 3, can you provide them?

    Lastly, I might have missed where you reported this, was there an interaction effect?

  6. Pendaran Roberts

    Hi Wes,

    The other experiment I mention is actually in Section III of the response we wrote.

    Regarding interaction effects, our analyses were conducted over SPSS using Friedman’s test, and a limitation of this is that interactions are not calculated.

Comments are closed.

Back to Top
%d bloggers like this: