Surveying Freedom: Folk Intuitions about free will and moral responsibility

Abstract

Philosophers working in the nascent field of ‘experimental philosophy’ have begun using methods borrowed from psychology to collect data about folk intuitions concerning debates ranging from action theory to ethics to epistemology. In this paper we present the results of our attempts to apply this approach to the free will debate, in which philosophers on opposing sides claim that their view best accounts for and accords with folk intuitions. After discussing the motivation for such research, we describe our methodology of surveying people's prephilosophical judgments about the freedom and responsibility of agents in deterministic scenarios. In two studies, we found that a majority of participants judged that such agents act of their own free will and are morally responsible for their actions. We then discuss the philosophical implications of our results as well as various difficulties inherent in such research.

Keywords:

• Free Will
• Moral Responsibility
• Folk Psychology
• Experimental Philosophy

Acknowledgements

For their helpful suggestions, we would like to thank Joshua Knobe, John Doris, Tamler Sommers, Mike McKenna, Brandon Schmeichel, George Graham, two anonymous referees, and, especially, Shaun Nichols, Manuel Vargas, and Al Mele.

Notes

 For discussions see, for example, Jackson (1998) and essays in DePaul and Ramsey (1998).

 See, for example, Adams and Steadman (2004), Doris and Stich (forthcoming), Knobe (2003a, 2003b, 2004), Mele (2003), Nadelhoffer (2004, 2005), Nahmias, Morris, Nadelhoffer, and Turner (forthcoming), Nichols (2004), Nichols, Weinburg, and Stich (2003), Weinburg, Nichols, and Stich (2001), Woolfolk, Doris, and Darley (forthcoming), and Turner and Nahmias (in preparation).

 However, Jackson's example of an uncontroversial case, Gettier examples in epistemology, has in fact been challenged by empirical results suggesting that people from different cultures or of different socio-economic statuses have different intuitions about such examples (see Nichols et al., 2003; Weinburg et al., 2001). See also Stich and Weinberg (2001).

 See also Smilansky (2003, p. 259), Pereboom (2001, p. xvi), O’Connor (2000, p. 4), and Campbell (1951, p. 451).

 Frankfurt cases aim to pump the intuition that an agent can be responsible for an action even if he could not do otherwise: for instance, if there were a neuroscientist ready to manipulate the agent to do A if he were not going to do it on his own, then the agent could not do otherwise than A, but if the agent did A on his own without the neuroscientist intervening, it seems the agent is nonetheless responsible for A-ing.

 Lycan (2003) argues that it is the incompatibilist who must provide a positive argument for her position, in part because “numerous commonsense claims of free action will always be more plausible than are the purely philosophical premises” of incompatibilist arguments (p. 120).

 Compatibilists and libertarians offer a parallel set of conflicting, but empirically untested, claims about the phenomenology of free will—for instance, our experiences of the ability to do otherwise and of the self as source of actions (see Nahmias, Morris, Nadelhoffer, & Turner, 2004).

 See also Nahmias et al. (forthcoming), Nichols (2004), and Nichols and Knobe (in preparation).

 Compare Fischer and Ravizza's (1998) conception of how to understand moral responsibility: “we shall be trying to articulate the inchoate, shared views about moral responsibility in (roughly speaking) modern, Western democratic society … . Here we shall be identifying and evaluating ‘considered judgments’ about particular cases—actual and hypothetical—in which an agent's moral responsibility is at issue” (pp. 10–11).

 For a helpful discussion of ‘revisionism’ in the free will debate see Vargas (forthcoming).

 For instance, Graham and Horgan (1994) offer a contextualist sort of explanation for why people might express incompatibilist intuitions in certain philosophical contexts, and Velleman (2000) developed an error theory for why our epistemic freedom leads us to believe we have metaphysical (libertarian) freedom.

 We should note that two of the four authors are compatibilists and two are incompatibilists. Every effort was made not to prejudice the data through misleading experimental design, and the diversity of opinions within the team was beneficial to this end (though frustrating at times!). We also sent our surveys to a dozen or so philosophers with various views in the free will debate and received no responses suggesting that the surveys were in any significant sense misleading or problematic.

 Examples of participants’ definitions of ‘determinism’ include: “Being unable to choose”, “That people have a set fate”, and “The lack of free will”. Many others thought it meant ‘determined’, as in ‘resolute’. Fischer (1994, p. 152) predicted that the phrase ‘causal determinism’ may seem threatening to people but that this alone would not indicate they have incompatibilist intuitions.

 Laplace defines ‘determinism’ in terms of an intelligent being such that if it had complete knowledge of the laws of nature and the current state of the universe, it could know all past events and predict all future events. Van Inwagen's (1983) technical definition of ‘determinism’ is related: a proposition describing the complete state of the universe at one time and the laws of nature logically entails a proposition describing the complete state of the universe at any other time. Laplacean determinism entails van Inwagen determinism, but the converse is not true.

 Participants were drawn from an Honors student colloquium and several introductory philosophy classes at Florida State University (before studying the free will problem). Any participants who indicated that they had taken a previous college philosophy course were excluded from the results. We also excluded those participants who missed the manipulation check and the few who answered “I don’t know” in response to the experimental question.

 The reasons participants offered for believing the scenario to be impossible were wide-ranging, including, for instance, that the computer could never acquire that much data, that people could undermine the predictions by learning about them, and that chaos theory or quantum theory makes such predictions impossible. See section 3 for a discussion of the problem of surveys that require conditional reasoning.

 This result (16:5) is statistically significant using a χ² goodness-of-fit test: χ²(1, N = 21) = 5.76, p = 0.016.

 For all of these cases we used different sets of participants to avoid any order or interference effects due to participants’ attempting to keep their judgments consistent across different cases—i.e., any given participant read just one survey and answered just one experimental question.

 Case 2 results (15:7) approach statistical significance: χ²(1, N = 22) = 2.9, p = 0.088). Case 3 results (15:4) are statistically significant: χ²(1, N = 19) = 6.37, p = 0.012. Binomial proportion comparison tests show no significant differences between the response patterns of participants in case 1 and case 2 (p = 0.56), or between those in case 1 and case 3 (p = 0.84).

 Case 4 results (15:3) are significant: χ² (1, N = 18) = 8, p = 0.005. Case 5 results (16:2) are significant: χ²(1, N = 18) = 10.9, p = 0.001. Binomial proportion comparison tests show no significant differences between the response patterns of participants in case 1 and case 4 (p = 0.58), between those in case 1 and case 5 (p = 0.30), or between those in case 2 and case 5 (p = 0.12).

 Binomial proportion comparison tests between the response patterns of participants in case 1 and those in case 6 (14:7) show no significant difference (p = 0.50).

 Binomial proportion comparison tests between the response patterns of participants in case 2 and those in case 7 (8:13) show that there was a statistically significant difference (p = 0.05).

 Binomial proportion comparison tests between response patterns of participants in case 3 and those in case 8 (6:8) show a statistically significant difference (p = 0.03).

 To test this interpretation, it would be better to have the same participants making judgments about both the ability to do otherwise and free will (or moral responsibility). See Woolfolk et al. (forthcoming) for empirical results suggesting that ordinary people have Frankfurtian intuitions.

 We suspect that it is particularly difficult to probe intuitions about modal concepts such as the ability to do otherwise (see note 35). See Turner and Nahmias (in preparation).

 Participants’ explanations of their answers suggest that some of them even had fledgling compatibilist theories in mind. For instance, some suggested that Jeremy acts freely because no outside forces compel him, because he controls his actions or consciously decides to do them, or because the prediction is based on what he decides to do. (Similarly, some of the minority who judged that Jeremy is unfree offered fledgling incompatibilist arguments.)

 We should add that in numerous pilot studies as we fine-tuned descriptions of the scenarios, experimental questions, and manipulation checks, we were surprised by the consistency of the results: in almost every set of surveys, 70–85% of participants judged that Jeremy acts of his own free will or that he is morally responsible for his action.

 We should note that we tried several variations of the wording of the Jeremy case, including one that said, “The computer then deduces from this information and the laws of nature that it is physically impossible for Jeremy to do anything other than to rob Fidelity Bank [save the child, etc.]” and results showed the same trends (as reported in note 28). For further discussion of this concern about the salience of determinism, see section 3.

 We wonder whether some participants, having reconciled themselves to the problem of God's foreknowledge and free will, were less inclined to see the supercomputer's foreknowledge as a threat to free will. We obtained information about participants’ self-reported religiosity, but we have not yet correlated this information with the results.

 Methods for study 2 were the same as study 1 except that participants answered two experimental questions (one about Fred and one about Barney). Fewer participants judged that the scenario was impossible and fewer missed the manipulation check (three participants who answered differently for Fred and Barney were not included in the analysis of results). Results for case 9 (26:8 judging that both agents acted of their own free will) were statistically significant using a χ² goodness-of-fit test: χ² (1, N = 34) = 9.53, p = 0.002.

 A binomial proportion comparison test shows no significant difference between the response patterns of participants in case 1 and those in case 9 (p = 0.98).

 Comparison tests show no significant differences between the response pattern of participants in case 9 and the pattern of blameworthy judgments in case 10 (21:14, p = 1.06), or in the response patterns of participants in case 9 and the pattern of praiseworthy judgments in case 10 (23:13, p = 0.25). Interestingly, a smaller majority of participants in these cases judged that the agents were morally responsible than those judging Jeremy to be morally responsible (in cases 4 and 5).

 Comparisons between the response patterns of participants in case 8 and those in case 11 (20:11) were not significantly different (p = 0.29). Because we asked the same participants about both Fred's and Barney's ability to do otherwise, and all participants made consistent judgments, we were unable to confirm the interesting results from study 1 suggesting that such judgments vary depending on whether the action is blameworthy or praiseworthy/neutral.

 As we suggested above, in some cases the pattern of responses suggests to us that people may have Frankfurtian intuitions whereby they are willing to assign blame despite believing the agent could not do otherwise. In other cases, the responses suggest people may be employing a conditional conception of the ability to do otherwise, such that the agents could do otherwise despite their actions being determined.

 Compare Double (1991 1996). We take very seriously the possibility that free will and moral responsibility are best analyzed as context-relative concepts (see Graham & Horgan, 1994). If so, empirical research aimed at mapping the contours of the relevant contexts would be important.

 This would suggest an interesting explanation (perhaps found in the work of William James and Ludwig Wittgenstein) for the long-standing free will stalemate. After decades of argument, compatibilists and incompatibilists tend to retain conflicting intuitions about crucial disputes in the debate; perhaps they end up where they do in part because they start off where they do—as ordinary folk with different intuitions about the concept of free will and its relationship to determinism.

 Libertarian theories require, at a minimum, indeterministic causation—in just the right place (Kane, 1996)—and often also an ontologically distinct type of causation, agent causation, that demands either substance dualism or emergent causation (O’Connor, 2000). Furthermore, in general, claims about the incompatibility of any two concepts require more evidence to establish than claims about their compatibility, in part because concepts that are not obviously incompatible should be assumed to be compatible barring an argument to the contrary. (See Lycan, 2003; Nahmias et al., forthcoming.)

 See Weinburg et al. (2001) and Nichols et al. (2003) for experimental work on epistemological intuitions suggesting that these intuitions vary among people with different educational backgrounds and between people from American and Asian cultures.

 Much of the work on people's “folk theories” of biology, physics, and psychology (e.g., theory of mind) involves such experiments. Some find differences between participants’ reported intuitions about the relevant situations and their actual behavior in such situations (see, for example, Keysar, Lina, & Barr, 2003).

 We ran pilot studies suggesting this. Scenarios describing an agent who was neurally manipulated to deliberate and act in a certain way elicited almost unanimous judgments that he was neither free nor responsible. When combined with the results from studies 1 and 2, this suggests to us that people do not have the intuition, sometimes advanced by incompatibilists, that determinism is no different than such covert manipulation (see Pereboom, 2001), though further tests of this question would be useful.

 Similar worries would apply to any description of determinism that suggests that the laws of nature control or constrain us, that our actions are fated by the past, that our conscious decisions are causally inefficacious, etc. Such question-begging descriptions of determinism would not show that the folk are incompatibilists about free will and determinism but only that they are incompatibilists about free will and constraint, fate, epiphenomenalism, etc. Notice that introductions to the problem of free will often use such misleading images to portray the threat of determinism.

 Nichols (2004) used a roughly similar set-up and found that adults and children are more likely to judge that a person could behave otherwise than that a physical object could. Nichols suggested that these results indicate that people have a theory of agent causation, but for a variety of reasons we think such a conclusion is too hasty. (See Turner & Nahmias, in preparation.)

 Compare Peter Strawson's (1962) claim that the reactive attitudes are “natural, original … in no way something we could choose or could give up” (note 7).

 This view is similar to van Inwagen's (1983) view, dubbed “metaphysical flip-flopping” by Fischer & Ravizza (1998), that if he were convinced determinism were true, he would become a compatibilist rather than a skeptic about free will.

 Kane (1999) suggests that students “subjected to [a compatibilist] argument may have the uneasy feeling they are being had” (p. 218). O’Connor (2000) writes, “Does freedom of choice have this implication [that causal determinism is false]? It seems so to the typical undergraduate on first encountering the question” (p. 4). We wonder whether these philosophers’ students have these reactions in part due to the incompatibilist sympathies of their professors.

 In fact, we ran one such test. One of the authors (E.N.), who happens to be a compatibilist, ran the Fred and Barney survey on his Intro students soon after a two-week section on the free will debate. The results, it turned out, were not significantly different from the results garnered from ‘untrained’ participants: 83% of the ‘trained’ participants judged that Fred and Barney acted of their own free will, where 76% of untrained participants had made such judgments). Participants were asked whether studying the free will debate had influenced their judgments, and roughly half said it had.

 Though we need to be wary of selling the folk short. My (E.N.) furniture delivery men, on discovering I was a philosopher, told me they had solved the old riddle of whether the cup was half full or half empty: “It's half full,” they explained, “if it's been filled halfway up and it's half empty if it's been emptied halfway.” An excellent example of contextualist semantics!