Nozick's experience machine is dead, long live the experience machine!

Abstract

Robert Nozick's experience machine thought experiment (Nozick's scenario) is widely used as the basis for a “knockdown” argument against all internalist mental state theories of well-being. Recently, however, it has been convincingly argued that Nozick's scenario should not be used in this way because it elicits judgments marred by status quo bias and other irrelevant factors. These arguments all include alternate experience machine thought experiments, but these scenarios also elicit judgments marred by status quo bias and other irrelevant factors. In this paper, several experiments are conducted in order to create and test a relatively bias-free experience machine scenario. It is argued that if an experience machine thought experiment is used to evaluate internalist mental state theories of well-being, then this relatively bias-free scenario should be used over any of the existing scenarios. Unlike the existing experience machine scenarios, when this new scenario is used to assess internalist mental state theories of well-being, it does not provide strong evidence to refute or endorse them.

Keywords::

• Experience Machine
• Experimental Philosophy
• Hedonism
• Status Quo Bias
• Well-Being

Acknowledgements

Thanks to Paul Jose for very helpful advice on the statistical analysis, and thanks to Nicholas Agar, Ramon Das, Simon Keller, David Gilbert, Andrew Moore, Simon Clarke, and anonymous referees for Philosophical Psychology for helpful comments on earlier drafts.

Notes

^[1] The following are just a sample of the authors who have stated or implied that the experience machine thought experiment is a knockdown refutation of hedonism or all internalist mental state theories of well-being: Attfield (1987, p. 33); Baggini and Fosl (2007, pp. 74–76); Becker (1992, p. 25); Bok (2010, pp. 24–28); Brink (1989, pp. 223–224); Brülde (2007, pp. 26–29 & 33); Darwall (1997, pp. 162 & 178); Feldman (2002, p. 615); Finnis (1980, p. 33; 1983, pp. 37–42); Griffin (1986, pp. 9–10); Hausman (2010, p. 329); Haybron (2008, p. 21); Hooker (2000, p. 39); Hurka (2011, pp. 68–70); Jollimore (2004, pp. 333–334); Kagan (1998, pp. 34–36; 2009, p. 253); Kazez (2007, pp. 51–54); Keller (2009, p. 657); Kraut (2007, pp. 124–126); Kymlicka (1990, pp. 13–14); Nozick (1989, pp. 99–117); Railton (1984, pp. 148–149); Rivera-López (2007, p. 75); Sobel (2002, p. 244); Thomson (1987, p. 41); Tiberius (2004, p. 311, n. 4); Tiberius and Hall (2010, pp. 214–215); van Wyk (1990, p. 109).

^[2] It is often assumed that all hedonistic theories of well-being must be internalist mental state theories, but externalist theories have been suggested. See Feldman's truth-adjusted intrinsic attitudinal hedonism (2004, pp. 112–114) for a contemporary example and Weijers (2011a) for discussion of the different types of hedonism. See Weijers (2011b) for an explanation of the experience machine objection to hedonism.

^[3] Status quo bias is best defined as an inappropriate preference for things to remain the same (Bostrom & Ord, 2006).

^[4] That is, winning lots of money tends to make people feel happier at least until they become accustomed to their newfound wealth or squander it away and revert back to feeling about as happy as they did before their windfall.

^[5] Since deliberative judgments are open to introspection, and intuitive judgments are not, judgments that appear to be formed because of a reason are more likely to have a large deliberative component and judgments that appear to be formed without any immediately obvious reason are more likely to have a large intuitive component. See Weijers (forthcoming) for the difference between deliberative and intuitive cognition in relation to thought experiments.

^[6] Intuitive cognition is opaque because as a process, it is unconscious until the very end point, when we experience a positive or negative visceral sensation or “gut feeling” (Lieberman, 2000; Myers, 2002; Weijers, forthcoming; Woodward & Allman, 2007).

^[7] See for example Druckman (2001); Gilbert (2006); Gilovich, Griffin, and Kahneman (2002); Kahneman, Knetsch, and Thaler (1991); Kahneman and Tversky (2000); Samuelson and Zeckhauser (1988); Tversky and Kahneman (1991). See Eidelman and Crandall (2012) for a useful summary of the psychological research on status quo bias.

^[8] This line of argument is developed further in Weijers (2012a).

^[9] Unfortunately, De Brigard did not test Nozick's scenario on any of his sample groups, so we can't easily get an idea of exactly how large the difference is between the responses to Nozick's scenario and to his reversals of it. The best we can do is to compare the results of De Brigard's Negative scenario with my test of Nozick's scenario. Less than 13% (3/24) of the participants responding to the De Brigard's Negative scenario and about 84% (66/79) of participants responding to my test of Nozick's scenario reported preferring to live in reality. This difference is certainly large, but it's impossible to say how much of the difference is caused by varying characteristics between the sample groups and other experimental conditions.

^[10] Where ‘reality’ refers to living in direct contact with reality in a way that we can freely interact with it.

^[11] Nozick doesn't explicitly state that your life has average experiences, but I'm assuming that most people who read Nozick's scenario are roughly average in experiences.

^[12] My Self scenario is discussed in more detail in section 5.

^[13] Smith (2011) criticizes De Brigard's (2010) experiments on several grounds, including: the sample groups being too small; a lack of information about the procedure of the experiment; and a significant disanalogy between De Brigard's scenarios and Nozick's scenario. The disanalogy, which involves an imbalance in how relationships with others and social capital is treated, is also likely to affect judgments about Kolber's (1994) scenario. Smith's own experiments demonstrate that the disanalogy may significantly enhance the appearance of status quo bias in De Brigard's results. But even if this effect could be taken into account, De Brigard's results would still indicate that the respondents had been non-negligibly influenced by status quo bias. Smith also expresses concern with the representativeness of De Brigard's all-student sample, but since this is also a potential issue for my experiments, it is addressed in note 29 below. Finally, Smith (2011) also argues that all surveys on experience machine thought experiments are destined to fail because survey respondents cannot fully adopt the mental states of an agent confronted with such an important decision. Weijers (2012b) argues against this objection at length, so it will not be discussed here.

^[14] The experiment consisted in handing out surveys to four different classes of first-year students on August 15, 2012. The surveys were handed out at the beginning of two lectures and the end of the other two. Each survey sheet contained one of seven different experience machine scenarios. No participants responded to more than one scenario, and participants who had completed earlier versions of one of the scenarios in a previous experiment were not counted. I was unknown to the vast majority of the respondents and did not communicate with them before or during the experiment except to tell them that participation was not mandatory, not linked to their grades, and completely anonymous, etc. Two classes were marketing and two were philosophy. Using a one-tailed Fisher's exact test, I found no statistically significant differences between the results for philosophy students and marketing students (the difference that was closest to being significant was for responses to Nozick's scenario: p-value = 0.143). This methodology assumes that we expect to see a relationship (a difference in the particular direction that was seen in the results), and this result means that we can only be about 85% confident that the difference between the groups is not a product of chance. A 95% level of confidence is usually required to consider the result “statistically significant.” Furthermore, since the differences between the philosophy and marketing groups vary in different directions for different (but relatively similar) scenarios and we lack a theory of why we would expect there to be a difference between these groups in a particular direction, a two-tailed Fisher's exact test should probably be used. Using a two-tailed Fisher's exact test, we observe: p-value = 0.229, which means that we can only be about 77% confident that the difference between the groups is not a product of chance.

^[15] There was also a third question, asking their gender, for a separate research project.

^[16] 12% (8/66) of the survey forms on Nozick's scenario provided un-informational main justifications.

^[17] These main justifications demonstrated imaginative resistance because their experiences would seem unpredictable and surprising while connected to an experience machine.

^[18] These stated justifications might not be the real reason for the choice the participants actually made because the main cause of their judgment could have been subconscious—see Weijers (forthcoming) for more on this. Regardless, the justifications given match up to the reasons that some philosophers have predicted would impact participants' judgments and it might be the case that the judgments of the participants who confabulated their justification are somewhat offset by those of the participants who couldn't articulate the real cause of their judgment.

^[19] See note 14 for more details about the experiment.

^[20] Using a one-tailed Fisher's exact test, we observe (p-value = 0.008), which means that we can be more than 99% confident that the difference in the responses between these two scenarios is not the product of chance (when we assume that there will be a difference in the particular direction it was in the results).

^[21] This assumption will not sound reasonable to anyone who thinks that the vast majority of these justifications are likely to be confabulations—vain attempts by our deliberative mind to explain the verdict of our intuitive cognition. Readers with this belief needn't accept this part of the argument because they are likely to be more heavily swayed by the main argument about the effects of status quo and other biases on our judgments about experience machine scenarios.

^[22] See note 14 for more details about the experiment.

^[23] Using a one-tailed Fisher's exact test, we observe (p-value = 0.015), which means that we can be nearly 99% confident that the difference in the responses between these two scenarios is not the product of chance (when we assume that there will be a difference in the particular direction it was in the results).

^[24] The exact changes in the Cousin scenario (from the Stranger scenario) are as follows: the second sentence in the scenario is replaced with: “you have just heard that your cousin Boris has been offered a permanent spot in an Experience Machine.” The exact changes in the Friend scenario (from the Stranger scenario) are as follows: the second and third sentences in the scenario are replaced with: “you have just heard that your friend has been offered a permanent spot in an Experience Machine. Although you would never actually tell him your opinion, you are trying to decide if you think he should accept.” All of the remaining six mentions of “Boris” in the Stranger scenario are replaced with “Y/your friend” in the Friend scenario.

^[25] Using a one-tailed Fisher's exact test, we observe (p-value = 0.457), which means that we can be nearly 55% confident that the difference in the responses between these two scenarios is not the product of chance (when we assume that there will be a difference in the particular direction it was in the results).

^[26] The previous version of this experiment was conducted in 2011 on first-year philosophy and business students at Victoria University of Wellington. The only difference in the relevant scenarios was the treatment of autonomy. The Stranger NSQ' and Stranger' scenarios included the following text: “when in the machine, you still made autonomous decisions and faced tough situations, such as striving for your goals and feeling grief, but your experiences were vastly more enjoyable and varied.” This text was replaced in all of my scenarios discussed in this article with the following: “when in the machine, it still felt like you made autonomous decisions and occasionally faced tough situations, such as striving for your goals and feeling grief, although you didn't really do these things. Your experiences were also vastly more enjoyable and varied in the machine.”

^[27] Over 60% (75/124) of participants responding to the Stranger NSQ' scenario thought Boris should connect to an experience machine, compared to about 48% (45/93) of participants responding to the Stranger' scenario. Using a one-tailed Fisher's exact test, we observe (p-value = 0.051), which means that we can be about 95% confident that the difference in the responses between the Stranger NSQ' and the Stranger' scenarios is not the product of chance (when we assume that there will be a difference in the particular direction it was in the results).

^[28] Using a one-tailed Fisher's exact test, we observe (p-value = 0.008), which means that we can be over 99% confident that the difference in the responses between these two scenarios is not the product of chance (when we assume that there will be a difference in the particular direction it was in the results).

^[29] I say “assuming” here because it is quite possible that I have inadvertently elicited some additional biases or other irrelevant factors in my new scenarios that were not existent in Nozick's scenario. Furthermore, it might be argued that my all-student samples are unrepresentative of reasonable people generally. However, Smith's (2011) test of his Pretend Neutral scenario on a two-country, multi-setting, and multi-demographic group of participants produced a fairly similar result to De Brigard's (2010) test of his analogous Neutral status-quo-emphasized scenario on a mono-country, mono-setting, and all-student group (71% and 59% respectively). Furthermore, most of the 12% difference between these two results might be explained by the extra emphasis that Smith put on having to start life anew in reality in his Pretend Neutral scenario. Therefore, there is at least some evidence to believe that student samples are fairly representative of reasonable people generally and that my results cannot be dismissed out of hand for being unrepresentative.

^[30] An alternate option to experience machine thought experiments for addressing questions about well-being might be to ask simple questions like “if you never directly or indirectly experience the effect of an event, can that event make your life go better or worse for you?” In my experience of asking these kinds of questions to anyone other than trained philosophers, they confidently offer answers that often contradict answers that they would give to related questions, such as “if you are insulted behind your back, but never directly or indirectly experience any of the effects of the insult, does being insulted in this way make your life go worse for you?” Since an example provides more information than just a principle, many philosophers believe that thought experiments can elicit more considered responses than consideration of principles in isolation. Thank you to an anonymous reviewer for bringing this to my attention.

^[31] We have reason to believe that the Stranger NSQ scenario is not much more complicated than Nozick's scenario because the number of un-informational responses to each is very similar. About 14% (11/77) of participants responding to the Stranger NSQ scenario gave un-informational responses, only very slightly more than the 14% (11/77) of participants responding to Nozick's scenario.

Additional information

Notes on contributors

Dan Weijers

Dan Weijers is a Postdoctoral Fellow at Victoria University of Wellington.