Content uploaded by Igor Douven
Author content
All content in this area was uploaded by Igor Douven on Sep 26, 2017
Content may be subject to copyright.
Nozick’s Experience Machine: An Empirical Study*
Frank Hindriks
Faculty of Philosophy, University of Groningen
f.a.hindriks@rug.nl
Igor Douven
SND/CNRS/Paris-Sorbonne University
igor.douven@paris-sorbonne.fr
Abstract
Many philosophers deny that happiness can be equated with pleasurable experi-
ences. In inuential work, Nozick introduced an experience machine thought
experiment to support the idea that happiness requires pleasurable experiences
that are “in contact with reality.” In this thought experiment, people can choose
to plug into a machine that induces exclusively pleasurable experiences. We test
Nozick’s hypothesis that people will reject this offer. We also contrast Nozick’s
experience machine scenario with scenarios that are less articial and offer options
which are less invasive or disruptive than being connected to a machine, specically
scenarios in which people are offered an experience pill or a pill that improves
overall functioning.
In order to show that happiness cannot be equated with pleasurable experiences, Robert
Nozick (1974) invented a thought experiment involving an experience machine. In this
thought experiment, we are to imagine that we can choose to plug into an experience
machine that ensures our having exclusively pleasurable experiences. Nozick (1974: 646)
supposes that we will reject this offer, because we want to live a life that is “in contact
with reality” (more on this below). is paper takes Nozick’s thought experiment as a
starting point for a survey study, aiming to thereby determine the level of support for the
intuition this thought experiment seeks to bring into relief.
Empirical work on Nozick’s “contact intuition” revealed that this intuition is not as
universally shared as many in the philosophical community have supposed (De Brigard
2010, Weijers 2014). We probe deeper by considering variations of Nozick’s original
scenario that feature interventions which are both more realistic and less invasive than
hooking someone up to a machine. Specically, in Experiment 1 we also present the
option of taking an experience pill that has the same effect as the experience machine as
well as the option of taking a functioning pill that improves one’s overall functioning.
*
All Supplementary Information as well as all data and the script for the statistical analyses are available
at https://osf.io/nrx2x/?view_only=d001aeff6ea24503b7e844e05f71419f.
1
Inspired by Nozick, we predict that the less invasive an intervention is—the less it severs
contact with reality—the more people will be prepared to accept it. We contrast our
ndings with those of the aforementioned earlier empirical studies. In a second experi-
ment, we consider still further variations of Nozick’s scenarios, which are parallel to the
materials from Experiment 1 except that now instead of offering an improvement over
the status quo the scenarios promise to prevent a pending loss. Recent research suggests
that the difference in “valence” between the scenarios from Experiment 1 and those
from Experiment 2 may have an effect on people’s willingness to accept the interventions
on offer. We investigate whether it does indeed by comparing the data from the two
experiments.
1 eoretical background
Hedonism is the view that happiness can be reduced to pleasurable experiences. When
considering this position, Nozick (1974: 644) asks: “What else can matter to us, other
than how our lives feel from the inside?” A hedonist will be inclined to read this as a
rhetorical question with “nothing” as the obvious answer. Nozick (1974: 646), however,
rejects hedonism and suggests instead: “Perhaps what we desire is to live (an active
verb) ourselves, in contact with reality. (And this machines cannot do for us.)” Dan
Haybron (2011: 27) formulates Nozick’s key point as follows: “Beyond having positive
mental states, it seems to matter both that our lives go well and that our state of mind
is appropriately related to how things are.” Nozick’s point appears to be that people
intrinsically value contact with reality in the sense that they value their mental states
being veridical and their experiences being genuinely of our world.
Nozick mentions a number of things other than pleasurable experiences that matter
to us. It matters to people that they do something rather than merely have experiences.
Furthermore, it matters to them what kind of person they are, whether, for instance, they
are courageous, kind, or intelligent. In light of this, Nozick (1974: 645) even concludes
that “[p]lugging into the machine is a kind of suicide.” e nal consideration he
mentions is that people care about being in touch with a reality that is deeper or more
important than one that is manufactured, one that is real rather than articial (ibid.).
1.1 Previous research
us far, empirical research concerning the experience machine has focused on factors
that might bias people’s responses to Nozick’s scenario (De Brigard 2010, Weijers 2014).
Wayne Sumner (1992: 216) suggests that reading that scenario might trigger a number
of strictly irrelevant thoughts:
We immediately begin to imagine the ways in which things could go horribly wrong.
How do we know that the technology is foolproof? What happens if there is a power
failure? Suppose the operators of the machine are really sadistic thrill-seekers, or
the premises are overrun by fundamentalist zealots.
2
In view of these considerations, one might doubt whether the thought experiment reveals
anything about what people value rather than about the factors that bias them. As a more
general concern, Sumner (1992: 216) mentions that “[f]or the experience machine to
yield any philosophically interesting results we must imagine ourselves in a world very
different from our own—so different that any choices we make in that world might tell
us very little about how we think our lives should go in the real world.”
Sumner notes that the philosophical point of the experience machine heavily depends
on the supposition that the biasing factors have been neutralized. Dan Weijers (2014)
tries to formulate a vignette that neutralizes at least a number of such factors. e rst
step he takes is to test the scenario as Nozick originally presented it. is lengthy passage
features “superduper neuropsychologists” who stimulate your brain and pre-program
your life’s experiences; you will be “oating in a tank, with electrodes attached to your
brain” (Nozick 1974: 42–43). From an extensive library that businesses have developed
by researching the lives of many other people, you can choose the experiences you want
to have for the next two years. en you get some time to select experiences for another
two years. It is stipulated that while you are hooked up to the machine, you will not be
aware of that fact.
Weijers (2014: 520) presents the participants to his rst study with the following
question: “What is the best thing for you to do for yourself in this situation?” Participants
can choose between “Permanently plug in to an Experience Machine” and “Never plug
in to an Experience Machine,” and they are asked to briey explain their choice (ibid.).
Of 79 participants, 13 (or 16 percent) indicate that they would connect to an experience
machine. is could be taken to indicate that a vast majority of people shares Nozick’s
intuition. Many of the justications people gave for not wanting to be connected to the
machine concerned autonomy or being in touch with reality. ese justications t well
with the reasons Nozick presents for rejecting the offer.
However, Weijers goes on to note that only 47 percent of the justications concern (in
our terminology) Nozick’s contact intuition.
1
Other explanations subjects gave for their
choices suggest that their responses were biased by a number of factors. One such factor
is overactive imagination, which is exemplied by a participant writing that “the machine
seems scary or unnatural” (ibid.). Another biasing factor is what Weijers refers to as
“imaginative resistance.” Participants put forward diverse considerations that t under
this heading, including “the machine might break down or not produce great experiences
in the future,” “unpredictable or surprising experiences are better than pre-programmed
ones,” and “bad experiences are required to appreciate good experiences or to develop
properly.” Weijers also classies justications along the lines of “I can’t because I have
responsibilities to others” under imaginative resistance. He concludes: “e fact that
47 % (31/66) of the participants who chose reality (and provided an informational main
justication) stated an irrelevant reason as the main justication for their choice gives
us reason to believe that Nozick’s scenario might not be very useful for assessing the
1
Similarly, De Brigard (2010: 45) claims that 60 percent of the explanations his participants gave for
rejecting the experience machine “had nothing to do with their preference for a real life over a virtual one.”
3
relative intrinsic prudential value of reality and how our experiences feel to us on the
inside” (Weijers 2014: 521).
Subsequently, Weijers tests a vignette that he takes to be affected less by confounding
factors:
You have had a go in an Experience Machine before and know that they provide an
unpredictable rollercoaster ride of remarkable experiences. 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.
You also recall that, while you were in the Experience Machine, you had no idea
that you had gotten into a machine or that your experiences were generated by a
machine. (Ibid.)
Almost all of the confounding factors mentioned above are addressed in this vignette in
that it is explicitly stipulated that they play no role. Some others are addressed by the
phrasing of the question:
Ignoring how your family, friends, any other dependents, and society in general
might be affected, and assuming that Experience Machines always work perfectly,
what is the best thing for you to do for yourself in this situation? (Ibid.)
Now 34 percent of the participants chose to connect to the experience machine, and only
31 percent provided an irrelevant justication for their choice.
Weijers goes on to investigate loss aversion as a potential source of bias. People who
experience loss aversion assign greater value to avoiding losses than to securing gains.
As it happens, such asymmetries in how people evaluate options all but disappear when
it comes to evaluating the options faced by strangers. In view of this, Weijers contrasts
the scenario just mentioned, in which a subject is asked what would be best for him or
her to do, to a version in which the subject is asked what would be best for a stranger to
do. In this Stranger scenario, 52 percent of the participants deem it best for the stranger
to connect to the experience machine against 34 percent in the earlier scenario to which
Weijers now refers as “the Self scenario.” Weijers explains this difference in terms of loss
aversion, arguing that when subjects evaluate the choice faced by a stranger, they are less
sensitive to risk and more impartial.
Loss aversion can also occur because the subject perceives the status quo as substan-
tially less risky than the alternative presented. In Nozick’s scenario, reality is the status
quo. Perhaps the reason why at least some people prefer the status quo is not because
they intrinsically value reality. Instead, it might be because they are averse to the loss they
might experience when taking the risk of being connected to an experience machine. To
rule out or at least mitigate this bias, Weijers formulates a version of the Stranger scenario
in which the stranger has already switched a number of times between real life and a life
generated by the experience machine (without being aware of this). As the stranger has
already spent 50 percent of the recent past being connected to the experience machine,
Weijers suggests, real life ceases to be the status quo. And he hypothesizes that the status
quo bias will all but disappear when both options are framed as equally familiar. Fiy
4
ve percent of participants now say that the stranger should choose “the Experience
Machine life” rather than “real life” (Weijers 2014: 526). Even though the difference
between this Stranger No Status Quo (NSQ) scenario and the Stranger scenario discussed
previously is not statistically signicant, there is a signicant difference with the Self
scenario (𝑝 = .008, using a one-tailed Fisher’s exact test; Weijers 2014: 533 n28).
Weijers concludes that Nozick’s scenario is less useful for evaluating hedonism than
it is commonly taken to be. At the same time, he is optimistic about the extent to which
the scenarios he uses succeed in ruling out biases. In particular, he suggests that the
Stranger NSQ scenario is “relatively bias free” (Weijers 2014: 528). He goes on to argue
that “the widespread agreement about Nozick’s scenario was guided more by status quo
bias and other irrelevant factors than it was by the value of reality” (Weijers 2014: 529).
e upshot is that the experience machine provides “defeasible” rather than “decisive”
evidence against hedonism (ibid.).
Prior to Weijers, Felipe De Brigard (2010) defended the more radical claim that,
because of the status quo bias, experimental ndings concerning the experience machine
are of little use. He argued that Nozick’s scenario is inadequate as a test of hedonism. In
contrast to Weijers, however, De Brigard does not test Nozick’s original scenario. Instead,
he uses a vignette in which subjects are to imagine that, unbeknownst to them, they have
been hooked up to an experience machine for some time now. e underlying idea is that
people are averse to abandoning the life they have been experiencing thus far, irrespective
of whether it is virtual or real. is means that the status quo bias might not only explain
the negative responses people (allegedly) tend to have toward Nozick’s original scenario,
it also provides reason to expect people to respond negatively to scenarios in which they
can choose to return to reality in a situation in which they have been connected to the
experience machine for some time. De Brigard (2010) nds that people are more or less
equally divided on the question of whether they would like to remain connected to the
machine (41 % of the participants wanted to disconnect). He concludes that, as in this
case there is not a vast majority that rejects the option, people’s responses are inuenced
by some biasing factor.2
De Brigard formulates one scenario in which people are hardly divided. In this
scenario, the alternative to remaining connected is living a very unhappy life, more
specically being a prisoner in a maximum security prison.
3
When that is the alternative,
the vast majority—87 percent—choose to remain connected to the machine (De Brigard
2010: 47). is suggests that people do not value reality more than avoiding a life full
of unpleasant experiences. In light of his experiments, De Brigard (2010: 51) concludes
that “some people may prefer to remain unplugged, not because they value reality, but
because they are averse to losing their status quo.”
2
e underlying premise, which De Brigard does not test, is that the vast majority will reject being
connected to the machine in Nozick’s original scenario.
3
In the positive scenario, people read that their real life is that of “a multimillionaire artist living in
Monaco” (De Brigard 2010: 47). e neutral condition says nothing beyond “you can go back to your real
life” (ibid.).
5
One might doubt that De Brigard’s ndings pose a signicant challenge to Nozick’s
contact intuition. De Brigard (2010: 48) notes that the fact that people are divided
in the neutral and positive conditions could be taken to mean that more than half of
the participants share Nozick’s contact intuition and prefer to go back to reality in the
neutral condition.
4
is interpretation certainly ts the ndings in the positive condition.
In that condition, the quality of the experiences inside and outside of the machine is
similar, which means that hedonists could respond both ways. However, the neutral
condition seems to provide evidence against hedonism, as the machine is clearly the
better option when evaluated only in terms of experiences. In the negative condition, the
fact that the vast majority favors the machine does not establish that people exclusively
care for pleasurable experiences. Instead, it supports the idea that avoiding unpleasant
experiences matters to people, possibly as one among several things, which could include
living a life in touch with reality.
e upshot is that, even though the vast majority (84 percent) rejects the option
of being connected to an experience machine when presented with Nozick’s original
scenario, there is some reason to believe that these responses are biased by loss aversion.
Attempts to correct for this bias, for instance in the way Weijers tried to do this, result in
acceptance rates up to 55 percent (59 percent when the option is to remain connected).
1.2 Plan
To decrease the chance that people’s responses are inuenced by biasing factors such as
imaginative resistance and overactive imagination, we used short and simple scenarios
in our own studies, presented below. Weijers used Nozick’s original description of the
experience machine, which contains a number of possibly distracting details (such as,
for instance, the mentioned reference to businesses that have developed an extensive
library by researching the lives of many other people). e vignettes that De Brigard
uses are also lengthy because the situation he presents as the status quo is that of being
already connected to the experience machine.
5
Because the status quo in our vignettes
is ordinary life, our vignettes can also be brief in this respect. We accommodate the
concern that one can only experience pleasure if one also has unpleasant experiences by
stipulating that the experience machine induces experiences that are almost exclusively
4
In response, De Brigard points out that in the neutral condition, people know nothing about what
their real life will be like. He goes on to test a vignette that ends with: “your life outside is not at all like the
life you have experienced so far” (De Brigard 2010: 49). Now 59 percent of the participants want to remain
connected (against 54 percent earlier).
5
De Brigard (2010: 47): “It is Saturday morning and you are planning to stay in bed for at least another
hour when all of the sudden you hear the doorbell. Grudgingly, you step out of bed to go open the door. At
the other side there is a tall man, with a black jacket and sunglasses, who introduces himself as Mr. Smith.
He claims to have vital information that concerns you directly. Mildly troubled but still curious, you let
him in. ‘I am afraid I have some disturbing news to communicate to you’ says Mr. Smith. ‘ere has been a
terrible mistake. Your brain has been plugged by error into an experience machine created by superduper
neurophysiologists. All the experiences you have had so far are nothing but the product of a computer
program designed to provide you with pleasurable experiences’.”
6
pleasurable.
6
In this way, we also try to succinctly exclude this further potential biasing
factor.
To be sure, permitting longer vignettes gives one more opportunity to explicitly
stipulate that certain factors are irrelevant, which might make subjects discount such
factors. We believe, however, that in particular imaginative resistance can be decreased
by excluding features that are far removed from ordinary life. e question of whether
this is indeed the case can be answered, at least in part, by comparing our ndings to
those of Weijers and De Brigard, as we do below.7
2 Authenticity and valence
As intimated, the intuition Nozick sought to address with his experience machine thought
experiment is a bit broader than the idea that we intrinsically value a life in contact with
reality. Two other ideas mentioned in the context of the thought experiment are that it
matters to people who they are, and that it matters to them what they do, not only what
they experience. While it is not entirely straightforward to determine what Nozick had in
mind when he identied these three aspects, we believe that each of the aforementioned
considerations can be taken to capture an aspect of being authentic, or of living an authentic
life.
8
is notion of authenticity may be hard or even impossible to make formally precise,
but that does not make it obscure. Indeed, we take it to be clear enough that authenticity is
reected in the traits persons have, in what they do, and in how genuine their experiences
as well as their relations to the outside world are. We take it to be equally that the notion
is of interest in and of itself, regardless of what Nozick had exactly in mind. (Nozick
exegesis is not the aim of this paper.) We hypothesize that people value authenticity, and
dub this hypothesis “the authenticity intuition.”
A main source of inspiration for our own study is the observation that authenticity
admits of degrees. Someone can live a life that is more or less in contact with reality, can
6
Recall that one of the participants of Weijers’ experiment rejected the option of being connected
to the experience machine, because “bad experiences are required to appreciate good experiences or to
develop properly” (Weijers 2014: 520). De Brigard (2010: 47) seems to accommodate some such concern by
including the following in his vignettes: “All the unpleasantness you may have felt during your life is just an
experiential preface conducive toward a greater pleasure (e.g. like when you had to wait in that long line to
get tickets for that concert, remember?).”
7
Weijers (2013) argues that including less background information does not necessarily prevent people
from considering irrelevant features. People might match the scenario they read to the closest real life
experiences, which include many features that are absent from the thought experiment. However, he also
notes that “most philosophical thought experiments stipulate features that are so unrealistic that we have
not experienced anything like them or we have experienced the very opposite of them” (Weijers 2013: 22).
At least some of the scenarios that we used in our studies are actually quite close to real-life situations (e.g.,
taking a pill that improves functioning, and certainly one that prevents deterioration, is not so different
from what we hope that taking vitamin pills and other supplements does for us).
8
Griffin (1986: 19), arguing against hedonism in much the same way as Nozick does, says: “ . . . what
affects well-being can only be what enters experience, and the trouble is that some of the things that persons
value greatly do not. My truly having close and authentic personal relations is not the kind of thing that can
enter my experience; all that can enter is what is common to both my having such relations and my merely
believing that I do. And this seems to distort the nature of these values.”
7
be closer or farther removed from her true self, and act to different degrees in accordance
with her own values and desires.
9
Based on this observation, we aim to take some rst
steps in investigating whether the authenticity intuition is shared by the folk. Do ordinary
people really care about living an authentic life, or is it just philosophers falsely supposing
that people care about authenticity?
Nozick’s experience machine scenario offers an excellent starting point for this
investigation, precisely because it is not hard to come up with scenarios that are like
it in important respects but that feature interventions which steer people less far away
from normality. e experience machine is exceedingly invasive in that it completely
detaches people from the life they ordinarily lead. People’s experiences can, however, be
inuenced in less intrusive ways. For this reason, we conducted studies featuring less
invasive scenarios. In particular, rather than with the option of being hooked up to a
machine, people are presented with the option of taking a pill that induces pleasurable
experiences. Because it is less invasive, the experience pill scenario could be said to
preserve authenticity, or contact with reality, to a greater degree than the experience
machine scenario does.
One might think that the very fact that subjective experiences are inuenced directly
makes them inauthentic. is thought led us to include a scenario in which a pill
inuences the subject’s overall functioning. Improved functioning is in turn likely to
have a positive effect on the experiences someone has. As the manipulation is indirect,
this scenario featuring a functioning pill would seem to imperil authenticity to a still
lesser degree than the one featuring an experience pill. e functioning pill can be related
directly with one of Nozick’s concerns, that for being a particular person: “A second
reason for not plugging in [to the experience machine] is that we want to be a certain
way, to be a certain sort of person” (1974: 645). Just as how close someone is to reality,
being a particular person can presumably contribute to someone’s happiness. is is how
the experimental condition under discussion ts the overall experimental design. It is
loosely inspired by the transformation machine that Nozick mentions, “which transforms
us into whatever sort of person we’d like to be (compatible with our staying us)” (1974:
646). By and large, people remain the same person when their functioning is enhanced.
It is their capacities that now function better. us, Nozick’s characterization of what
people care about supports our claim that people whose functioning is improved live a
more authentic life than people whose experiences are manipulated directly.
In contrast to Weijers, we use only rst-person scenarios, scenarios in which partici-
pants are to imagine that they face the choice at issue. We do not question that in general
people are more averse to risk when they evaluate options from their own perspective as
compared to options faced by a stranger. However, we doubt that a scenario featuring a
stranger can capture a concern for authenticity. Authenticity, it seems to us, is primarily
a rst-person concern.
9
Note that this is not to say that authenticity is a measurable quantity. But although love is (probably)
not a measurable quantity, it still makes sense to say that Jim loves Harriet more than he loved Susan. All
we need to assume for the purposes of this paper is that similar comparative statements make sense for
authenticity.
8
A second source of inspiration for our study consists in a wide range of valence
asymmetries discovered in experimental philosophy.
10
As it turns out, the way in which
people apply apparently non-normative notions such as intentional action, freedom, and
happiness is sensitive to normative factors. In particular, they turn out to be sensitive to
whether or not the situation experienced or affected is good or bad. For instance, Joshua
Knobe (2003) nds that people qualify the behavior of the chairman of a company whose
business strategy happens to affect the environment as unintentional when this effect is
benecial, and as intentional when it is harmful.11
Valence also enters De Brigard’s investigation of how people respond to scenarios in
which they are to imagine themselves having been connected to an experience machine.
He compares three versions that differ with respect to the life that awaits them outside of
the machine. at life might be good, neutral, or bad. In other words, it might have a
positive, neutral, or negative valence. De Brigard nds that 87 percent of the participants
want to remain connected in the negative condition, against 50 percent in the positive
condition, and 46 percent in the neutral condition. is suggests that the negative valence
of the outcome makes people accept the option of being hooked up to the experience
machine. De Brigard takes this to be due to people’s aversion to loss. We test whether
valence matters in this way when the status quo is the ordinary life a subject leads in which
he or she is neither very happy nor very unhappy. e alternative is a life in which they
have almost exclusively unpleasant experiences. People can avoid this future by being
connected to an experience machine (or by taking an experience pill or a functioning
pill). As this pending future has a negative valence, loss aversion leads us to also expect
that when real life is the alternative, more people will want to connect to the experience
machine (or take one of the pills).
3 Research questions
e authenticity intuition can be tested by presenting people with a vignette that features
Nozick’s experience machine. For the reasons explained previously, we used a version
of Nozick’s scenario stripped down to its essence. Our rst research question (Q1)
is whether, using such a version, we can replicate Weijers’ nding that only a small
percentage of participants (16 percent, in his study) accepts the offer of being hooked up
to the experience machine.
A stronger test of the same intuition can be developed once it is recognized that
authenticity is a matter of degree. e experience machine is rather invasive in that it
completely disconnects people from reality. By considering less invasive manipulations of
people’s experiences, we can test whether people become more accepting of the envisaged
10
See Knobe (2010) for an overview. Knobe argues that a unied explanation can be provided of all
valence asymmetries; see Hindriks (2014) for a critical discussion.
11
See also Philips, Misenheimer, and Knobe (2011), who nd that people discount reports of happiness
when they live an objectively bad life, but not reports of unhappiness when they live an objectively good
life. is suggests that unpleasant experiences suffice for unhappiness, while pleasurable experiences do not
suffice for genuine happiness and less subjective values have to be present as well.
9
scenario. To this end, we formulated a vignette that features an experience pill that has
the same effect as the experience machine. In contrast to people who are hooked up to an
experience machine, people who take an experience pill still live a life in their ordinary
surroundings, meaning, among other things, that they have genuine personal relations
(see note 8). Given that people taking the pill would still live in contact with reality in
the sense that their mental states would be veridical and their experiences genuinely of
our world, we expect that people will be more inclined to take the experience pill than to
connect to the experience machine. Whether we are right to expect this is our second
research question (Q2).
Nozick’s remark regarding people caring about the kind of persons they are suggests
that part of what might be problematic about the experience machine is that their per-
sonality is not reected in the kind of experiences they have. To mitigate this feature of
the original thought experiment, we prompt people to imagine that, rather than their
subjective experiences, the way they function is enhanced by means of a pill. In other
words, by taking the functioning pill, people’s overall functioning improves substantially.
As it is their capabilities that are enhanced or their way of functioning, the experiences
they will have are more authentic than the ones that are directly induced. In the extreme
case, people might see little or no reason to discount the resulting experiences as articial.
Indeed, it could be argued that we live our lives most authentically when we can be our
best possible selves. is leads us to expect that more people will choose the functioning
pill than the experience pill. Whether this is correct is the third research question (Q3).
Underlying our expectations is the thought that being hooked up to an experience
machine is a major intervention that completely disconnects an agent from reality. Taking
a pill rather than being hooked up to a machine is less invasive. e infringement on
authenticity is even less (or might be deemed to be completely absent) when the pill
inuences people’s functioning rather than their experiences directly.
Our fourth and nal research question (Q4) concerns the valence of the outcomes.
is question is inspired by the ndings concerning loss aversion discussed in the pre-
vious section. More indirectly, it is also motivated by many recent empirical ndings
within experimental philosophy that reveal the signicance of normative factors for
apparently normatively neutral concepts (such as intentional action, causation, freedom,
and happiness). Although the interpretation of these ndings is a matter of ongoing
controversy, they suggest that normative factors such as moral valence, norm violations,
normative reasons, or responsibility attributions inuence our intuitions concerning
concepts of the designated kind.
To investigate the role of valence in the context of Nozick’s thought experiment, we
formulated vignettes in which, rather than inducing pleasurable experiences or enhancing
functioning (the positive condition), a machine or a pill prevents people from having
negative experiences or from their functioning deteriorating (the negative condition).
We hypothesize that people will be more inclined to accept measures that prevent a bad
outcome as compared to induce a good outcome, or differently put, that because people
may suffer from loss aversion (De Brigard 2010 and Weijers 2014), they will tend to
10
care less about living in contact with reality when it comes to preventing unhappiness as
compared to bringing about happiness.
4 Experiment 1
is experiment was designed to investigate questions Q1–Q3.
Participants
ere were 249 participants in this experiment. ey were recruited via CrowdFlower,
which directed them to the Qualtrics platform on which the experiment was run. e
participants were nancially compensated for their time and effort. Repeat participation
was prevented.
We rst excluded participants who returned incomplete responses sets, nonnative
speakers of English, and participants who indicated that they had not responded seriously
(see Aust et al. 2013). is le us with 224 participants. From those, we removed the
fastest 2.5 percent of responders, as well as the slowest 2.5 percent, which le us with 210
participants for the nal analysis (𝑀age = 38,𝑆𝐷age = 13; 120 females).12
Materials and procedure
Participants were divided into 6 groups, each group receiving a different question. e
questions concerned either the possibility of being connected to a machine that guarantees
pleasurable experiences or the possibility of taking a pill, where the pill either guarantees
pleasurable experiences or improves functioning. Specically, each of the following three
vignettes was offered to two groups:
Machine:
Imagine that you are presented with a choice to plug into an experience ma-
chine. Due to the machine, the experiences you will have for the rest of your life are
almost exclusively pleasurable. If you accept this option, you will be permanently
oating in a tank with electrodes attached to your brain (without being aware of
this).
Experience pill:
Imagine that you are presented with a choice to start taking an experi-
ence pill. Due to the pill, the experiences you will have for the rest of your life are
almost exclusively pleasurable. If you accept this option, you will be taking the pill
permanently on a daily basis. (e pill will have no detrimental long-term effects
on your health.)
Functioning pill:
Imagine that you are presented with a choice to start taking a pill that
affects the way you function. Due to the pill, the way in which you function in a
12
In choosing these cut-off points for slowest and fastest responders, we are following the practice of
other experimenters who have used the same cut-off points for the purpose of enhancing the quality of data
from online surveys.
11
wide range of respects will substantially improve for the rest of your life. Specically,
your physical, cognitive, and social functioning will improve (e.g., you will become
healthier, smarter, wittier, and more social). If you accept this option, you will be
taking the pill permanently on a daily basis. (e pill will have no detrimental
long-term effects on your health.)
We refer to these vignettes as MP, EP, and FP, respectively. (e P stands for “positive,”
because the vignettes refer to positive experiences and later on they will be contrasted
with vignettes that refer to negative experiences, which will be used in Experiment 2.)
For control purposes, we elicited responses in two different ways: One of the two
groups that received a given vignette was asked for a yes/no answer—“Would you choose
to plug into the experience machine/to start taking the pill?”—while the other group
was asked to respond on a Likert scale going from 1 = very unfavorable to 7 = very
favorable, the exact question being, “What attitude do you have towards plugging into
the experience machine/starting taking this pill?”
Results
Question Q1 asked whether Weijers’ nding that the vast majority of his participants
rejected the offer of being hooked up to the experience machine could be replicated
given a version of Nozick’s scenarios stripped down to its essentials. e data bearing
on this question came from the group of 35 participants who had received the MP
vignette followed by the yes/no question. Of these, 10 answered positively. is is a
considerably higher percentage (29 percent) than was found in Weijers’ (2014) study, in
which 16 percent accepted the offer of being hooked up to the machine (De Brigard’s 2010
nding was exactly the same as Weijers’). A binomial test showed that this difference in
percentage was signicant,
𝑝 = .043
(one-sided).
13
On the other hand, it is still the case
that a vast majority rejected the offer of being connected to the machine.
To answer questions Q2 and Q3, we started by comparing the response frequencies
to the three yes/no questions. ese are displayed in Figure 1. is gure shows that the
proportion of positive responses to the question concerning the functioning pill is very
high and (much) higher than the corresponding proportions for the other two conditions,
and further that the proportion of positive responses to the question concerning the
experience pill is considerably higher than the proportion of positive responses to the
question concerning the machine. To be exact, the percentage of positive responses in the
EP condition was 53 and that in the FP condition, 89. A chi-square test of independence
conrmed that the association between condition (MP/EP/FP) and response (yes/no)
is highly signicant and very strong:
𝜒2(2) = 25.97
,
𝑝 < .0001
; Cramér’s V = .50. We
followed up the chi-square test with a series of pairwise Fisher’s exact tests, which revealed
13
Using the BayesFactor package for R(Morey and Rouder 2015), we also conducted a Bayesian
proportion test, obtaining a Bayes factor of 3.18 against the null hypothesis. is indicates that the data are
3.18 times more likely assuming the hypothesis that the probability of a positive response is greater than .16
than assuming the null hypothesis. According to Jeffreys’ (1961: 432) classication scheme, this means that
our data substantially support the alternative hypothesis.
12
0
10
20
30
MP EP FP
Condition
Count
yes
no
Figure 1: Yes/no responses for the three vignettes used in Experiment 1.
a highly signicant difference between the MP and FP conditions (
𝑝 < .0001
) and a
marginally signicant difference between the MP and EP conditions (
𝑝 = .054
); the
difference between the EP and FP conditions also came out signicant (
𝑝 = .002
; all
reported 𝑝-values are Bonferroni–Holm adjusted).14
We obtained further information relevant to answering Q2 and Q3 from the responses
to the three Likert scale questions. An overview of these responses is given in Figure 2.
Visually, the trend in those responses is not too different from to the trend in the responses
to the yes/no questions, with participants being more favorably inclined toward the EP
and FP options than to the MP option, and also seemingly somewhat more favorably
inclined to the FP option than to the EP option. Here, this impression was conrmed by a
one-way ANOVA, with condition (MP/EP/FP) as independent variable and rating (from
very unfavorable to very favorable) as dependent variable:
𝐹(2,101) = 28.05
,
𝑝 < .0001
;
there was a very large effect of condition on ratings:
𝜂2
p= .36
. Post-hoc comparisons
using Tukey’s HSD indicated that both the difference between the mean rating for the
MP condition (
𝑀MP = 2.59
,
𝑆𝐷MP = 1.89
,
𝑁MP = 34
) and that for the EP condition
(
𝑀EP = 5.09
,
𝑆𝐷EP = 1.96
,
𝑁EP = 35
) and the difference between the former mean
rating and the mean rating for the FP condition (
𝑀EP = 5.71
,
𝑆𝐷EP = 1.62
,
𝑁EP = 35
)
14
Again, we obtained basically the same results conducting a Bayesian analysis using the BayesFactor
package, which offers an implementation of Gunel and Dickey’s (1974) contingency table Bayes factor test.
Specically, we found a Bayes factor of
67 × 103
in favor of an association between condition (MP/EP/FP)
and response (yes/no), and conducting separate contingency table tests for pairs of levels of the condition,
we found a Bayes factor of 2.34 in favor of an association between condition (MP/EP) and response (yes/no);
a Bayes factor of
231 × 103
in favor of an association between condition (MP/FP) and response (yes/no);
and a Bayes factor of 65.72 in favor of an association between condition (EP/FP) and response (yes/no).
13
2
4
6
MP EP FP
Condition
Rating
Figure 2: Likert scale responses for the three vignettes used in Experiment 1. (Individual
responses are displayed with jitter to enhance their visibility.)
were signicant: both
𝑝s < .0001
; there was no signicant difference between the mean
ratings for the EP and FP conditions: 𝑝 = .326.15
Discussion
We presented participants with Nozick’s experience machine scenario in a bare form. is
made close to 30 percent of the participants prefer the option, which is signicantly more
than the percentages found in some previous research using more detail-rich versions of
the experience machine scenario, but it is a minority nonetheless. So, the answer to Q1 is
essentially positive: while we did not strictly obtain a replication of Weijers’ result for
our version of the experience machine scenario, our results conrm his general nding
that a majority of people appear to oppose being connected to the machine.
Furthermore, our data—both the yes/no responses and the Likert scale responses—
warrant an unequivocally positive answer to Q2. Participants appeared signicantly
more inclined to accept the offer to take a pill, whether one guaranteeing pleasurable
experiences or one guaranteeing improved functioning, than to accept the offer to be
hooked up to the experience machine. e answer to Q3 is positive as well, at least as
far as the yes/no responses go: signicantly more participants were willing to take a pill
guaranteeing improved functioning than one guaranteeing pleasurable experiences. As
15
A series of Bayesian ANOVAs, conducted using the BayesFactor package again, and all of which had
rating as response variable, yielded very similar results: We obtained a Bayes factor of
45 × 106
favoring a
model with condition (MP/EP/FP) as predictor over the intercept only model; a Bayes factor of
13 × 102
favoring a model with condition (MP/EP) as predictor over the intercept only model; a Bayes factor of
24 × 106
favoring a model with condition (MP/FP) as predictor over the intercept only model; and a Bayes
factor of 1.64 favoring the intercept only model over a model with condition (EP/FP) as predictor.
14
for the Likert scale responses, the mean response for taking the functioning pill was
higher than that for taking the experience pill, but the difference was not signicant. In
fact, both means were rather close to the maximal rating.
As stated earlier, someone taking a pill guaranteeing pleasurable experiences could
be said to be more in contact with reality than someone who is connected to a machine
guaranteeing pleasurable experiences, while someone taking a pill that improves func-
tioning could be said to be in still closer contact with reality. us, we predicted positive
answers to Q2 and Q3, based on what we called “the authenticity intuition.” Accordingly,
that we found a positive answer to Q2 and also a mostly positive answer to Q3 supports
that intuition.
5 Experiment 2
e second experiment was designed to address Q4. It was basically a rerun of Experi-
ment 1, except that now all vignettes had opposite valence in that the options offered in
them prevented negative consequences, rather than generated positive consequences.
Participants
In this experiment, there were 256 participants. Recruitment, testing, and nancial
compensation of participants was as in Experiment 1. We also used the same selection
criteria as in that experiment. is le us with 231 participants for the nal analysis
(𝑀age = 40,𝑆𝐷age = 13; 134 females).
Materials and procedure
e materials and procedure were as in Experiment 1, the only difference being that
now in the machine and experience pill vignettes the participants were told that the
machine/pill would prevent them from a pending future in which they would experience
almost exclusively unpleasant experiences (instead of guaranteeing almost exclusively
pleasurable experiences) while in the functioning pill vignette participants were told that
the pill would prevent them from a pending future in which the way they function in a
wide range of respects would substantially deteriorate (instead of guaranteeing improved
functioning). Because of their negative valence, we refer to these variant vignettes as MN,
EN, and FN, respectively. (See the Supplementary Information for the full vignettes.)
Results
Figure 3 gives an overview of the responses from the three groups that were asked yes/no
questions. When this gure is compared with Figure 1, only one difference stands out: in
the FN condition, acceptance of the offer to take the pill is, while still the majority response,
not at the same high level as in the FP condition from Experiment 1. Specically, in the
MN condition, only 26 percent gave a positive answer (29 percent in the MP condition
15
0
10
20
30
MN EN FN
Condition
Count
yes
no
Figure 3: Yes/no responses for the three vignettes used in Experiment 2.
from Experiment 1) while in the EN and FN conditions the percentages were 55 and 64,
respectively, the corresponding percentages from Experiment 1 being 53 and 89.
A series of chi-square tests showed that, while there is no signicant difference
between the results from the MP and MN conditions, nor between the results from the
EP and EN conditions, the difference between the results from the FP and FN conditions
is signicant:
𝜒2(1) = 4.66
,
𝑝 = .031
, Cramér’s V = .29. A chi-square test for the
results from Experiment 2 showed that there was again a signicant association between
condition (MN/EN/FN) and response (yes/no):
𝜒2(2) = 11.38
,
𝑝 = .003
, Cramér’s V =
.32. Following up this nding by pairwise Fisher’s exact tests revealed that the proportion
of yes/no responses in the MN condition differed signicantly from both the proportion
of yes/no responses in the EN condition and the proportion of yes/no responses in the FN
condition (
𝑝 = .025
and
𝑝 = .006
, respectively, these being Bonferroni–Holm adjusted
𝑝
-values); there was, as expected, no signicant difference between the proportions of
yes/no responses in the EN and FN conditions.16
e results from the Likert scale questions are represented in Figure 4. As a compari-
son of this gure with Figure 2 already suggests, the differences between the conditions
here were smaller than the differences between the corresponding conditions in Experi-
ment 1 (
𝑀MN = 2.54
,
𝑆𝐷MN = 1.83
,
𝑁MN = 37
;
𝑀EN = 4.58
,
𝑆𝐷EN = 2.10
,
𝑁EN = 38
;
𝑀FN = 4.70
,
𝑆𝐷FN = 1.91
,
𝑁FN = 33
). Nevertheless, a one-way ANOVA with condi-
tion (MN/EN/FN) as independent variable and rating (from very unfavorable to very
16
We obtained similar results from a series of contingency table Bayes factor tests, like the ones reported
in note 14: we found a Bayes factor of 21.42 in favor of an association between condition (MN/EN/FN)
and response (yes/no); a Bayes factor of 7.23 in favor of an association between condition (MN/EN) and
response (yes/no); a Bayes factor of 52.50 in favor of an association between condition (MN/FN) and
response (yes/no); and a Bayes factor of 2.73 in favor of the null hypothesis of no association between
condition (EN/FN) and response (yes/no).
16
2
4
6
MN EN FN
Condition
Rating
Figure 4: Likert scale responses for the three vignettes used in Experiment 2.
favorable) as dependent variable turned out signicant:
𝐹(2,105) = 13.97
,
𝑝 < .0001
,
𝜂2
p= .210.17
Most relevantly to Q4, we pooled the Likert scale data from Experiments 1 and 2 and
conducted a
3×2
ANOVA with condition (M/E/F) and valence (P/N) as between-subjects
factors and response (yes/no) as dependent variable.
18
is revealed a main effect of
condition:
𝐹(2,206) = 40.28
,
𝑝 < .0001
,
𝜂2
p= .279
. However, that was not surprising,
given the outcomes of the separate one-way ANOVAs for Experiments 1 and 2, and it
also has no direct bearing on Q4. Relevant to that question is the fact that there was
also a main effect of valence,
𝐹(1,206) = 3.94
,
𝑝 = .049
, although the effect was small:
𝜂2
p= .019
. ere was no signicant interaction between the factors:
𝐹(2,206) = 1.14
,
𝑝 = .323,𝜂2
p= .011.19
17
Here, the Bayesian analysis concurred as well: a Bayesian ANOVA with condition (MN/EN/FN) as
independent variable and rating as dependent variable yielded a Bayes factor of 4400.08 favoring the model
with the condition as independent variable over the intercept only model. Furthermore, we found a Bayes
factor of 684.13 favoring a model with condition (MN/EN) as predictor over the intercept only model;
a Bayes factor of 1940.77 favoring a model with condition (MN/FN) as predictor over the intercept only
model; and a Bayes factor of 3.98 favoring the intercept only model over a model with condition (EN/FN) as
predictor.
18
M is a label for the union of the responses in the MP and MN conditions in Experiments 1 and 2,
respectively, and similarly for E and F. P is a label for the responses from Experiment 1 (in which the
vignettes had positive valence) and N is a label for the responses from Experiment 2 (in which the vignettes
had negative valence).
19
A Bayesian ANOVA yielded a Bayes factor of
2.5 × 1012
favoring the model with condition as indepen-
dent variable over the intercept only model, but a Bayes factor of 1.35 favoring the intercept only model
over the model with valence as independent variable, a Bayes factor of
2.4 × 1012
favoring the intercept
only model over the model with both condition and valence as independent variables, and a Bayes factor of
5 × 1011
favoring the model with condition, valence, as well as their interaction as independent variables
17
Discussion
Based on previous literature, there was reason to reckon with the possibility that (what
we called) the valence of the vignettes might make a difference to how participants judge
them. Most relevantly, in Section 1.1 we mentioned De Brigard’s scenario that offers
participants, as an alternative to remaining connected to Nozick’s experience machine,
living a very unhappy life as a prisoner in a maximum security prison, an alternative
that was rejected by 87 percent of his participants (they chose to remain connected to
the machine). is can be explained in terms of loss aversion, and it would lead one
to expect a signicant difference between the P- and the N-scenarios. However, this
expectation was only weakly conrmed, in that we found no more than a small effect of
valence, which moreover was due to only one scenario (the functioning pill scenario).
While the main focus of Experiment 2 was Q4, it is to be noted that the results from
this experiment are very similar to those from Experiment 1. In particular, we found
again that a majority of participants in the experience machine condition rejected the offer
to be connected to the machine while both in the experience pill and in the functioning
pill condition, the majority of participants accepted the offer to take the pill. Also, the
trend was again that the percentage of positive responses was highest for the functioning
pill vignette, lower for the experience pill vignette, and still lower for the experience
machine vignette. ereby, the current results shed further light on questions Q1–Q3 as
well, and more broadly offer further support for the authenticity intuition.
6 Conclusion
ere has been much discussion in the philosophical literature about Nozick’s experience
machine thought experiment. We took that thought experiment, together with some
of Nozick’s commentary on it, as a starting point for investigating what we called “the
authenticity intuition”: the idea that people value living their lives in contact with reality,
and also that they care about who they are and what they do. We investigated this intuition
by means of two experiments. e design of these experiments was primarily motivated
by the thought that authenticity comes in degrees: we can be more or less in contact with
reality, stay more or less close to who we actually are, and feel to a greater or lesser degree
that our actions are genuinely our own. Our materials consisted of vignettes that featured
different degrees of detachment from reality and self, and were meant to established
whether those differences were reected in our participants’ acceptance rates, or their
attitudes toward the situations described in the vignettes.
ere were two secondary motivations. First, earlier work on the experience machine
had cast some doubt on the usefulness of this scenario, and had pointed to various
possible confounds, mostly in the form of cognitive biases that might skew people’s
reactions to the scenario. Our aim was to present a simple version of Nozick’s scenario
that would keep such biases at bay, as much as possible. Second, it has been recognized
in unrelated literature that valence—bringing about something good versus preventing
over the intercept only model.
18
something bad from happening—can inuence people’s value judgments in sometimes
unexpected ways. We wondered whether valence also mattered to how people valued
authenticity.
e results we obtained supported earlier ndings in that participants were in majority
disinclined to accept the offer of being connected to the experience machine or, in the
Likert scale conditions, held largely unfavorable attitudes toward that option. We also
found that participants were more likely to accept the offer of taking an experience pill and
still more likely to accept the offer of taking a pill that would improve their functioning
in important ways. (Although not all differences were signicant here, the trend was
clear and consistent across relevant conditions.) Because these scenarios were less (the
experience pill scenario) to much less (the functioning pill scenario) disruptive than
Nozick’s, in the sense that the result of the featured intervention would move people
less far away from normality than the result of connecting to the experience machine,
these ndings corroborated precisely what had been predicted on the basis of our main
hypothesis (the authenticity intuition).
It was further encouraging to nd that the answers to questions Q1–Q3, which con-
cerned the authenticity intuition most directly, were the same from both experiments,
meaning that the conclusions we reached about that intuition are robust and largely
independent of the valence of the interventions proposed in our vignettes. ere was
one small effect of valence, which was due to one specic scenario in one specic condi-
tion (the functioning pill scenario in the yes/no condition). Whether this nding is a
coincidence or whether it points to something deeper is an issue we leave as an avenue
for future research.20
References
Aust, F., Diedenhofen, B., Ullrich, S., and Musch, J. (2013)
“Seriousness Checks Are
Useful to Improve Data Validity in Online Research,” Behavior Research Methods
45: 527–535.
De Brigard, F. (2010)
“If You Like It, Does It Matter if It’s Real?” Philosophical Psychology
23: 43–57.
Griffin, J. (1986)
Well-being: Its Meaning, Measurement and Moral Importance, New
York: Oxford University Press.
Gunel, E. and Dickey, J. (1974)
“Bayes Factors for Independence in Contingency Ta-
bles,” Biometrika 61: 545–557.
Haybron, D. (2011)
“Happiness,” e Stanford Encyclopedia of Philosophy (Fall
2011 Edition), available at
http://plato.stanford.edu/archives/fall2011/
entries/happiness/.
Hindriks, F. (2014)
“Normativity in Action: How to Explain the Knobe Effect and its
Relatives,” Mind & Language 29: 51–72.
20
We are greatly indebted to three anonymous referees for valuable comments on previous versions of
this paper.
19
Jeffreys, H. (1961) eory of Probability (3rd ed.), Oxford: Oxford University Press.
Knobe, J. (2003)
“Intentional Action and Side Effects in Ordinary Language,” Analysis
63: 190–194.
Knobe, J. (2010)
“e Person as Moralist Account and Its Alternatives,” Behavioral and
Brain Sciences 33: 353–365.
Morey, R. D. and Rouder, J. N. (2015)
BayesFactor: Computation of Bayes Factors for
Common Designs,Rpackage version 0.9.11-1,
http://CRAN.R-project.org/
package=BayesFactor.
Nozick, R. (1974) Anarchy, State, and Utopia, New York: Basic Books.
Phillips, J., Misenheimer, L., and Knobe, J. (2011)
“e Ordinary Concept of Happiness
(and Others Like It),” Emotion Review 3: 320–322.
Sumner, L. W. (1992) “Welfare, Happiness, and Pleasure,” Utilitas 4: 199–223.
Weijers, D. (2013)
“Intuitive Biases in Judgments About ought Experiments: e
Experience Machine Revisited,” Philosophical Writings 41: 17–31.
Weijers, D. (2014)
“Nozick’s Experience Machine Is Dead, Long Live the Experience
Machine!” Philosophical Psychology 27: 513–535.
20
