2.1. Highlighting the structural isomorphism

Before I dive into details, I mark one important rationale for introducing and defending the proposed isomorphism. Exegetically, there seems to be an initial tendency to move from manipulation in ordinary cases to manipulation in simulation. In the original Frankfurt-style Case, the manipulator manipulates one particular action of one particular agent, and Frankfurt (Reference Frankfurt1969) asks: What our judgment about the agent’s moral responsibility (or the absence thereof) should be? After that, Mele and Robb (Reference Mele and Robb1998) further ask: Is there any change regarding our judgment about the agent’s moral responsibility (or the absence thereof) if the manipulator manipulates all actions of the particular agent? In a similar vein, I want to ask: Is there any change regarding our judgment about an agent’s moral responsibility (or the absence thereof) if the manipulator manipulates all actions of all agents? To this extent, § 2.1 answers to an unexplored tendency in the relevant literature from Frankfurt-style Cases via Global Frankfurt-style Cases to simulations.

To begin, I’d like to demarcate what I mean by “simulation” and “manipulation,” as there are various readings of either concept in the rich literature. The idea of simulation ultimately goes back to computer science, in which a simulation is usually defined as a program using step-by-step methods to explore the approximate behavior of a mathematical model that is run on a computer.Footnote ¹¹ On this conception, the following features are arguably essential to a simulation qua a computer program. First, a simulation is pre-programed in the sense that it starts with some initial parameters set by the simulator.Footnote ¹² Consider a (toy) simulation of a car accident for example. The simulator has to input some data like the vehicle’s initial velocity, the friction of the road surface, etc., before the simulation is run. Second, after the initial parameters are set, the following simulating process nevertheless proceeds in a seemingly autonomous manner.Footnote ¹³ Here is how. Again, consider the simulation of a car accident. Suppose the crush in the simulation is about to take place at time t. Intuitively, the occurrence of the crush at time t is the result of the system’s own evolvement given these initial parameters set prior to t. That said, provided these pre-programed parameters, the crush naturally follows without any need for the simulator’s further intervention at t. To this extent, a simulation seems to manifest a kind of autonomy.

Bearing these two features of simulation in mind, I continue demarcating the concept of manipulation. There are obviously abundant possible conceptions and distinctions made about manipulation in the literature. However, what is directly relevant to our discussion is just a tiny tip of the iceberg. Quite remote from manipulations common to ordinary life in the form of guilt trips, gaslighting, peer pressure, negging, or emotional blackmail, the kind of manipulation discussed by free will/moral responsibility specialists is instead a highly artificial kind that is “imagined as happening via decidedly extra-ordinary methods” (Noggle Reference Noggle and Zalta2022: § 1). My discussion in this paper only focuses on this artificial kind.Footnote ¹⁴ This kind of manipulation “typically includes a powerful agent [i.e., the manipulator] who is capable of altering her dupe’s [i.e., the manipulatee] external or internal conditions to ensure that the dupe does what she wants” (Deery and Nahmias Reference Deery and Nahmias2017: 1258). More particularly, two points about this kind of manipulation are worth marking. First, at least for hard-liners, the means of manipulation is arguably irrelevant to its manipulative nature. Manipulation could be done by directly interfering with the manipulatee’s mental states by sending signals to affect her brain (e.g., Pereboom’s Case 1), or via disposing the manipulatee to a designed deliberation outset such that she could only act in the way the manipulator wants (e.g., Pereboom’s Case 2), or via the manipulator’s pre-design in the (remote) past that causally necessitates the manipulatee’s choice that aligns with the manipulator’s purpose (e.g., Mele’s Zygote case). Second, in manipulation, the manipulatee still seems to act through her own Compatibilist Agential Structure, i.e., features of her own psychology that compatibilists typically judge as jointly (and minimally) sufficient for free will and moral responsibility.Footnote ¹⁵ In other words, even if the manipulatee’s decision inevitably aligns with the manipulator’s purpose, from the manipulatee’s own perspective, she still has control over her choice and action. For a clear presentation of this feature, recall Pereboom’s Case 2 and Mele’s Zygote case: In either case, via her pre-designing, the manipulator does not directly bypass any of the manipulatee’s relevant capacities at, or before, the time of action.Footnote ¹⁶

By now, it is not hard to see the similarity between simulation and manipulation: They both manifest the same structure (call it the key structure). Consider the following contrast for illustration:

• Simulation: [Input A as pre-programming] [Computing rules]→ [Output B as the result of the system’s own evolvement]

• Manipulation: [Interference C as pre-design] [Laws of causality]→ [Performance D as the result of the agent working through his Compatibilist Agential Structure]

• The key structure: [Prior state set by party E] [Independent rules/laws]→ [Post state realized by party F that seems to bestow autonomy]

That said, both simulations and manipulations manifest the same structure composed of five key elements, as illustrated in the chart:

Chart. The Structural Isomorphism Between Simulation and Manipulation.

In short, by manifesting the key structure, simulation is isomorphic with manipulation.Footnote ¹⁷ By virtue of such structural isomorphism, I argue, there is no relevant difference between simulation and manipulation regarding whether, in their presence, an agent should be held morally responsible for her actions.Footnote ¹⁸ That said, any reason to hold an agent for (or exempt her from) moral responsibilities in cases of manipulation also applies to cases of simulation, and vice versa.Footnote ¹⁹ I only mark two examples of such reasons here. First, a hard-liner compatibilist might hold that agents under manipulation should be deemed morally responsible because any action she performs is still the result of her working through her Compatibilist Agential Structure and no relevant capacities of her are ever bypassed in the process. Similarly, a parallel reason could say an agent in simulation is also morally responsible in the sense that any action performed by her also seems to be the result of her working through a structure of her own psychology which is similar to Compatibilist Agential Structure in cases of manipulation. Second, on the contrary, a hard-liner hard determinist may hold that agents under manipulation should be exempted from moral responsibility because, despite the dubious autonomy they manifest, their later action is causally necessitated by the manipulator’s pre-design, which arguably undermines moral responsibility. Similarly, a parallel reason could say an agent in simulation is also not morally responsible because her later action is necessitated by the simulator’s pre-programming, too. That said, if manipulation undermines moral responsibilities, then so should simulation; and if simulation is compatible with moral responsibilities, then so should manipulation.Footnote ²⁰

I deal with one objection at the end of § 2.1. Consider the following depiction of the dialectical landscape:

One may say: Suppose simulation is compatible with moral responsibility; for your master argument to work, the case of simulation should presumably be added to Place 1; however, since simulation also resembles (causal) determinism, we may put it in Place 2 or 3, too; and if so, then hard-liner hard determinists may still reasonably hold the intuition that “Manipulation undermines moral responsibility,” and the purpose of your paper is jeopardized.

My reply is this. I agree that simulation, manipulation, and determinism are similar in the sense that, in all of them, the prior state and the rules/laws taken together ensure the inevitable realization of the post-state.Footnote ²¹ However, one apparent difference is that the prior state for simulation and manipulation is set up by simulators or manipulators; while the prior state for determinism is probably some events in the natural history (e.g., the occurrence of the Big Bang), or at least determinism is silent about where the prior state comes from.Footnote ²² Thus, because of the presence of a second party (i.e., a simulator or a manipulator), simulation is more similar to manipulation than to determinism and it probably should not be put in Place 3.

As for whether simulation belongs to Place 1 or 2, let’s continue considering Pereboom’s transitional Cases 2 and 3. Arguably, Pereboom’s Cases 1, 2, and 3 are all manipulative; though the closer it gets to Case 4, the more disguised and ambiguous the means of manipulation becomes. The manipulation in Case 1 is done by directly controlling the manipulatee’s mind. The manipulation in Case 2 is done by preprograming the manipulatee’s deliberation outset. The manipulation in Case 3 is done by rigorous training from the manipulatee’s family and community. Given these different means of manipulation, some would say simulation should be put in Place 2 since it is more similar to Case 2 (as preprogramming is present in both cases) than to Case 1.

I think it might be true with respect to the indirectness of the influence a simulator has on the simulatee. However, I think there are other aspects in which simulation is more similar to Case 1 than to Case 2. One such aspect is the power imbalance. The power imbalance in Case 1 seems to be greater in the sense that the manipulator in Case 1 can directly invasively violate the integrity of the manipulatee’s body/brain while the manipulator in Case 2 cannot. And the power imbalance in simulation, I argue, is even greater than that in Case 1. The simulator can, by simply adding or revising several lines of code, in principle directly invasively violate the integrity of the simulatee’s body (though he may choose not to) and preprogram his deliberation outset (as suggested by the key structure) in the same time. That said, the manipulation techniques a simulator can employ are arguably the combination of techniques present in both Cases 1 and 2, which makes a simulator more powerful and God-like than a manipulator. In light of this, I argue it is not obviously true that simulation is more similar to Case 1 than to Case 2 and should be put in Place 2 rather than Place 1.

At the very least, even if simulation in fact resembles Case 2 more, it should not end up in Place 2 for the following reason. Regarding why simulation does not undermine moral responsibility (to be discussed in § 3), my resource is the discussion of virtual reality which, as far as I am concerned, no moral responsibility specialist has particularly appealed to. That is, my later argument will invite hard-liner hard determinists to a new dialectical environment. If they are on board with my judgment about moral responsibility, if any, in virtual reality, and provided the isomorphism between simulation and manipulation (as I defended earlier in this section), hard-liner hard determinists will be pushed to introspect the intuition they appeal to. To this extent, the introduction of simulation should initiate a new thread, rather than merely adding a new dot to the old thread of the four-case transition.

2.2. Dismissing some irrelevant differences

The precedent section argues: There is no MR-relevant difference between simulation and manipulation. I now anticipate several objections in the subsequent section.

One objection says: Though simulators and manipulators affect one’s action in similar ways, simulators are outside the simulatee’s world while manipulators are inside the manipulatee’s world. And this in-/outside distinction might be MR-relevant. But I argue this alleged difference, though seems plausible, is MR-irrelevant. To see how, consider any video game as a toy simulation. The game developer can, from time to time, enter the game world by logging in to her avatar. That said, the boundary of a simulation is not in principle impenetrable, and simulators are not necessarily outside. Meanwhile, manipulators are not necessarily inside. To make an intuitive case, consider the comic world of John Constantine.Footnote ²³ In that world, the God lives in Heaven, the Devil lives in Hell, and both agree to not enter the Earth and interfere with humans directly. However, the Devil may still manipulatively seduce humans by making a sequence of events that leads a person to, say, commit suicide.

Another objection says: There might be an MR-relevant difference between simulation and manipulation regarding whether a particular manipulative intention is present.Footnote ²⁴ For example, in Pereboom’s Case 1, the manipulation is directed by the particular intention of having Plum kill White. Simulation, by contrast, is usually directed by the intention of making the state of the world evolve in a particular pattern (e.g., a world in which Nazis won WWII). Such an intention does not primarily care about particular agents like Plum and particular actions like killing white. That said, even if simulation is a kind of manipulation that does not undermine moral responsibilities, other kind(s) of manipulation may undermine moral responsibilities as well. I think this line of reasoning might be true, but it is arguably soft-liner in spirit. Recall Mele’s Zygote case or Pereboom’s Case 3. In these two cases, the manipulator, if at all, does not manipulate the manipulatee by implanting particular intentions like killing White to him, either. However, at least for hard-liners, there is no relevant difference between the Zygote case, Case 3, and Case 1 regarding whether the manipulatee, if at all, should be held morally responsible for his killing White. That said, at least for hard-liners, the presence (or absence) of particular manipulative intention is probably moral responsibility-irrelevant.

A third objection says: Manipulation requires bad intentions while simulation doesn’t; therefore, a simulator is not necessarily a manipulator. My reply is this. Sure, the simulator’s intention needs not to be bad, but it is arguably irrelevant to whether manipulation in fact takes place. For example, consider the debate over God, free will, and manipulation in scholastic philosophy. A Catholic hard determinist who defends the simulation hypothesis (or its ancient relative, say, external world skepticism) can consistently believe the following: The God doesn’t intend to play us around as his puppets (because of his love for us); he has the ultimate control of everything happening in the world; and the world he creates is a simulation (or a dream). In this case, it seems plausible for the Catholic hard determinist to say we are manipulated by God, even though there is no bad intention involved. Besides, recall Pereboom’s Case 1. In that case, it is true that these scientists manipulate Plum, but they may do so for good, paternalist reasons (like to avoid Plum’s suffering), though their “good” intention arguably does not change the fact that Plum is manipulated. To sum up, the manipulator’s intention, no matter good or bad, is moral responsibility-irrelevant.Footnote ²⁵

3.1. The simulator-centric reason

The simulator-centric reason says: A good simulation, from the simulators’ perspective, should be designed to allow its inhabitants to be morally responsible because moral responsibility is usually connected to the meaning in life and the meaning in life motivates simulatees to help realize the simulator’s goal. Expectedly, this reason is committed to a moral responsibility-meaning in life connection, which holds that moral responsibility matters deeply to “our conception as deliberative agents,” to whom the absence of healthy “reactive attitudes essential to good human interpersonal relationships and meaning in life” would be disastrous deprivation of life hopes (i.e., aspirations for achievement) (McKenna and Pereboom Reference McKenna2014: 262–79). The moral responsibility-meaning in life connection is canonical in the relevant debate. However, it has recently been challenged by, most notably, some hard determinists.Footnote ²⁷ They object: There might be a sense in which an agent can satisfy her life hopes or achievements without the presence of moral responsibility, though this sense is probably not as robust as we might naturally suppose. The picture of a meaningful life without moral responsibility can be drawn in three ways.Footnote ²⁸ My goal here, then, is to show how these alternatives are, though arguably making sense, probably not the best options for simulators.

One way of accommodating the meaning in life in a world without moral responsibility is to connect the meaning in life to purely consequentialist considerations.Footnote ²⁹ That said, one can still live a meaningful life without moral responsibility by committing herself to some purely consequentialist doctrines, e.g., to always maximize her welfare. I understand this idea may seem attractive, especially for those who have the platitude that sims are just robot-like followers of codes. However, I mark two concerns about the possible insufficiency of a purely consequentialist base for meaning in life from a simulator’s perspective. Firstly, merely consequentialist doctrines may be incompatible with some simulations. Consider, for example, a trivial case – the simulation of a world in which all agents are virtuous moral saints. For this simulation, arguably, some virtue ethicist principles are also needed for its simulatees to live a meaningful life. More realistically, consider a simulation of the actual world at present: Since not everybody is (and in fact many of us are not) stubborn consequentialist, to simulate our moral deliberation and action, something over and above purely consequentialist considerations seems to be required. Secondly and more crucially, merely consequentialist doctrines may fail to approximate the complexity that some simulation requires. Consider, for example, a simulation of how a certain policy affects the global society. If everybody acts solely in accordance with some purely consequentialist doctrines, then, seemingly, their actions are highly predictable (i.e., they always act to maximize welfare). However, one important goal of the simulator running this simulation is to properly approximate the complexity and unpredictability of people under the given circumstance. That said, at least for this simulation and the like, guiding sims with purely consequentialist doctrines undermines the simulator’s goal.

Another way of accommodating the meaning in life in a world without moral responsibility is to connect the meaning in life to some objective attitudes.Footnote ³⁰ Some would say that apart from reactive attitudes like resentment or appreciation we have toward our fellow agents that are closely connected to moral responsibility, some objective attitudes we have toward non-agential entities like the appreciation of nature or art may also constitute the meaning in life. I agree with this point. However, I think this point only applies to a small number of simulations (e.g., a simulation of a world governed by an artist king). But I doubt if objective attitudes like the appreciation of nature or art alone are enough, for example, to motivate sims in a simulation of WWII. It seems hard for me to imagine a sim Jewish French patriot devotes herself to the resistance movement merely because of some objective attitudes like “(sim) Hitler launched a war and genocide, though he is not responsible for it in the genuine sense and I do not resent him.”

A third way of accommodating the meaning in life in a world without moral responsibility is to connect the meaning in life to some religious commitments.Footnote ³¹ That said, by detaching oneself from taking credit for her deeds, one’s life could still be deemed meaningful in the Stoic sense (for her peace of mind), the Christian sense (for her humility), or the like. I think this point might be true. But my concern, for starters, is that religious commitments might be incompatible with some simulations, consider, for example, a simulation of a world in which everybody is an atheist philosopher. Besides, a more serious problem is this: Arguably, a simulator deliberately converting a population of sims to a certain religious commitment is, by analogy to a cult’s brainwashing, morally questionable itself, even if not straightforwardly wrong.

Summing up my point, I conclude: At any event, moral responsibility still seems to be the most unproblematic base for a simulator to ground the meaning in life for her simulatees, as most people conventionally assume.

Finally, one objection to the simulator-centric reason may run like this: It might stand for benevolent simulators; however, what if the simulator is malevolent? Then why would he even want to create a good simulation that allows its inhabitants to live meaningful lives? I don’t have a strong position against this objection. But I think it is practically unwise, given the cost-benefit calculus, to create simulations that forbid the meaning in life, even for malevolent simulators (like an evil Demon). In practice, simulators use energies and resources (i.e., the cost) to create simulations that serve their purpose (i.e., the benefit). Creating simulations will use a lot of energy and resources, that said, the cost is not cheap (Bostrom Reference Bostrom2003: 245). Given such a cost, the simulator will presumably expect reasonable benefits. That said, simulators would naturally prefer their purposes – no matter what, even if it is something narcissistic or perverse like to be worshiped or to play God – to be achieved, which requires some actions to be done by simulatees (even if it is something like praying).Footnote ³² Having the meaning in life by being morally responsible – that is, taking credit for one’s achievements in the genuine sense – can probably better motivate actions (including but not limited to simulatee’s). Consider, for illustration, a person finds the cure for cancer. What she has done, in Scenario one, may be deemed her accomplishment (in a genuine sense) and people praise and congratulate her for that, or it may not be deemed her accomplishment (in a genuine sense) in Scenario two and people’s praise and congratulation to “her,” if any, feels exactly the same with their praise and congratulation to a random stranger. Intuitively, the person will have a stronger motivation to work hard in Scenario one, and this intuition in principle generalizes.Footnote ³³ In light of this, from a practical perspective, in most cases, it seems preferable to create simulations that allow the inhabitants to live meaningful lives by being morally responsible.

3.2. The simulatee-centric reason

There are two classes of standards, as I see it, to judge whether there are moral responsibilities in simulations. One class of standards is internal, which focuses on some internal qualities of (possible) moral responsibility-holders. However, appealing to these internal conditions is probably bootstrapping because, given manipulation arguments, there are concerns about whether usual (compatibilist) internal conditions for MR-susceptibility – e.g., the ability to identify and reflect on desires, the ability to respond to (both moral and non-moral) reasons, and the ability to not act on compulsive or compelled desires, among others – are enough.Footnote ³⁴ That said, by looking into manipulation arguments, compatibilists (especially soft-liner compatibilists) intend to figure out whether MR-susceptibility further requires a certain “no manipulation” condition: The efficiency of available internal conditions itself is being questioned.

A better option, then, seems to be turning to the other class of standards, i.e., some key external conditions of the surrounding environment in which agents are held morally responsible for their actions. In § 3.2, I will examine three such external conditions. They are: 1) the presence of persons/agents, 2) the presence of events/actions, and 3) the presence of a certain inter-personal, social structure. These key external conditions seem natural. Just consider our standard practice of holding others morally responsible: We usually hold an agent (clause 1) of some moral community (clause 3) morally responsible for an action (clause 2). The question, then, is: Can a simulation satisfy these conditions? My answer is yes. A simulation can satisfy the first two conditions. I report that there seem to be good reasons to believe 1) sims/digital lives deserve moral status and may further bestow autonomy or agency, and 2) virtual events really take place, though I cannot provide a full-scale defense.Footnote ³⁵ The third key external condition, that moral responsibilities require a certain inter-personal, social structure, probably needs more demonstration. Inspired by Peter Strawson’s famous insight (Reference Strawson1962) that moral responsibilities are closely tied to our reactive attitudes, the third condition captures something deep in our conception of moral responsibility. That is: When one person is held morally responsible, other members of her society are also put in the position to resent or appreciate. I argue that there is indeed a certain inter-personal, social structure in a simulation such that a sim can have reactive attitudes toward other sims. Imagine a simulation of human history from 1900 to 2020. In that simulation, a sim mother Terressa can – at least we can interpret her actions as – resent sim Hitler for launching sim genocide. Similarly, a sim member of sim Nobel Prize Committee can also – at least we can interpret their actions as – appreciate sim mother Terressa for her philanthropy.Footnote ³⁶

I now propose this argument: Both virtual reality and physical reality satisfy these key external conditions of MR-susceptibility; there are moral responsibilities in physical reality (at least hypothetically); therefore, if there are moral responsibilities in physical reality, it is highly likely that there are moral responsibilities in virtual reality, too.

Three remarks are in order. First, the second premise is a relatively weak claim. For realists about moral responsibilities, it can be strengthened to be “There are in fact moral responsibilities in physical reality.” Since I only intend to argue for a conditional conclusion here, the weak version suffices.

Second, there might be a problem if I argue for a stronger conclusion that “If there are moral responsibilities in physical reality, there are moral responsibilities in virtual reality, too.” Since these three key external conditions are best understood as (strong) indicators, rather than as sufficient conditions, of the existence of moral responsibility, the presence of these external conditions does not guarantee the presence of moral responsibility. Hence, the strong conclusion does not follow from my premises, though the weaker conclusion I proposed shall not face this problem.

Third, I clarify “physical reality” in this argument does not necessarily mean the actual world. One potential objection says: Were the “physical reality” here to mean the actual world, this argument dubiously begs the questions against hard-liner hard determinists (take Pereboom for example).Footnote ³⁷ Pereboom’s position, charitably put, should be something like “Determinism is true for the actual world in which there is no free will (and moral responsibility), regardless of the world being physical or virtual.” In this sense, Pereboom would probably reject this premise in the first place.

I agree it is a legitimate concern. My reply is this. Imagine three possible worlds: Wa, Wv, and Wp. Wa is the actual world. And provided the relevant disagreement among philosophers, let us suspend judgment on whether there is free will or moral responsibility in Wa for now. Wv is a virtual possible world, which is very similar to Wa. That said, the only difference between Wa and Wv is that, from a God perspective, Wa is essentially a host of particles while Wv is essentially a sequence of digits. But for any person in Wa and her counterpart in Wv, their first-person experiences are indiscernible. Wp is a physical possible world that is similar to Wa in all aspects except that there is free will (and moral responsibility) in Wp.Footnote ³⁸ Suppose Wv satisfies these key external conditions of MR-susceptibility. Consider, then, the following adaption of my argument: (P1) Wa, Wv, and Wp all satisfy these key external conditions of MR-susceptibility; (P2) there are moral responsibilities in Wp; therefore, if there are moral responsibilities in Wp, it is highly likely that there are moral responsibilities in Wa and Wv, too.

This adaptation, as far as I can tell, does not beg the question against Pereboom. Instead, it may serve as an argument showing why Pereboom is probably wrong about Wa, i.e., the actual world. At any rate, I think Pereboom is disposed to some further issues, even if he rejects the conclusion in the end. Firstly, to reject P1, Pereboom may have to tease out the relevant difference among Wa, Wv, and Wp. But I doubt if it is a viable option since the way I introduce these three worlds leaves little space for further informative, non-ad hoc ramifications. Secondly, to reject P2 by questioning the intelligibility of Wp, Pereboom may have to strengthen his position by arguing “If determinism is true, it is necessarily true” and Wp is unintelligible in the first place. But I am not sure whether Pereboom will be on board with this idea (cf. fn. 38). Thirdly, to dismiss the strength of the conclusion, Pereboom may have to argue “very likely” does not guarantee truth, and Wa, Wv, and Wp may still be different in other MR-relevant ways despite their mutual satisfaction of my three key external conditions. I think it is Pereboom’s most promising way-out. However, I doubt how, at the end of day, any such alleged “other MR-relevant way” would not turn out to be a “no manipulation” condition that begs the question against me in the first place.