Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

In this post, I will defend the first premise of the Value Change Problem: that realistic models of human values take them to be malleable, rather than fixed.  I call this the ’value malleability’ claim (VM). 

I start by clarifying what I mean by value malleability before making a case that human values are in fact malleable. My general approach is to consider empirical observations about the practical reasoning of real-world agents (humans) and consider to what extent these are best explained by an account that posits fixed values vs one that posits value malleability. Concretely, I look at changes over the course of the human developmental lifespan; cases of value uncertainty, underspecification and open-endedness; and finally two existing accounts in the philosophical literature - transformative experiences and aspiration - and consider whether they represent genuine cases of value change. Finally, I consider some further objections. 

This is the longest of the three premises of the Value Change Problem, and the discussion is relatively long and philosophical. If you are excited about diving into some of the details, great - read ahead! If instead you’re looking for a minimal summary, the following is what I would say if I could only give one argument in favour of the value malleability claim: 

Argument from open-endedness:

Relative to (cognitively) bounded agents (such as humans), the world is open-ended. Humans do not come into this world with a complete representation of the world (e.g., a full hypothesis space). As a result, people occasionally get exposed to situations which they have not considered before and with respect to which they have not yet formed a meaningful evaluative stance. Thus, for a bounded agent to be able to function in such an open-ended world, it must be the case that they sometimes undergo value changes. In other words, the only way an agent could come to adopt an evaluative stance with respect to a genuinely novel situation is one that involves value change.

Clarifying what mean by value change

By claiming that values are malleable, I claim that they can (and typically do) undergo some form of change over, e.g., the course of a person’s life. 

But what do I mean by change? To clarify this question, let’s disambiguate three different notions of value change: value replacement, value development and value refinement. 

At a first glance, let us say that…

  1. Value replacement occurs when we adopt a novel value relative to an issue such that the new value replaces a value that was held earlier with respect to the same issue. 
    1. Example: Anna used to eat meat, and later, as a result of changing her values with respect to the moral patienthood of animals, she comes to adopt a vegetarian diet based on her new moral beliefs. 
  2. Value development occurs when we adopt a novel value such that its adoption does not cause the replacement of an existing value. 
    1. Example: Bob, on top of and while continuing to courage, comes to additionally adopt a value compassion.
  3. Value refinement occurs when we adopt a more nuanced, elaborate or contextualised version of an existing value. 
    1. Example: Clara used to understand her value for honesty as referring to ‘not knowingly saying false things’ and later expanded her conception of honesty to also concern cases of deceptive behaviour that does not involve saying literally false things.

For the present purposes, I am less interested in discussing whether or if so how these three notions represent mutually exclusive and collectively comprehensive definitions of value change. Instead, I think it is a useful exercise to start familiarising ourselves with the notion of value change, and building some vocabulary that might come in useful later. Furthermore, thinking about the above example allows us to ask, for each of these cases, whether these represent genuine cases of value change (rather than, say, being better explained by a story that does not involve a change of values). In doing so, we can notice two things - let’s think of them as markers of genuine value change - worth emphasising. 

First, what makes each of these an instance of value change is that, with respect to the total set of values that describes a person and their behaviour, the person has undergone a substantive, non-trivial change which manifests in their practical reasoning and behaviour. 

Second, for something to be a genuine case of value change, it must be the case that the person undergoing the change could have, in the relevant sense,[1] taken a different path, e.g., they could have adopted a different value, or refined their existing value in an alternative direction. Fore example, rather than refining her value of honesty in the way she did, Clara could have (in the relevant sense) formed values over, for example, whether someone acts honestly either from virtue or from duty (i.e., being intrinsically motivated to act honestly vs. acting honestly even if reluctant to), over the (in)acceptability of certain types of dishonesty (e.g., white lies), over different standards for how much someone is morally required to ‘go out of their way’ to avoid (intentionally or unintentionally) misleading others, etc. All of these are a priori reasonable ways to refine a value of ‘honesty’ which Clara could have (in a relevant sense) adopted, which I argue makes it such that we can consider this a case of genuine value change in the sense relevant to us. If that was not the case—if the value change could not have, in the relevant sense, taken a different path; in other words, if the nature of the change was fully inevitable—one may object that said ‘change’ should instead be better understood as the mere ‘unfolding’ of a preexisting, fixed, yet latent value. This would not count as an example of value malleability. 

Lastly, in asking what we mean  by value change, we could also ask what we do not want to count as genuine value change. There is a bunch more that could be said there, but the following seems worth mentioning in passing. There are cases (let's call them cases of ‘pretend’ change) in which someone is forced or finds themselves otherwise compelled to adopt certain behaviours which could be interpreted as the result of a change in values, but which is more appropriately understood as forced behaviour adopted against someone’s genuine values. For example, I may do things while, say, held at a gunpoint, which conflict with my (earlier) values, which nevertheless, do not result from an underlying value change.

To summarise, all three types of change described above—replacement, development and refinement—are all cases of genuine value change and thus ways in which the malleability of human values may manifest.

The case for value malleability

Do I have the same values as my younger self?

Let us assume, for a moment, that values are fixed (i.e., non-malleable). What would this imply? 

For one, it would imply children already possess the fully formed and complete sets of values that they will also have as adults.  If you’re like me, this will sound like a wild thing to believe. In other words, I don’t think this belief lines up well with what we observe. 

First, children certainly do not seem to have the same values as their adult selves in any trivial or straightforward sense. For example, a child that cared about animals and wanted to become a veterinarian will sometimes change their mind, start to care about, say, the arts and want to become a violinist instead. Second, children do not typically have formed values over a number of issues their later selves will have come to form values about, such as various political or philosophical stances (e.g., whether they are a pacifist, anarchist or capitalist), lifestyle choices (e.g., whether they will want to get married, have children, etc.) or aesthetic preferences.

You might object, however, that these aren’t genuine cases of value change. Instead, we might be able to make sense of these cases by proposing that children already hold their future values in a latent state, even if they may not be able to consciously access them. Those latent values might not yet show fully, but only take the form of some sort of ‘proto’ version of the child’s later, fully fledged values. Under this interpretation, rather than representing cases of genuine value change, what we are observing is instead the realisation, over the course of the child’s growing up, of those already existing, latent values. 

Recall that, in order to reject the VM claim, the latent values would have to be such that the child will come to develop their full versions invariably, i.e., that the development could not have, in the relevant sense, taken a different path. Inversely, if there are different ways in which a value could have developed, we treat this, according to our definition above, as a case of value change. However, as I will try to show, the invariability condition is not met in the case of human development, and thus that the explanation via latent values remains unsatisfactory.

To make this point, take, for example, a child who shows an inclination (i.e., ‘proto’ values) to judge any form of violence as atrocious. It seems like said child’s early tendencies could just about as well be interpreted as a ‘proto-pacifist position’ or as a ’proto-real political’ one. In the former case, the dispreference of violence could lead to the belief that any forms of violence must be resisted, while in the latter case, the same early dispreference could lead to the belief that the state’s monopoly of power is critical to maintaining peace and order  and thereby reduce the overall expected violence and harm. In other words, there is uncertainty about what values the child will come to adopt. What does this imply for the theory of latent values?

First, we must distinguish between two places from which the uncertainty could originate. In the first case, the uncertainty about the child’s future values resides in the subjective beliefs of the observer. In other words, while there is a true answer to what the child’s latent values are that it will come to hold, the observer does not have direct and full access to what those values are, and as a result, (the observer’s subjective beliefs about) the values are uncertain.

This type of uncertainty is compatible with claiming that the child has latent values it will inevitably come to adopt. However, this theory does not seem to afford the theoriser any epistemic purchase (e.g., improved predictive power). As we have seen in the example above, simply posing the existence of latent value doesn’t on its own tell us anything about what those latent values are. Furthermore, it is hard to imagine what epistemic position one could realistically come to inhabit that would make it justified (as well as pedagogically unproblematic) to act according to the belief that a child’s future values are already fully pre-determined. As such, one might critique that, while a theory of latent values requires us to posit an unobservable but purportedly real ontological entity (namely, latent values), this cost in parsimony is not met with equal or greater epistemic gains.

 In the second case, the uncertainty about the child’s future values resides not in the observer’s subjective beliefs but ‘in the territory’. In other words, while the possessions of a proto value biases how likely different value trajectories are, the child’s values could ultimately come to develop in different ways, depending on, say, what the child comes to experience over the course of its life. This picture, however, does not conflict with VM, and thus does not serve as a basis for rejecting it. Under this interpretation, the child’s inclinations (i.e., their ‘proto’ values) are (on their own) an underspecification of what future values the child will come to adopt. In other words, the child’s values could develop in different directions, depending on what experiences it will have, what ideas it will be exposed to, etc. If, however, the child, given its current inclinations and proto values, could come to adopt different sets of values in the future,  this development must involve some form of value change (in the sense defined above). As such, the basic premise of this section holds that the observation that children, over the course of their developmental process, come to hold different values provides evidence in favour of VM. This also shows how VM does not mean to exclude the possibility that certain (inherent, genetic, etc.) tendencies influence (but do not fully determine) a person’s future values

Dealing with value uncertainty, underspecification and open-endedness

Let’s consider another set of empirical observations about human practical reason, and check whether or not value malleability is a compelling way to make sense of these cases. Our premise is that humans often find themselves in one of the following positions: they are uncertain about their values; they hold values that are underspecified; or they may face novel situations over which we have not yet formed coherent values. I will argue that handling any of these situations will sometimes involve value change. 

Value uncertainty and underspecification

First, let’s consider the case of value uncertainty—the idea that we are often uncertain about what we value. 

Consider Fin, who is uncertain about what his endorsed stance is towards such issues like the moral status of non-human animals, the moral value of people who don’t exist yet, abortion, etc. I argue that in some (not necessarily all) cases where value uncertainty gets (partially) resolved, this occurs in a way that involves value change. (Of course, the inverse process is possible to—where one becomes more uncertain about a given value.)
For example, when dealing with the fact that he’s uncertain about the moral status of non-human animals, Fin might e.g. refine an existing value, or extend it to a novel issue. I have argued before that value reinforcement and development (on top of valve replacement) are genuine cases of value change. 

A similar argument applies to the case of value underspecification—the idea that we often hold values that turn out to be insufficiently specified. We have already seen an example of this earlier, when Clara’s initial underspecified notion of honesty got refined to include both the idea of ‘not saying false things’ as well as ‘not knowingly acting to deceive’. Underspecification is resolved by value refinement, and might sometimes reveal conflicting values, resolving which may further require the replacement of some of one’s value. I have already argued earlier that these are all forms of genuine value change. 

(Note, in neither case am I claiming that all cases of uncertainty or underspecification are noticed or resolved; nor do I need to, for my argument to succeed.) 

Value open-endedness

Finally, I argue that because the world is open-ended relative to (cognitively) bounded agents (such as is the case for humans), value change is commonplace. Let us unpack this argument further. First, humans are bounded, meaning that they operate with a limited set of (cognitive) resources as they are navigating the world (see, e.g., Simon, 1990; Wimsatt, 2007). Given that, humans do not come into this world with a complete representation of the world (i.e.e.g., a full hypothesis space). Instead, from the point of view of an individual, the world appears open-ended, i.e.e.g., the world’s limits and its ontology are not clearly defined or unchangeable from the get-go. As a result, people sometimes get exposed to situations they have not considered before and with respect to which we have not yet formed a meaningful evaluative stance. 

One might object that the agent could have held the relevant value latently, and thus, instead of involving genuine value change, these should be understood as cases where a fixed, latent value is being ‘unfolded’ or ‘come to be consciously apprehended’. While this might sometimes be what is in fact what is happening, in order to refute VM, one would have to argue that every value a person will ever come to adopt must have already been held latently. This would imply that a person’s model of the world already contained all representational entities with respect to which they will ever come to hold evaluative stances. Given the cognitive boundedness of real agents, this seems exceedingly unrealistic. Instead, the open-endedness of the world will predictably expose a bounded agent to genuinely novel situations over which they will not yet have had the chance to form any evaluative stances. Because the situation must be understood as genuinely novel, the only interpretation of how an agent could come to adopt an evaluative stance with respect to such novel situations is one that involves genuine value change.[2]

Transformative experiences and Aspirational pursuits

In the following, I will briefly describe two accounts from the philosophical literature which provide extensive treatments of cases of genuine value change. These two accounts are Paul’s (2014) Transformative Experiences, and Callard’s (2018) Aspiration

‘Transformative experiences’ (Paul, 2014) are experiences that permanently and fundamentally transform a person in hard-to-imagine ways, including with respect to their values. An example of a transformative experience is becoming a parent, immersing oneself in a completely novel culture, or going through a near-death experience. It is not uncommon that such experiences change people (including their values) significantly, and, from the point of view of the person going through the experience, it is hard, maybe impossible, to fully anticipate how they will be changed as a result. Nevertheless, most people will experience at least one or a few such transformative experiences throughout their lifetime, and humans do not generally seem to go out of their way to avoid undergoing such a change. Insofar as Paul’s account of transformative experiences holds up—i.e., insofar as these experiences are truly transformative, including with respect to people’s values—they provide a central example of the malleability of human values. 

As for the second account, Agnes Callard (2016, 2018) discusses how, not only in the case of transformative experiences, humans seem to regularly take actions on the basis of what she calls proleptic reasons. These are reasons which are based on value estimates which the reasoner cannot fully access yet, even if they might be able to partially glean them. As such, Callard argues that humans sometimes actively aspire to explore and ‘unlock’ new domains of value before they have become fully able to value said things. For example, one may aspire to be able to appreciate the joys of haute cuisine before one is fully able to do so, or one may aspire to cultivate certain virtues (say courage) without yet being able to fully inhabit all the things that may be good about said virtue. In other words, the aspirant wants to want something, before having full access to why they want that something. As such, and according to Callard’s account, the notion of aspiration refers to a rational process of value development. In Callard’s own words, to aspire is to pursue an activity based on an 'inchoate, anticipatory, and indirect grasp of some good' (2016, p. 132), and as a result, 'we have guided ourselves to the new values or desires or commitments that our experience engenders’ (p. 153). Again, insofar as Callard’s account holds up, it provides an example, and thus existence proof, of value malleability. 

Counterargument : ‘shallow’ vs ‘fundamental’ values

One might object that all of the above examples can be explained by postulating that more ‘shallow’ types of values can change while more ‘fundamental’ values remain fixed. Further, such changes of ‘shallow’ values (e.g., in response to epistemic changes) always stand in an instrumental relationship to the pursuit of the more ‘fundamental’ values which themselves stay fixed.

To illustrate the shape of what I have in mind with this counterargument, consider the following two examples. Suppose that I like chocolates, and that I believe that there are chocolates in a yellow box. I will then also value having the yellow box at hand. Now, suppose I find out that there are no chocolates left in the yellow box. I subsequently lose interest in it. In a certain sense, my values have changed—I no longer value having the yellow box at hand. At the same time, my valuing of the yellow box appears rather shallow, and instrumental to my valuing of chocolates, which seems more fundamental. Furthermore, the latter value has not undergone any change. 

Or, consider the case of Garry, who aspired to, and eventually succeeded at, coming to value poetry. Maybe what happened is, rather than Garry developing a new value in his love of poetry, he simply learnt more about how he can effectively seek pleasure—pleasure being his more fundamental value which remains fixed throughout his pursuit. Under this account, poetry is ‘merely’ an effective strategy to achieve pleasure. 

Do these examples provide a counterargument to VM? To answer this question, let us first distinguish between two ways we might interpret the notions of ‘shallow’ and ‘fundamental’ in this context. 

On one account, the more fundamental a value, the less likely it is to change. However, it is still in principle possible for (most) values to change, and in particular, the more shallow values change regularly. This interpretation of the notion of ‘shallow’ and ‘fundamental’ does not conflict with VM. In fact, this appears to present a realistic picture of the nature of values in that, for example, it can account for the intuition that some values are more (evolutionary) hardwired (e.g., the value to not be hungry)—and therefore less likely to be subject of change—while others are more contingently acquired (e.g., a preference for some foods over others) and more likely to change.

As for the second interpretation, however, we do find it to stand in conflict with VM. According to this view, the changes of the ‘shallow’, derived values (such as valuing the yellow box, valuing poetry) can be understood as ‘merely’ a case of rational sensitivity to epistemic learning in light of, and instrumental to, more fundamental (and fixed) values. As such, ‘shallow’ values (under this interpretation) don’t constitute values of the same standing or importance as ‘fundamental’ values. Some might argue that they don’t count as genuine values at all (or something similar).[3] As such, under this view, what we observe in the yellow box and the poetry example does not constitute cases of genuine value change, but instead merely rational re-evaluation of the most effective strategies for realising one’s fundamental (and fixed) values. 

However, this counterargument faces difficulties and, as such, I am not convinced it succeeds. For one, it is always possible to postulate some invariant higher-order value with respect to which changes to shallower values can be explained (e.g., my value for chocolates explains my ‘shallow’ ‘value’ for the yellow box). Within this account, nothing constraints us from constructing arbitrary ‘deep’ hierarchies of value which can be used to rationalise any changes of the ‘lower-level’ values in terms of fixed ‘higher-level’ values (e.g., my value for tasty food explains my more ‘shallow’ value for chocolates; my value for chocolates, in turn, is explained as merely an instrumental strategy in service of my ‘more fundamental’ value for hedonic pleasure, etc.). I find this problematic because it explains too much (and thereby nothing at all). 

Second, one concern I see with this approach is that we risk abstracting away too much from concrete, contextualised and thick notions of value, replacing them by increasingly more abstract ones (e.g., ‘pleasure’ or ‘utility’), thereby stripping away more information than is desirable. From the perspective of trying to explain and predict phenomena around us, we risk ridding our ‘theory’ of any substantive empirical or predictive content, and the notion of (fundamental) value we end up with ceases to be useful in understanding the (often contextual!) practical reasoning and behaviour of actual humans. In the example above, saying that Garry seeks to maximise (some abstract notion of) ‘pleasure’ or ‘utility’ does not constraints whatsoever my predictions about whether, e.g., he will get into liking jazz, far from having anything to say about why or how he might do so.  

Beyond being unsatisfying from the point of view of the theorist, this approach is also problematic from the perspective of being moral reasoners ourselves. In particular, adopting overly abstract notions of value risks impoverishing our very ability to recognise and derive value, and to deliberate with each other about what we value and why. Imagine if Garry’s poetry-loving friend Harriet, who, in one case, is able to talk in rich and nuanced terms about what she finds so marvellous and inspiring about poetry, and, in another case, describes poetry simply as ‘an effective strategy to seek pleasure’. Most people, I believe, would be more motivated to aspire towards coming to appreciate the value of poetry in the former rather than the latter case. Garry’s own ability to aspire—the quality of his proleptic reasons, his inkling of what might be valuable about poetry—is likely also stronger if his friend is able to talk to him about her love for poetry in thick and nuanced terms. In other words, by reverting to overly abstract notions of value, we impoverish our very ability to reason and deliberate, collectively and with ourselves, about our values. Elizabeth Anderson, in Value in Ethics and Economics, makes this point better than I can:[4]

[I] deny that it makes sense to reason about the good and the right independent of thick evaluative concepts. [...] It disabled us from appreciating many authentic values. It suppresses the parallel evolution of evaluative distinctions and sensibilities that make us capable of caring about a rich variety of things in different ways. It cuts off fruitful avenues of exploration and criticism available on a pluralistic self-understanding. (p. 118.) 

There are a number of other things that could be said about this problem, both in favour of and against defending some notion of fixed fundamental values. I am happy to hear about arguments (in either direction) that I seem to have missed here. For now, this is where I will leave the discussion. I remain unconvinced, for the time being, by this line of counterargument against value malleability.

  1. ^

     Yes, saying ‘in the relevant sense’ (i.e. invoking a notion of ‘relevant counterfactuals’ sweeps a bunch of nuance under the rug. I will touch on this question again, at slightly higher resolution, later in this post. For the time being, I suggest that a common sense/pragmatic interpretation is sufficient for the argument at hand.

  2. ^

     A practically interesting area in which this dynamic occurs regularly concerns innovation and technological development. The transhumanist literature, for example, is full of examples of how changes to the biological constraints faced by humans raise significant novel moral questions. Consider, for example, how novel means of contraception have caused the transcendence of earlier constraints set by the fallibility of natural methods of contraception. This expansion in contraceptive possibilities has raised questions and changed people’s attitudes concerning, among others, the social role and rights of  women, sex and marriage. Or, consider how the development of spacesuits has caused the transcendence of the earlier constraints set by humans’ inability to survive without an atmosphere. The resulting possibility of space travel in turn raises questions about whether there is a moral case to be made for or against humanity becoming space-faring, and at what cost relative to, say, other social priorities (e.g., differential investment in education or health care). Finally, consider how the possibility of artificial wombs, if they come to be, will affect people’s values related to gender, parenthood, people’s rights over their bodies, etc., in ways that cannot be conclusively predicted at this point in time. In other words, innovation creates situations of moral relevance which an individual and/or a society has not yet formed a coherent stance towards, and which thus necessitates value change. 

  3. ^

     For example, I find it worthwhile noting that under expected utility theory (e.g., following the Von Neumann–Morgenstern utility theorem), any case of value change is necessarily understood as undesirable (e.g. value drift) and/or the result of a failure of rationality (i.e. irrationality). Accordingly, the only way one could rationally change one's mind about ‘valuing the yellow box’ is for the latter to not actually be a genuine ‘value’ but merely a contingent and instrumental ‘strategy’ which it is fine to change in light of new information.

  4. ^

     The original quote is part of a defence of pluralist over mostist theories of values. However, the claim about the role of ’thick evaluative concepts’ is directly relevant to our discussion at hand. 

New Comment
1 comment, sorted by Click to highlight new comments since:

I think whether or not people change their values is a matter that can be resolved sufficiently by polling them at various ages to see what they think about stuff.