In this chapter and the next, I will present various ideas on how to reduce suffering in a complex intelligent system acting in a complex world—such as humans. I derive various ways how information-processing should be changed, i.e. how the agents should be reprogrammed, based on the theory presented in this book. Since the systems in question, such as our brain, have largely learned their function from input data, an important part of such reprogramming is retraining the learning system by inputting new data into it.
The methods discussed here are not original: almost all come from Buddhist and Stoic philosophy or related systems. The goal here is to interpret them from a computational AI perspective, using the theory developed in this book. Thus, we gain more understanding on how they work, why they work, and what could be done to improve them.
The main starting point in this chapter is the frustration equation we just encountered (🡽). We can try to reduce suffering by changing any of the terms on the right-hand side of the equation, since that inevitably implies that the frustration on the left-hand side of the equation is reduced. We can see from the equation that the obtained reward should be increased because it has a negative sign in the difference computed. In contrast, all the other terms on the right-hand side should be decreased because their contribution to frustration is positive.
Maximizing the obtained reward is really a very conventional way to try to reduce suffering, based on the wide-spread view that happiness comes from having achieved all your goals and having got what you wanted.1 However, that is difficult for reasons which are rather obvious. Many resources are limited: not everybody can have the best cars, the best wines, and the best sex partners. There is fierce competition over such resources, and not everybody can win. Besides, expectations are adapted to the obtained level of rewards, so what used to feel good no longer brings happiness after a while, as discussed in Chapter 5 (see 🡽).
So, instead, we attempt here to reduce all the terms other than the obtained reward in the frustration equation. In this chapter, we start by considering how it is possible to reduce two of those terms: the (perception of) expected reward, and the certainty attributed to the perception of reward loss. (The next chapter will consider reducing the remaining terms, as well as some further methods.) Such reduction also includes reducing self-needs as a special case, thus complementing the frustration equation by the logic of the flowchart in Fig. 15.1. Ultimately, such practices lead to reducing all desires and aversions. This approach may be rather unusual in the context of modern Western psychology and philosophy, but it is thoroughly standard in Buddhist and Stoic philosophy.
Let us first look at the term “perception of the difference of expected and obtained reward”, i.e. perception of reward loss, in the frustration equation (🡽). This should be made as small as possible, ideally zero or even negative. As already mentioned, the most conventional way to reduce it would be to try to increase actually obtained rewards, but that is very difficult. So, we need to do something more clever. A well-known idea in Buddhist and Stoic philosophy is to lower your expectations. Then, your reward loss should be smaller; it will perhaps vanish altogether.
The expected reward is typically a product of two things: the probability the agent assigns to obtaining the reward, and the actual amount of the reward if it is obtained (considering the basic case where the amount of reward, if obtained, is fixed). Thus, reducing the expectation of a reward can be accomplished in two ways: either reducing the probability the agent assigns to the reward, or reducing the value it sees in the reward. This can be compared to a lottery. Suppose your initial chance of winning a Porsche is 1%. Obviously, the lottery would be made less attractive if the probability of winning is lowered to 0.01%; it would also be less attractive if you realized that the Porsche is second-hand and not so cool after all. In both cases, your expected reward is reduced.
Most importantly, rewards in the real world are always a bit subjective, and so are the probabilities we assign to them. A new Porsche may feel like a great reward to one person, while it may matter very little to another; this is why we have to talk about perceived reward. People will also have very different guesses of the probability of winning it. Since these quantities are subjective, it is possible to change our estimates of them by changing our beliefs, perceptions, and associations, even if the actual physical reality remains unchanged.2
A key goal of Buddhist and Stoic systems is exactly such re-evaluation of the probabilities and rewards. To accomplish this, Buddhist philosophy talks about the “three characteristics of existence”, which are impermanence, no-self, and unsatisfactoriness. They map roughly to our concepts of uncertainty, uncontrollability, and unsatisfactoriness we discussed in the preceding chapter.3 Each of these characteristics gives a reason why the rewards are actually lower than what they would otherwise be, or what they appear to be, as will be explained next.
Uncontrollability, discussed in Chapter 13, is a key concept here. The level of controllability is clearly related to the level of expected reward. If you think the world can be controlled, you will expect to achieve high rewards, because you think you are able to take courses of action that give you the very highest rewards, and you are reasonably certain that you can achieve them. Thus, you’re exposed to strong frustration since your expectations are high. In contrast, if you think the world is uncontrollable, you assign a low probability to achieving any rewards, and the higher rewards may seem to be completely out of your reach. Then, your expectation of reward is smaller, and you are less likely to suffer from a reward loss, i.e. frustration. This is how considering the world to be uncontrollable reduces suffering.
To transform this logic into a method for reducing suffering, the trick is to acknowledge the fact that you have little control and there can never be very much control, and firmly believe in that fact. We saw earlier how the Stoic philosopher Epictetus emphasizes how little we can control (🡽). He continues by explaining that if we are mistaken about this point, suffering is inevitable:4
The things in our control are by nature free, unrestrained, unhindered; but those not in our control are weak, slavish, restrained, belonging to others. Remember, then, that if you suppose that things which are slavish by nature are also free, and that what belongs to others is your own, then you will be hindered. You will lament, you will be disturbed, and you will find fault both with gods and men.
To put this idea into practice, on every occasion where you are inclined to develop a desire or aversion towards an object or event, you should ask yourself whether you can control it or not. Basically, you cannot control anything external to you, such as your possessions, other people, or their opinions. Epictetus suggests you should not see any such external, uncontrollable things as good or bad, or as bringing any reward in our terminology.5
Likewise in Buddhist philosophy, the original form of the no-self philosophy says that nothing is part of me, which is a way of saying that nothing can be controlled, as we saw in Chapter 13. Understanding this is crucial according to the Buddha:6
All [mental phenomena], whether past, future or present, internal or external, gross or fine, inferior or superior, far or near, should be seen with one’s own knowledge, as they truly are, thus: ’This is not mine, this I am not, this is not my self.’ (...) [S]eeing thus, [the disciple] grows wearied of form, wearied of feeling, wearied of perception, wearied of volitional formation [i.e. desire and aversion], wearied of consciousness. Being wearied, he becomes passion-free (...), he is emancipated [from processes leading to suffering].
Here, I interpret “growing wearied” as signifying that the reward expectations are lowered, or little enjoyment anticipated. Thus, the point is that recognizing uncontrollability, or inexistence of self, reduces expectations of reward, and this reduces suffering.
Buddhist philosophy further emphasizes the importance of understanding “causality”. Such causality means that events in the world just follow from each other based on natural laws, for example those depicted in Figure 15.2. This thinking minimizes the importance of free will and the control that the agent can have over the world; it is related to what is called determinism in Western philosophy. I would think, therefore, that the Buddhist emphasis on what they call causality is just another viewpoint on uncontrollability; seeing such causality is one way of realizing that the world is uncontrollable. Stoic philosophers advocated the study of the natural sciences7 (which they simply called “physics”), with a similar goal.8
Uncontrollability is closely related to the concept of uncertainty. Uncertainty feeds into uncontrollability: if the workings of the different objects in the world are uncertain, even quite random, the world cannot be very well controlled. Likewise, uncontrollability leads to uncertainty about whether rewards will be obtained. In some sense, these are two sides of the same coin.
Buddhist philosophy focuses on the related concept of impermanence, which can be largely seen as a special case of uncertainty. Impermanence means that the world is constantly changing, and usually in unpredictable ways.9 For example, any object that you possess can break or get lost. Any enjoyment that you get is likely to be fleeting. In fact, even your feelings and opinions are impermanent: today you like one thing, but perhaps tomorrow you’re already bored with it and want something else; what you consider important today may have no significance to you next month. Obviously, impermanence thus interpreted leads to uncertainty.10
Going back to our frustration equation, the consequences of uncertainty are very similar to the consequences of uncontrollability. The central point is that any future rewards are uncertain, i.e. unpredictable. Rewards and the circumstances leading to rewards can change, so an agent cannot really know whether it will get any reward after executing its plan. Thus, the agent should lower the probability it assigns to any future reward. If the agent acknowledges such uncertainty of the world, its expectations regarding rewards will be lowered, just like in the case of uncontrollability. Consequently, frustration will be reduced. (Later, I will talk about perceptual uncertainty, which has a different effect on suffering.)
It is quite paradoxical that Buddhist practice, which turns your attention to uncertainty and uncontrollability, tends to reduce stress and suffering. In Chapter 6 we saw that uncertainty and uncontrollability are usually thought to lead to more stress, not less. I think the paradox has a lot to do with one’s attitude to uncertainty and uncontrollability. Somehow Buddhist philosophy seems to result in a particularly appropriate attitude, related to their acceptance, which will be considered in more detail in Chapter 18.11
In Buddhist philosophy, the two characteristics of impermanence and no-self (roughly, uncertainty and uncontrollability) are complemented by a third characteristic: unsatisfactoriness, which has many meanings and interpretations. On the one hand, it expresses the idea that whatever we try to achieve, we often fail due to uncontrollability and uncertainty. In this sense, it simply recapitulates those two aforementioned properties. On the other hand, unsatisfactoriness can be seen as an extremely general characteristic which penetrates all phenomena and all existence. In fact, in the original Indian texts, the single word dukkha is used to express such unsatisfactoriness as well as suffering, i.e. this very thing we are trying to reduce. One could express the relation between these two meanings by saying that all phenomena are unsatisfactory in the sense that they can produce suffering, one way or another.12
In Buddhist philosophy, it is recommended to acknowledge the unsatisfactoriness of all phenomena. A basic justification for this can be constructed using our frustration equation: if the agent is strongly convinced about the unsafisfactoriness of all phenomena, its expectation of reward will be very low, and the reward loss will be small and rarely even occurs. Thus, here we are talking about a very general, if a bit vague, strategy for lowering the expectations of rewards. As a training method of great generality, the Stoics suggested reviewing any plan of future action with the view of anticipating what could go wrong and how the plan will not lead to great enjoyment after all. Epictetus gives a famous example of going to a Roman bath:13
If you are going to bathe, picture to yourself the things which usually happen in the bath: some people splash the water, some push, some use abusive language, and others steal.
With this mindset, you will not expect much enjoyment, i.e. reward, and you will not be disappointed. Such a scenario could be analyzed in terms of uncontrollability and unsatisfactoriness as well, but unsatisfactoriness may be a more natural viewpoint.14
Using our computational theories, we can penetrate still deeper into the meaning of unsatisfactoriness. Suppose you can get chocolate quite easily and there is little uncertainty about its great taste. It is not obvious how uncertainty or uncontrollability would be a problem here. But, we can still say that the chocolate is unsatisfactory because there are various negative long-term side effects hidden in the apparently rewarding object.
To see what such negative long-term effects might be, recall how in Chapter 15, we defined unsatisfactoriness in a more specific way, based on two computational ideas. First, we had the concept of insatiability (Chapter 5). An intelligent system programmed to maximize reward will never be satiated or satisfied, by the very construction of the system. It will never find that it has had enough, because in the long run, getting more reward will increase the expectation of rewards. In a word, the system is infinitely greedy. The second aspect of unsatisfactoriness in our framework was evolutionary obsessions (Chapter 5). Even the very goals pursued and the rewards obtained can be questioned. Perhaps the evolutionary system gives you a certain reward for drinking a sugary drink. But we know very well that such a reward is misleading: the sweet drink is not good for you when all its effects are considered in the long run.
These two computational viewpoints of unsatisfactoriness point at mechanisms which are very different from uncontrollability or uncertainty, or even simply reducing reward expectations. The implication is that even if we could totally control the world and everything were certain, the result of our strivings would not be that great anyway because it would not produce a lasting satisfaction or pleasure. While uncertainty and uncontrollability are more about the probability of getting various kinds of rewards, unsatisfactoriness (both in our sense and the Buddhist sense) is really about the real worth of the rewarding objects or events themselves, when considering the bigger picture. Even the very best chocolate, if you eat it every day, will ultimately leave you indifferent, and may ruin your health in the long run.
Our definition of unsatisfactoriness actually works a bit outside of the frustration equation because it is not that the rewards or their probabilities (or any other terms in that equation) are changed: it is rather understood that even if the rewards are obtained, there will be side effects in the distant future. The frustration equation is in a sense short-sighted: it only considers the direct, immediate effects of rewards or their simulation.15 In contrast, the ideas of insatiability and evolutionary obsessions bring a longer time scale into the picture, pointing out that obtaining rewards now may actually increase frustration and suffering in the long run. 16
Another term in the frustration equation that we can reduce is the “level of certainty attributed to that perception”. As we saw in Chapter 12, perception is uncertain. To recapitulate the main ideas: perception is based on limited data, thus necessitating unconscious inference, which may not always be much better than guessing. Perception is also subjective: different people can have different priors and thus different perceptions. Subjectivity is made even more serious by the strong selection of incoming information by attentional mechanisms. Since the computational capacity is always fundamentally limited, and the world is awesomely complex, it is not possible to build a perceptual system that always makes correct inferences, let alone one that perceives the “true” reality. This should imply a fundamentally skeptical attitude towards any perception: we should not make too strong conclusions based on sensory input.
Thus, we see that uncertainty has two different aspects. There is the objective unpredictability of the world: surprising and unexpected things can happen, the world is to some extent random—this is the kind of uncertainty we focused on earlier in this chapter when talking about the importance of recognizing uncertainty and impermanence. But here, we focus on the uncertainty in our perceptions and beliefs of the world, which I here call perceptual uncertainty. The point is that we don’t know with any great certainty what the state of the world is, since we have neither enough data nor enough computation to perceive it properly.17 Such perceptual uncertainty increases the effects of unpredictability and uncertainty that we saw earlier, since it makes the world even more unpredictable for the agent.
If the agent is intelligent enough, it will take perceptual uncertainty into account when evaluating the reward loss or frustration. Suppose the agent has completed an action sequence in view of getting reward, and it tries to evaluate the reward loss. Now, the agent should understand that it cannot know with certainty how much reward it got. A drink may have tasted good, but you cannot know if it was actually good for you. That is why in the definition of reward loss, we should really be talking about perceptions of rewards instead of any objective quantities; this is precisely what is done in the frustration equation.18 Since the reward loss is uncertain, any conclusion drawn from it should not be given too much weight.19 This is an implication of the basic principles of Bayesian inference as used in AI; many philosophers over the centuries have also pointed out that what first appears to be a negative outcome may even turn out to be positive, and vice versa.20
If the agent is programmed to take account of the fact that all its perceptions are uncertain, it would likely have weaker reward loss signals. Consider an agent that attempts to get some chocolate. Suppose that after executing a plan, the agent is able to eat some, but its program “understands” that it does not really perceive the amount of chocolate with any certainty; perhaps because it swallows all of it immediately without really taking a look. Intuitively, it does not then make a lot of sense to send a strong reward loss signal: such a signal would be too much guesswork and would not provide a proper basis for learning better behavior. In other words, uncertainty about the correct signal to send should lead to a weaker signal.21 Thus, taking account of the uncertainty of the perception of reward would reduce suffering.22
The perceptual kind of uncertainty has a central role in the later Mahayana schools of Buddhism. While the “three characteristics” (impermanence, no-self, unsatisfactoriness) form the core of the Buddha’s original philosophy, later Buddhist philosophers found them somewhat simplistic. The emphasis shifted to the properties and limitations of perception and cognition, as opposed to characterizing the outer world. The inaccuracy of perceptions and beliefs became essential as part of the multifaceted concept of “emptiness” widely used in Mahayana Buddhism—although rarely by the Buddha himself.23
Emptiness has many meanings. In the framework of this book, we can consider emptiness as an umbrella concept encompassing several of the ideas related to information-processing that we have seen in this book, in particular uncertainty, fuzziness, subjectivity, and contextuality. To summarize it in a single word well-known in Western thinking, we could call it “relativity”.24 What the different aspects of emptiness have in common is that fully appreciating them should make us take the contents of our minds less seriously.
While it seems fashionable to discuss such concepts in terms of Buddhist philosophy, very similar ideas can be found in Greek philosophy. We already saw the Skeptics questioning the reliability of any sensory information in Chapter 12.25 On the other hand, Plato’s famous theory of “ideas” (or “forms”) describes a kind of true reality behind the sensory phenomena, thus denying the true existence of the phenomena. Seneca explains how this theory is related to reducing desires:
[A]ll these things which minister to our senses, which arouse and excite us, are by Plato denied a place among the things that really exist. Such things are therefore imaginary, and though they for the moment present a certain external appearance, yet they are in no case permanent or substantial; none the less, we crave them as if they were always to exist, or as if we were always to possess them.26
Indeed, Seneca reads Plato as if he were a Buddhist philosopher propounding emptiness philosophy.
Concepts and categories are considered particularly empty in Mahayana philosophy. It proposes that the objects in the world do not really exist as separate entities, but are just part of a complex flux of perceptions happening in our consciousness. In this sense, there are really no separate objects or crisp categories in the world; they are purely constructions of the mind. Zen texts use the parable of confusing the moon and the finger that is pointing at the moon. Here, I would interpret this in the sense that the finger is a category, perhaps expressed by a word, that merely points at a phenomenon in the real world, that is, the moon. Ceasing to think in terms of categories and concepts, based on a recognition of emptiness, is something that generalizes the idea of reducing the certainty attributed to perception, or in fact, to your cognitive processes in general. It reduces frustration according to the logic given above for recognizing uncertainty of perception. Furthermore, any valence that you would typically associate with a category cannot be considered certain anymore.27 What may be Epictetus’s most famous quote says: “Men are disturbed not by the things which happen, but by the opinions about the things.”28
In the frustration equation above, we didn’t have any terms explicitly related to self. Yet, self is obviously an extremely important concept from the viewpoint of suffering, as seen in Chapters 6 and 13. In our framework, self creates its own kind of frustration, by bringing aspects such as self-preservation, self-evaluation (or self-esteem), and control into play. As such, self-related suffering is covered by the frustration equation as a special case.
Many philosophical traditions such as Buddhism encourage reducing self-related thinking as a means to reduce suffering. One case of self-related thinking is related to the self-evaluation system. In Chapter 6, it was proposed that a self-evaluation system constantly computes whether we have gained “enough” reward recently, looking at the relatively long-term performance of the system. (This long-term evaluation system is different from the one which computes the ordinary, short-term reward losses in the first place.) Such self-evaluation creates, as it were, another frustration signal on a higher level, in case the result of the self-evaluation is worse than some set standard.
Logically, there are three ways of reducing negative self-evaluations. The first is similar to the “conventional” approach we discussed above regarding ordinary frustration: it is to really gain a lot of reward, so that you surely reach the standard required. This is obviously easier said than done. Furthermore, such striving may not reduce suffering at all because gaining a lot of reward may increase the expectations in the future, resulting in insatiability on a “meta-level”.29 The second approach, in line with the main proposals in this chapter, is to lower the standard of expected reward. For example, the aforementioned philosophical viewpoint that everything is unsatisfactory should work here as well. If the system expects little reward even in the long run, the self-evaluation should not claim that the agent did not gain enough.
However, there is clearly a third option: shut down the system that evaluates your long-term success. Such a shut-down is possible by convincing yourself of the total futility of the self-evaluation. The Buddhist philosophy of no-self should be particularly useful here. Admitting the lack of control, even lack of free will, implies that there is little to evaluate. If we cannot influence the world and the level of obtained rewards, what is the point in evaluating my actions and learning strategies? On a deeper philosophical level, if it is not me that actually decides my actions—say, it is my neural networks—-who is to be evaluated? Perhaps my neural networks and my body could still be evaluated, but not “me” really. On the other hand, what if “my” actions are ultimately determined by the input data or the environment, not “myself”?
Suppose an agent were somehow able to shut down its self-evaluation system. It could be objected that such an agent with no self-evaluation might no longer be functional. However, even if the long-term self-evaluation were completely shut down, the system could still achieve most of its goals, and it will even be able to learn. Learning might just be slightly impeded because the learning system would not be optimally tuned to the environment. Thus, only “learning to learn”, a kind of meta-learning, would be shut down, while the agent would be perfectly functional otherwise, even without self-evaluation.
I should emphasize another crucial point about self-evaluation. As long as the self-evaluation is based on evolutionary fitness, including what I called evolutionary obsessions, it does not actually make a lot of sense for us. It is too often based on criteria that are not in line with what humans should strive at, according to mainstream ethical principles. We need better criteria to decide if our actions were “good enough”; criteria that would be more in line with what we consider a good human life should be about.30
Likewise, reducing the survival instinct, or information-processing aiming at self-preservation, would seem to be useful for reducing suffering. Again, it could be objected that it is not good for the agent: such reduction may increase the probability of injury and even death. If I had no survival instinct, I might just happily go and pat a tiger I see in the jungle. This is a valid point, but we could still try to reduce the intensity of suffering incurred. In fact, religions and spiritual traditions invariably propose some method to cope with fear of death and mortality. Fear of death may sometimes be paralyzing, and quite often, it is unreasonable since I may even suffer from seeing a tiger on TV. Therefore, a moderate reduction in survival instinct might have mainly positive consequences. One method would be to reduce the mental simulations of injury and death; we will get back to this point in the next chapter, where we look at reduction of simulation by meditation.
In addition to reducing the two specific self-needs just considered, we can aim at a more general “reduction of self”, which can take many forms. To begin with, if we see the self as the source of control, and then we recognize uncontrollability as discussed earlier in this chapter, this can actually be seen as a way of reducing the power of the self. As far as the self is about control, giving up control is, figuratively speaking, giving up part of the self. More precisely, it is rejecting part of the power that self-centered processing has on us.
Another approach is limiting the number of things that belong to “myself”. Typically, I would consider that a number of things belong to me: perhaps my family, my home, my job, and so on. If I think of them as “mine”, I invest them with a certain power because I think I should be able to control them, as well as keep them intact. In other words, I think that they are in a sense part of myself; some would say I “identify” with them. Then, if anything bad happens to them, or anybody tries to take them away from me, I will have a strong negative emotion as if my self were threatened—and in a sense the intactness of my person or self is threatened.
It is clear how one can reduce suffering coming from such possessions: as a first approach, just own fewer things. If you have very few things that you consider yours, it is less likely that you will experience them breaking down, being stolen, or getting lost. Many spiritual traditions do recommend giving up most of your material possessions. Further, you can try to change your attitude towards such external parts of yourself. Epictetus proposes that you should think of all your possessions, your family, and so on, as not really belonging to you, but as things that have been temporarily lent to you:31
Never say of anything, “I have lost it”; but, “I have returned it.” Is your child dead? It is returned. Is your wife dead? She is returned. Is your estate taken away? Well, and is not that likewise returned?32
Finally, the reduction of self can be approached from the viewpoint of reducing thinking in terms of categories. Typically, I divide the world into things that are part of myself and things that are not part of myself. This is how I construct the category “self”. Like with other categories, it would be useful not to take this category too seriously, and understand its fuzziness and arbitrariness, or emptiness. “Self” can be seen as the ultimate category that should be deconstructed and given up. Such giving up of the whole category of self, in a sense, encompasses all the other aspects of no-self philosophy described above. If the very category of self does not exist, or, to put it simply, if self does not exist, what would be the point in self-preservation or self-evaluation, or any attempt to control? Any such self-related thinking should vanish if the underlying category of “self” is given up. The Buddha said that when a monk is advanced enough, “any thoughts of ‘me’ or ‘mine’ or ‘I am’ do not occur to him”.33 This is the most general way of reducing suffering based on no-self philosophy.34
While so far we have focused on reducing the frustration of desires, many philosophical traditions propose that desires themselves should be reduced—as always, this includes aversions. In Buddhist philosophy of the Theravadan school, it is traditionally the main focus of the training, and it is the main point of the Buddha’s teaching as expressed in the Four Noble Truths. After describing what suffering is (quoted at 🡽), he proposed that it is born of desire, and that one can be liberated from suffering by eradicating desire by following a path of meditative and other practices.35 Epictetus was equally clear about the importance of not having desires or aversions, especially towards things we cannot control:36
Remove aversion, then, from all things that are not in our control (...) But, for the present, totally suppress desire: for, if you desire any of the things which are not in your own control, you must necessarily be disappointed; and of those which are, and which it would be laudable to desire, nothing is yet in your possession.37
Humans can indeed reduce frustration simply by giving up some unnecessary goals: you don’t really need a fancy car. It is possible to consciously decide not to strive for certain goals, and we can modify our desires to some extent without any special techniques. In our framework, this in particular means reducing intentions, i.e. commitment to plans, also called attachments in Buddhist terminology. Intentions can, in fact, be easier to reduce than desires themselves, as may be intuitively clear and will be discussed in more detail in the next chapter. Suffering will then be reduced since if there are no desires and no goals that need to be achieved, frustration will not appear, and neither will suffering. As such, reduction of desires is a central mechanism through which reduction of frustration is possible.
Many ideas in this chapter can be seen as mental techniques serving the very goal of reducing desires. Consider, for example, reducing expected rewards as considered above: why would the agent want anything if it has arrived at the conclusion that the expected rewards are zero, or very small? Likewise, desires will be reduced by adopting the belief that many desires are pointless and even bad for you, they are just evolutionary obsessions. As such, reducing desires is closely related to the earlier ideas of facing uncertainty,38 uncontrollability, and unsatisfactoriness, and in fact, in a traditional Buddhist account, the main justification for such philosophical attitudes is precisely that they reduce desires.39
There are also special techniques to reduce desires. One example is choosing to pay attention to good things that one already has, instead of things that one might obtain. This reduces desires and the tendency of insatiability; it is central in mental exercises based on gratitude, which will be treated in more detail in Chapter 18. Reduction of desires is also facilitated by a simple, possibly ascetic lifestyle where there are fewer stimuli, or “temptations”, that might elicit desires. A fashionable example is taking a break from the internet or social media; the monastic rules of Buddhist monks and nuns give a more radical example. Epictetus also proposed a rather extreme form of training for this end, namely contemplation of death:40
Let death and exile, and all other things which appear terrible be daily before your eyes, but chiefly death, and you will never entertain any abject thought, nor too eagerly covet anything.41
Yet, there are also desires that are really “hot”, hard-wired, and difficult to modify, let alone reduce, based on the rather purely philosophical or intellectual considerations presented in this chapter. What is needed are special techniques that work on deeper levels of the mind than philosophical thinking. Meditation is one such method, as we will see in the next chapter.42