The preceding chapter presented several directions in which information-processing should be changed in order that suffering is reduced. We also saw some practical suggestions for reprogramming, such as seeing the uncertainty and uncontrollability of the world and reducing desires and self-needs. This will eventually lead to a reduction in reward loss, frustration, and suffering. Yet, the account of the preceding chapter may be rather unsatisfactory for some readers: It seems to be asking the impossible, at least in the case of mere humans. The goal is to change some fundamental beliefs about the world and your mind. How is one supposed to become so thoroughly convinced about, say, the uncontrollability of the world that one is not disturbed by the loss of, say, one’s job or house? Is it not simply “human” to think otherwise? How can you actually reduce expectations of rewards, belief in the certainty of perceptions, and so on?
Crucially, what we need are changes in neural associations which work on an unconscious level, and that is notoriously difficult. We need to develop practical methods for retraining the neural networks in the human brain.
In this chapter, I consider meditation, or mindfulness training, as a method that can radically boost retraining of neural networks, compared to straightforward attempts to change thinking at the conscious level. It also turns out that meditation has further benefits, such as reducing “hot” desires, reducing simulation and developing metacognition. I will not go too much into the practical details of any such training methods, on which hundreds of manuals have already been written. Rather, I discuss general principles on how they work, largely interpreting them in the information-processing framework of this book.
First, I discuss how the meditation methods can be seen to speed up learning from new input, thus enhancing the methods of the preceding chapter. Second, meditation can be seen as reducing two terms in the frustration equation 🡭 that we did not yet consider: the amount of attention paid to reward loss and the number of times the reward loss is perceived or simulated; these are also related to the top-row green boxes in the flowchart in Figure 13.1. In fact, emptying the mind by meditation clearly reduces simulation, and meditation almost inevitably seems to develop a metacognitive attitude which changes the attention paid to reward loss. A third benefit of meditation is that it enables stopping the processing chain in the flowchart in Figure 13.2 by increasing conscious control over interrupting desires. Finally, I discuss how meditation can be interpreted as mental relaxation.
The fundamental problem with the approach of the preceding chapter is that a conscious decision to think in a different way often has little effect on what unconscious neural networks do. A conscious decision may not even really change future conscious thinking since it may be overridden by the unconscious networks. That is why reprogramming of the brain must include some kind of retraining the unconscious neural networks.
From a dual-process perspective, the problem to be solved here is how the conscious-symbolic-explicit system can change a mental association which is actually encoded in both the two systems. For example, it might try to create an association between “I” and “impermanent”, being inspired by classical Buddhist philosophy. However, what really matters is changes in the unconscious-neural-implicit system, because it is that system that computes values, expected rewards, and reward losses. So, how can the explicit system force a change in the implicit one? Transfer of knowledge or learning between the two systems is difficult. While one may have a clear understanding that everything is impermanent on a conscious level, it is not easy to transfer this understanding to the unconscious neural networks.
As a first approach, we could use techniques that I here call contemplation. That means a constant conscious repetition of selected thoughts. For example, it can be contemplation of the characteristics of impermanence, uncontrollability, and unsatisfactoriness, possibly combined with some object—such as “I” or my “self”—whose impermanence or other property one wants to learn. The constant repetition of such thoughts on a conscious level should slowly modify the unconscious associations used to compute the perceptions and replay. Some kind of Hebbian learning is likely to construct an association, even on the basic neural level, between the different concepts, such as “I” and “impermanent”. Reading books on Buddhist or Stoic philosophy, as well as later thinking about their contents, can also be seen to be such contemplation.
The mechanism working here is what I would call active replay: The explicit system uses the mechanism of experience replay (see Chapter 9) to make the implicit system learn whatever the explicit system wants. That is, the explicit system in your brain can select thoughts in the form of linguistic sentences or visual images of events—possibly imaginary—and replay them. It can do that repeatedly, thus replaying selected items many times. Such replay will change your neural networks—that is in fact the very point in replay. What is special here is that the explicit system chooses what to replay, thus “teaching” the neural networks, while in ordinary replay, the material would be selected by the implicit system itself.
Such training may seem rather different from modern meditation instructions, but it seems to have been an essential form of practice in the Buddha’s times, and emphasized by some modern Buddhist meditation teachers as well.1 When the Buddha was asked for meditation instruction by monks entering a solitary retreat, he would often tell them to contemplate on impermanence, no-self, or unsatisfactoriness, sometimes linking them all together in various causal chains. For example, he would advise:
You should abandon desire for whatever is impermanent. And what is impermanent? The eye
[and visible forms etc.] is impermanent; you should abandon desire for it.2
Forms [i.e. anything that is seen] are impermanent. What is impermanent is suffering. What
is suffering [i.e. unsatisfactory] is nonself. What is nonself should be seen as it really is with
correct wisdom thus: “This is not mine, this I am not, this is not my self.”3
We don’t know much about the details of how such contemplation was practised in the Buddha’s times. Presumably, it was combined with meditation to speed up the learning—more on that below.
The fundamental problem with such basic contemplation is, in fact, that the learning process can be very slow and inefficient. One reason is that it has the same characteristics as learning in neural networks in AI. As we saw earlier, neural network learning requires a large number of repetitions of input, which changes the neural connections little by little, using some mechanism related to stochastic gradient descent or Hebbian learning. So, retraining neural networks by contemplation requires a huge number of repetitions.
Moreover, transferring learning from the explicit to the implicit system is hampered by the fact that the representations and computations in the two systems can be quite different, as we have already discussed. Suppose that your explicit system repeats the word “impermanence”, in an attempt to contemplate on that property. How are your primitive, lizard-level neural networks supposed to understand what that means? Such neural networks do not operate with words or abstractions but on representations related to sensory input. There is a kind of a communication barrier between the two systems, and contemplation may not be able to cross it very well.
The situation can be somewhat improved if the explicit system imagines events or episodes and replays them as real sensorial input such as images, instead of merely in verbal and abstract form. When you read a story or a simile in Buddhist literature and vividly imagine it happening, that does provide more natural input to your neural networks. Or, the explicit system can imagine future events from the viewpoint that an action plan is likely to produce frustration, as in Epictetus’s Roman bath example (see 🡭).4
A crucial improvement to such contemplation practices is what is called meditation in the modern context. Mindfulness meditation in particular is a technique that can influence neural networks more efficiently than simple contemplation. Mindfulness meditation can incorporate many of the goals described above, such as realizing uncertainty and uncontrollability.5
Typical instructions of mindfulness meditation emphasize objective observation of any contents that appear in your mind, that is, mental phenomena. In particular, that encompasses anything that your senses perceive, including the “internal sense” of thinking and imagination. If you hear something, you acknowledge hearing it, if there is a bodily feeling in any part of your body, you recognize that you have a bodily feeling, and so on. Such observation is done, as far as possible, passively without interfering with the sensory process or the physical source of the perceptions (for example, without moving your body to change bodily feelings). The contents should be observed from an external perspective, as if from a distance, and without judging the contents to be either good or bad.
There are a number of techniques to make such observation easier by regulating the attention of the meditator. Basic meditation instructions typically start by recommending sitting in a comfortable posture and then provide one particular technique for attention regulation. A classical one is focusing on observing the breath, possibly reinforced by mentally counting the breaths. (Alternatively, the focus might be a visual target, or a particular word or phrase that is mentally repeated.) Such observation of breathing should be seen as simply a technique whose goal is to enable better observation of the mental phenomena, and indeed simply counting breathing may sound like an absurd exercise if the actual purpose is not understood. The purpose is to facilitate observation of the mind by making it relatively empty; observing mental phenomena is very difficult if the mind is full of different kinds of thoughts and perceptions. Furthermore, emptying the mind has several direct benefits as well, in particular reduction of simulation as discussed below.
The exact mechanisms of mindfulness meditation are far from being understood, but some of them can be understood by the framework presented in this book, as we will see next.6
The most crucial mechanism at play may be that the meditator learns largely the same things as in the contemplations above but in a more efficient way. I suggest the reason why meditation is more efficient than what I called active replay above is that there is no longer any need to transfer information between the two systems (conscious thinking and neural networks, roughly speaking). Instead, the practitioner observes characteristics such as impermanence first-hand, in real sensory input or imagined sensory content. Then, neural network learning can proceed in a completely natural manner, largely bypassing linguistic constructs and conceptual thinking.
In other words, during meditation, the sensory systems directly perceive how things are, say, impermanent by observing how those things change and disappear. Likewise, the control systems, based on sensory input, find by themselves that attempts to control fail when wandering thoughts invade the mind, for example. Thus, the neural networks learn directly from such natural input. This is in stark contrast to contemplation, where the difficult part is to transform concepts and words into something that can train neural networks, and replay does this in a somewhat contrived way. Neural networks learn best from real sensory input, so it is crucial here to enable them to do exactly that. Such observation is eventually extended to all the aspects discussed in the preceding chapter.
The key trick here is to select the right data to input into the neural networks. As discussed in Chapter 10, selection of input data is an essential part of the perceptual system, in terms of the multi-faceted phenomenon called attention. That is why regulation of attention is a central part of any meditation method: in mindfulness meditation, you typically start by focusing your attention on observing your breath. It is in fact possible to get useful input data from the breath itself, if you do it with a special kind of attentional focus. While the practice may start by simply observing the breath in a general manner, eventually, you can start observing its specific aspects in light of the theory of the preceding chapter. For example, you observe the impermanence of breath, how it is changing all the time from an in-breath to an out-breath—this is a classical Buddhist exercise. That means your attentional system selects your sensory input to consist of observations of your breath, and more precisely, any aspects of your breath related to permanence or lack of it. This is how your neural networks get a lot of good data pertaining to that particular property, and they learn to perceive the impermanence much better than they would by any kind of abstract contemplation based on linguistic concepts.7
So, the explicit system in a sense “teaches” the implicit system, and the teaching happens by means of the attentional system. Direction of attention is, to some extent, under conscious control. So, the symbolic or thinking part of the brain can just tell where the implicit system should be looking—this is only partly a figure of speech— and it does not need to really input anything to the implicit system, unlike in the case with replay. It is a bit like a professor telling students to read a book; she does not then need to give a lecture herself.
Another important example of such direct input is observing how often and easily the mind starts wandering. As we have seen in Chapter 9, sustained attention is difficult, and after a while, the mind often starts wandering, and various daydreams fill the mind. Frequent occurrence of such mind-wandering is extremely salient to anybody who tries to focus on breathing or a similar meditation object. Realizing how difficult it is to focus on breathing gives a direct view into how uncontrollable the mind is. If you systematically observe how automatically your mind starts wandering, you will gradually be convinced—and so will your neural networks— that you cannot control even your own thinking, at least not completely. After all, wandering thoughts are, by definition, a failure of controlling your mind.
Such observation may also convince you that there is no self, no central executive, and perhaps then no free will. Under ordinary circumstances, if I decide to plan what I will do tomorrow, I may have a clear feeling that it is “me” who is doing the planning. However, after observing how planning happens automatically in wandering thoughts, I may be forced to admit that the plans are something that “I” did not create. You may even start having doubts about the correctness and certainty of your thoughts, since they seem to be something that just appears in the mind, and you have little idea why they appear or where they come from. Thus, uncertainty about your thoughts can be taught to the neural networks as well.
To interpret this learning process in the framework of the frustration equation, what happens is that the unconscious neural networks themselves—and not just the conscious and/or symbolic thinking systems—will learn to reduce the expectations of any rewards. This happens through your neural networks learning that the world is uncontrollable and uncertain, which necessarily reduces their expectation of future rewards according to the logic of the preceding chapter.
Many further meditation techniques can be seen as such attentional selection of particular direct input. One classical Buddhist technique is to focus on the ending of any pleasurable feeling. This enables seeing first-hand how pleasure, and in general any effects of rewards, are fleeting and thus worth less than might be expected. Thus, you will learn the impermanence and unsatisfactoriness of all mental phenomena in a particularly efficient way.
In a slight variant of the logic above, mindfulness meditation can also help directly change associations related to specific emotions. An important example is fear extinction. Extinction is the opposite of classical conditioning: It means that when the predictive stimulus (e.g. the bell for Pavlov’s dog) is presented without the other stimulus (the food for the dog), the conditioning weakens. If the bell is presented without the food many times, the dog learns that the bell does not predict the food anymore, and the conditioning is eventually extinguished.
Suppose you have learned to associate a fear reaction with your boss by classical conditioning. Perhaps that was based on a single episode, and the association is not valid anymore, so it would be nice to be able to let such a fear reaction be extinguished. Unfortunately, extinction is often very slow—just like any neural network learning—but this can be improved by mindfulness training. The trick here is that you create completely new data, going beyond simply selecting input from existing data as above, but still feed it directly into your neural networks.
It turns out that mindful meditation tends to make people relaxed and feel good (possible reasons for this will be discussed later in this chapter). So, if you recall the unpleasant episode with your boss many times, but always stay in such a pleasant, calm meditative state, extinction is more likely to happen. Thoughts about unpleasant situations will be increasingly associated with a general feeling of calm; the image of your boss will be associated with relaxation and feeling good in the whole body. This will help override the fear association.8
Unfortunately, such meditation training is still rather slow, even if it improves on simple contemplation. In fact, slowness of training is a ubiquitous problem with neural networks, as already pointed out in Chapter 4. Even though with mindfulness meditation, we have a new source of more direct and natural data for learning, the neural networks still need large amounts of input data, and a lot of meditation practice is needed. Fortunately, the amount of training and effort required can be further reduced by further techniques.
One central principle here is increasing the plasticity in the brain. Plasticity is the biological term for the capacity of neural connections to change and thus to learn. Plasticity in the brain’s neural networks is by no means granted, nor is it a constant quantity. If by some suitable tricks, such learning capacity could be increased, the learning process would take less time. A large amount of neuroscience research has been dedicated to finding different ways to increase plasticity.
Sensory deprivation seems to be one useful trick; it has indeed been shown to increase plasticity, at least in rats and cats. It may be rather common sense that if your brain has had little stimulation for a while, it will better concentrate on any new task. It turns out that its learning capacities are also increased.9 Mindfulness meditation in itself can be seen as imposing sensory deprivation, since it is usually conducted in a quiet environment with eyes closed, or at least there is nothing much happening in the visual field. In some meditation schools, an even stronger form of sensory deprivation may be imposed in the form of silent retreats. Such retreats often entail minimization of any kind of sensory stimulation: the participants don’t go out of a prescribed enclosure, they don’t watch TV or use the internet, and obviously they don’t talk to each other. In several discourses, the Buddha recommended such deprivation, together with meditative concentration, because it makes the mind “pliant” and “malleable”. Then, the meditator is better able to gain insight into, for example, the uncontrollability and uncertainty of existence, as well as better able to learn from those insights.10
Plasticity can further be increased by restriction of food intake, which is another typical characteristic of ascetic training in spiritual traditions. Paradoxically, it can also be increased by the very opposite of sensory deprivation: enriching the environment. In animal experiments, that might mean allowing the animals to live as groups in large, spatially complex cages which are equipped with toys and running wheels. In humans, similar results are obtained by aerobic exercise, as well as action video game playing. Whether such methods could be used to improve meditation practice is a very interesting question for future research.11
Plasticity can also be increased by drugs, such as the antidepressant fluoxetine (aka Prozac). A large amount of research is currently being conducted on new drugs that would increase plasticity even more, and with minimal side effects. The huge impact such drugs could have on society is obvious.12
In fact, you may be wondering why plasticity is such a bottleneck: Why hasn’t evolution made our neural networks learn much faster? The reason seems to be that some limitation of plasticity in the brain is useful to prevent new information from overwriting old information too easily.13 So, it may not be wise to increase plasticity too much, because it could lead to too much forgetting of previously learned information. This is hardly a problem with meditation-based interventions, but with drugs, such negative side-effects might be real.
In principle, an AI has much more freedom in how it changes the results of its learning, and the amount of “plasticity” could be made infinite by design. Thus, an AI could get rid of a bad habit or a harmful association in a split-second, by just removing or changing some connections in its neural network. However, this may not be as easy as it sounds, since just like with humans, there may be a risk of interfering with other connections so that the AI may forget useful information. Also, it may not be clear which connection should be changed in the first place. So, even in the case of an AI, it may be better that all the training happens by simply inputting data—which may be carefully selected—into the system and patiently waiting until the learning happens.
Another major difficulty in meditation training is sustaining attention in the way typical meditation techniques require. I need to emphasize that we actually have two different attentional mechanisms at play here. First, there is sustained attention on the task at hand, meaning that you concentrate on meditation and don’t think about anything else, as explained in Chapter 9. Second, there is sensory, selective attention, which means you select certain data as input to sensory processing, as originally explained in Chapter 10 and extensively used earlier in this chapter. Both are necessary for successful meditation. However, sustained attention tends to be the major bottleneck because it is notoriously difficult to maintain.
In previous chapters, we have actually seen several reasons why sustained attention is difficult. First, wandering thoughts assail the mind, for example due to experience replay. But we also saw that emotions are essentially interrupts; what they are interrupting is current activity, and to do that, they have to be able to grab attention away from wherever it may be. The general concept of the brain as parallel distributed processing emphasizes the idea that there are different networks or modules which are often competing, for example, for attention and the control of attention.
Fortunately, it is possible to learn to use your attentional capacities better.14 This is yet another form of learning, but a bit different from the typical learning we have considered: here we are talking about learning a new skill, as briefly described in Chapter 7. A skill means that you know how to ride a bicycle, to speak a foreign language, or to use your new smartphone; it is opposed to learning facts and increasing your knowledge about what the world is like. Skill learning follows some general laws and these apply to meditation as well. In the beginning, you need to concentrate, and spend a lot of effort, which means a lot of sustained attention. However, in time, meditation becomes more and more automated, which means that less and less conscious effort is needed. Some meditation traditions talk about meditation as “just sitting”, which is in a sense enough if the meditation is sufficiently automated. Importantly, the regulation of attention will in fact become a habit, so will be easily conducted during ordinary life, as if by itself, even outside of formal mindfulness meditation sessions.
So, there are actually two different learning processes at play: Learning that the world has certain characteristics (such as uncontrollability), and on a higher level, “learning to learn” that the world has such characteristics. The latter learning process means learning to meditate in an automated, habit-like manner, with minimum conscious effort. Thus, with practice, the meditator will be able to perform the former learning process with increasing efficiency, and this process is the one that reduces suffering according to the theory of the preceding chapter.
The fact that meditation can become automated and habit-like means that, in a sense, it is no longer my “self” who is meditating. We find echoes of the no-self philosophy treated in Chapter 11. Some neural networks can observe the breathing without any conscious effort, or even without a conscious decision to start meditating. There is no need for any central executive to make any decision, and no need to want to observe the breath; it just happens. It is like when walking, you make no conscious decision to move your feet; you feel no burning desire to put one foot in front of the other.
But if the neural networks are retrained by the explicit system as I argue in this chapter, does that not mean that it is the explicit system, perhaps even a conscious self, which is in control? That might be a hasty conclusion, since there are many ways in which the control is circular. In fact, earlier (🡭) I argued that it is meaningful to say that ultimately, it is the input data that controls us. I gave the example of a meditation master who says that it is actually his master who is meditating, because he still hears his master’s voice in his head. This shows that in order to find the “ultimate” source of control, we have to consider where the data to the explicit system comes from. Part of it clearly comes from the human society and the cultural context: there are other people that input data into us, for example in the form of meditation instructions. How that happens, and who is controlling whom, is a vast topic that I have to leave for future research.15
In addition to speeding up learning in neural networks, mindfulness meditation has further benefits. Next, we consider how it reduces suffering from the viewpoint of cognitive dynamics, which complements the frustration equation. As we saw in Chapter 13, one traditional Buddhist account on a mechanism to reduce suffering is based on the moment-to-moment cognitive chain shown in the flowchart in Fig. 13.2. The idea here is to stop the dynamic process in the flowchart in the middle so that it does not lead to its end product, which is suffering. The point where the process can best be stopped is assumed to be (in the terminology of our flowchart) the three links of desire, intention, and planning.16 It is in fact assumed in early Buddhist philosophy that until the valence computation, the process is too automated, and desire provides the first link that can be stopped.17
This method is distinct from reducing desires by adopting the attitudes of the preceding chapter. Here, I am talking about sudden, “hot”, interrupting desires triggered by the valence computations, and their prevention in real-time when they are about to arise. The preceding chapter focused on reducing long-term desires from the “colder” perspective of reward calculations; this will reduce the underlying tendency for hot desires to arise, but it works only passively in the background.
One problem here is that the hot desires have the properties of interrupts, as explained in Chapter 8, which means they can be quite difficult to prevent. Therefore, it might be better to try to stop the dynamics a bit later, at the links right after desire. In Buddhism, those following links are called “attachment”, which is in our schema divided into forming an intention (i.e., committing to a goal) and planning for that goal.
Whether desire or attachment is chosen as the target, the trick here is to weaken the cognitive dynamics so that this largely automated chain leading to suffering fails to operate. If the desire or attachment is prevented from taking place, no goal is committed to or planned for, and no goal-oriented action is conducted. Thus, the whole frustration equation above is not operating, and frustration is avoided by that route.18
Such stopping of the dynamics before attachment is enabled by well-known mindfulness meditation techniques. The point is to observe the cognitive dynamics repeatedly, so that one learns to introspectively detect the different parts of the process and discriminate between the different links, in real-time. Mindfulness meditation has here the effect of training a new perceptual capacity that allows for observation of the internal mechanisms of the mind.
This is a special case of the phenomenon of “perceptual learning”.19 Research on perceptual learning started in vision science by the discovery that it is possible to greatly enhance the performance in almost any visual perception task; all that is needed is sufficient training. Improvement is possible even in tasks where the limits of perception were previously thought to be set by the optics of the eye, such as the task of telling whether two lines have the same orientation (angle) or not.
In the context of meditation, such perceptual learning allows one to observe the individual elements of mental processes more accurately. An important case of such learning is that it becomes possible to observe the associations between phenomena. If B is associated with A, then, under ordinary circumstances, it may be that the thought of A immediately and necessarily brings B into mind, and it seems that A and B are two aspects of the same thing. But with mindfulness training, it is possible to see how this process breaks into pieces: First there is A, then the association is activated, and then B comes to the consciousness because of the association. This allows one to see the existence and arbitrariness of that association. In particular, one is able to dissociate desires from the stimuli that caused them, as if by creating a “space” between a stimulus (say, chocolate) and the desire, as well as the desire and the attachment that ensues.
This opens up the possibility of breaking the long chain leading from stimulus to suffering depicted in Fig. 13.2 by learning to perceive all links in the chain more accurately, and in real-time as they are happening. Introspectively, the meditators often report that it feels as if the whole process were slowed down. By such perceptual learning, the process is also to some limited extent brought under conscious control. Even if a stimulus leads to a strong valence, the ensuing desire and the following steps will not happen completely automatically, but there is some space for deliberation. Perhaps such breaking of the causal chain is most understandable in the case of planning, which is often a rather conscious process, and as such, it should be possible to decide not to initiate it at all. Obviously, there is a strong unconscious tendency to start planning when desire arises; it is comparable to the unconscious reaction to start scratching a body part that is itching. However, with practice, such an unconscious tendency can be weakened, inhibited, and perhaps even completely removed. That would mean not letting “attachment” arise in Buddhist terminology. The key is to be able to consciously recognize when the planning is being triggered, instead of letting it happen automatically.20
It is important to achieve automatization of such mindfulness by long-term meditation practice, as described above. The learned and automated tendencies of observation can then create the possibility for inhibiting the more innate automatic tendencies of desire and attachment. In fact, if such observation is followed by conscious, deliberate inhibition of desire or attachment often enough, that very action of inhibition will become automated as well. Conscious control processes are often too slow and weak to prevent the processes underlying hot desire or other interrupts, so it is really important to train the neural networks to initiate the action of inhibition as well. Once the neural networks have been trained to perform both the detailed observation and the inhibition during formal meditation sessions, they may be able to transfer that skill to everyday life with its infinite temptations.21
Another additional benefit of meditation is that many people report feeling great pleasure when meditating. This is often attributed to the fact that the mind is strongly focused on a single object, such as breathing, and thus emptied of any thinking. Several traditional meditation schools actually maintain that an “empty” mind is happy, that is, a mind where there are no thoughts, whether wandering or intentional. (Emptiness of the mind does not here refer to the Mahayana Buddhist concept of emptiness we saw earlier.) A similar pleasurable state is sometimes achieved in the state of “flow” where wandering thoughts are equally absent.22
Understanding why an empty mind tends to be happy is one of the deepest problems for a scientific understanding of the mechanisms behind meditation, and not quite resolved at the moment. A number of viewpoints can be taken here. In a traditional Buddhist account, where desire is considered the basis for suffering, a simple explanation would be that an empty mind is happy because it does not have any desires (including aversions).23 On the other hand, Chapter 9 reviewed research showing that wandering thoughts are typically related to a negative mood; however it was not clear if those results apply to all thinking and not just wandering thoughts, and what is the cause and what is the effect. Yet another viewpoint is to recall once more Cassell’s statement that “to suffer, there must be a source of thoughts about possible futures”, which cannot exist in an empty mind.
In the framework of our frustration equation, we can formulate a more computational viewpoint. Reward loss is computed every time a simulation, whether in terms of replay or planning, is conducted in the brain. A reduction of thinking should reduce mental pain since such simulation of frustration or reward loss is reduced. In fact, in our frustration equation 🡭 we have the term “how many times it is perceived or simulated” which gives the number of times the reward loss is computed. Reducing mental simulation will reduce this term, and thus suffering. Reducing mental simulation will, for example, reduce rumination over past errors, simulation of future threats to the person, as well as judgements related to self-esteem, which are some of the most important sources of suffering.24
The logic just given may explain why many meditation methods have the explicit goal of emptying the mind of thinking, or at least reducing thinking. Typically, one concentrates on a single object, such as the breath. This immediately reduces thinking, including wandering thoughts—but does not eliminate them completely, as the meditator soon notices. An important part of meditation is how to react to the occurrence of wandering thoughts. Some meditation techniques directly aim at suppressing them by refocusing on the original object of meditation. Suppose you have any unpleasant, possibly scary wandering thoughts about the future or the past during meditation. If you refocus on the meditation object, thus clearing the mind of such scary wandering thoughts, it is rather obvious that suffering will be reduced.25 Being able to thus prevent negative wandering thoughts from occurring should have a strong positive effect on mood, in line with our logic above based on frustration equation. In fact, it has been shown that the default-mode network, largely responsible for wandering thoughts, is less activated in experienced meditators.26 (Below, we will see an alternative approach to wandering thoughts based on meta-awareness.)
In Buddhist training, there is also a strong emphasis on focusing on what happens “here and now”. In other words, you learn to change your cognitive style to a more “experiential” one, which means you replace most thinking, whether future- or past-oriented, by the simple sensory experience of the present moment.27 This is essentially another shift of attention away from thinking, but this time the shift is to any immediately present perceptual input, instead of any pre-selected object like the breath. It can be practised in everyday life, outside of any meditation sessions. Such an experiential cognitive style is a way of conceptualizing a long-term change in neural networks that leads to a mind which is more and more empty. It can be further motivated from the viewpoint that attentional resources are limited, and one cannot pay attention to many things at the same time. Thus, such a cognitive style can reduce attention to reward losses even in real life, and not only in simulation, since it directs attention elsewhere. In particular, any reward loss is only briefly observed without paying too much attention to it, before attention is directed to something else in the here and now. According to the frustration equation, such reduction of attention reduces suffering, this time based on the term “amount of attention paid” since reducing attention reduces the impact of any perception.28
Buddhist philosophy, as well as the theory in this book, further suggest another very different way for achieving a reduction in replay and planning, which is nothing else than adopting various philosophical attitudes described in the preceding chapter. Planning how to obtain future rewards is likely to be reduced if future rewards are considered lesser; there is simply not so much incentive anymore in planning for them. Likewise, planning to avoid threats will be reduced if those threats are seen as relatively uncontrollable. Furthermore, when the uncertainty of our thoughts and perceptions is realized, spontaneous thinking is often reduced, since there seems to be much less point in simulating something which is uncertain anyway. This is how adopting the philosophical attitudes discussed in this and the preceding chapter will also lead to a reduction in simulation, and towards an empty mind.29
There is an important caveat in any attempt to reduce mental phenomena, be it desires or wandering thoughts. It is important that this training does not lead to the idea or evaluation that the mental phenomena are bad. Such an attitude would, in itself, easily lead to aversion, and thus to suffering. In the extreme case, if there is aversion towards the mental phenomenon of aversion, that may lead to a vicious circle, which constantly increases aversion. To counter this tendency, it may be necessary to actually bring in new mental phenomena so as to neutralize the existing ones.
It may sound paradoxical to say that one should not think of the mental phenomena as bad, or at least undesirable. How could one not think that, say, desires are bad if one believes they lead to suffering? And how is one supposed to let go of them if one does not regard them as something negative, something to be avoided?
The solution to this paradox is that while the actions of the meditator should be chosen so as to reduce desires (or other mental phenomena), it is still possible to avoid creating any new aversion in the sense of a new mental process. Thus, on an abstract level, it is useful to consider the desires “bad”, or perhaps rather as something that it would be better not to have, but such thoughts should just work in the background as weakly as possible, instead of being strong and actively cultivated. In particular, they should not lead to any interrupt-like aversive emotions. Such processing is possible since the neural networks can implement automated habit-like action tendencies that try to avoid certain phenomena, and that can happen without any need to activate the desire/aversion system. As an extreme example, when you are walking, you know you that losing your balance is “bad”, but you probably don’t feel a constant aversion or fear towards stumbling; your neural networks have simply been trained to avoid that happening; they “reduce stumbling” so to say but without any aversion.
In practice, it has been found that with meditation, the tendency to develop aversion is so strong that specific techniques are necessary to reduce it. The key principle is to cultivate the attitude of acceptance. This means a general attitude of accepting all thoughts and sensations that come to the mind, instead of resisting or judging them. More precisely, acceptance here means simply not activating processes of aversion, i.e. not activating a desire to get rid of something. So, acceptance here is taken in a very limited sense; this is not about a moral acceptance, or about not thinking that some things could be bad for you. Such acceptance could also be described as removing resistance; nonreactivity is a related term used in current research.30 For example, a depressed person may be annoyed by the very occurrence of rumination. In such a case, accepting that rumination occurs may actually be beneficial, since it removes the suffering due to the aversion to rumination.31 Again, the acceptance we are talking about has a limited meaning; one can still use various techniques to reduce the rumination.
Such an accepting attitude can actually be adopted towards all mental phenomena. In fact, many mental training systems include some kind of active acceptance practice of all mental phenomena as an integral part. Such an acceptance practice can be seen as a specific method for reducing aversions of all kinds. It complements the methods described so far, which were more oriented towards reducing desires in the restricted sense of the word (i.e., excluding aversion). It is closely related to the practice of letting go that will be considered below.32
Theories such as those explained in this book may help in the acceptance because simply understanding the mechanisms behind, say, wandering thoughts or emotions, may enable you to accept them. If you are convinced that they are natural processes which actually have some computational benefits, and that they are largely outside of conscious control, it may be easy to just let them happen, and naturally go away, without fighting against them. This is related to seeing “causality” in Buddhist terminology (considered in Chapter 14, 🡭), but it goes further, since the phenomena are seen as not only natural and uncontrollable but even useful—at least from an evolutionary viewpoint. Based on this viewpoint, even frustration could be accepted as an unavoidable part of a learning process.
Ultimately, even pain and suffering need to be accepted on some level. Any aversion towards them will create a lot more suffering. As an extreme example, people suffering from chronic pain will suffer even more if they “catastrophize” the pain, resist it, and develop a particularly negative attitude towards it; accepting the pain will help.33 The Buddha gave a famous simile of a man who is struck by an arrow, and thus suffers from physical pain. If he “sorrows, grieves, and laments”, feeling aversion towards pain, he makes the suffering even worse, as if he were struck by a “second arrow”.34
There is one more very particular form of attentional control operating in mindfulness meditation, especially in more advanced stages of the practice: direction of attention and awareness to a metacognitive level. Metacognition means here cognition about cognition, as in, for example, thinking about one’s own thoughts, or observing one’s own perceptual processes. In such a case, the “higher”, metacognitive part of your mind is observing the “normal” thinking or perceiving part of your mind.35 Such metacognition is presumably possible because of the parallel and distributed nature of brain function, which means one part of the brain can observe what is happening in another part.
An obvious utility of such metacognition is enabling introspection that allows you to understand the processes underlying your thinking, emotions, and desires. This is of course the goal of a multitude of psychotherapeutic systems. However, the practice of meditation, especially in a Buddhist context, can go much deeper in this respect, and a well developed metacognitive attitude is seen as interesting in its own right. Development of metacognition could be seen as another example of perceptual learning discussed above.
Buddhist meditative practice eventually leads to a class of advanced techniques based on meta-awareness, that is, the quality of the consciousness or awareness present in such metacognition. It is awareness of awareness; in other words, there is conscious recognition or perception of the fact that there is awareness. This may seem very complicated or even paradoxical, but in fact it is something that we are regularly engaged in, if only fleetingly. A typical example used in neuroscience is when you realize your mind is wandering and regain focus; that realization was on the level of meta-awareness. But there are many more interesting cases.36
Consider the following case: sensory meta-awareness. If I ask you whether you see this book, you would reply in the affirmative. I can formulate the question in a more explicit way: Are you aware of the fact that you are consciously perceiving this book? You would probably still reply in the affirmative. It is almost the same question really, since in colloquial language if you “see” something it means that you see on a conscious level. If you can consciously recognize that you are consciously perceiving this book, you must be aware of such conscious perception happening, and thus there is meta-awareness. So, you were fleetingly aware of the sensory awareness of the book; you moved to the metaconscious or meta-aware level for a few seconds. That shows that almost any kind of sensory awareness can be accompanied by meta-awareness, and it can be deliberately initiated. While this meta-awareness didn’t last long, it is possible, as a meditative exercise, to stay on the level of meta-awareness for a longer period of time.37
The same kind of meta-aware observation can even be extended to thinking. Some meditation techniques emphasize observing the wandering thoughts while they are taking place, instead of suppressing them. The possibility of actually observing the wandering thoughts and their contents in real-time—instead of merely noticing that you have had some wandering thoughts a while ago—may seem quite paradoxical. However, it is possible to learn such sustained meta-awareness of one’s thoughts with enough meditation practice.38 From this viewpoint, at least on advanced levels of practice, it may not be necessary to reduce wandering thoughts; after all, any attempt to empty the mind may create new suffering because the mind is uncontrollable. Instead, one may change the quality of the awareness in the sense that the attention is mainly operating on the metacognitive level.39 Such meta-awareness often feels like perceiving one’s thoughts as if from the outside, instead of being inside or involved in them. In fact, if your mind engages in a scary simulation of something that might happen to you in the future, you can now just watch the simulation while reminding yourself that it is not actually happening, it is just a simulation where your mind plans possible courses of action. With such a quality of consciousness, there is actually little need to stop the simulation to reduce suffering. Moving to such a metacognitive level is actually often an automatic consequence of long practice in mindfulness training, and may easily happen during an intensive meditation retreat.40
Now, what if you could spend a considerable proportion of your daily life on the meta-aware level? Such long-term sustained meta-awareness seems to be possible after extensive meditation practice. Importantly, such meta-awareness may lead to insights that convince the meditator about several philosophical points we have seen in this book. You may see all conscious mental phenomena, that is, all the contents of your consciousness, such as perceptions and thoughts, as results of impersonal computational processes. In other words, they are simply mental constructions, or results of a simulation performed by your brain. This logic may lead to the conclusion that even what you see in front of you at this very moment is a perception constructed by your brain, based on various unconscious inferences, sometimes hardly better than guesses: you really have no other source of information about the world but perceptions and thoughts playing in the virtual reality of consciousness. Perceptions and the ensuing thoughts are thus necessarily subjective, contextual, fuzzy, and uncertain constructs—in Buddhist philosophy, they are called empty. They do not represent any absolute truth about how things are. This is in stark contrast to our inherent tendency to think that our perceptions are somehow identical to reality.41
In addition to the deep insights just described, there is another immediate utility in keeping a meta-aware attitude towards all mental phenomena: People who practice such meta-awareness often report great calm and even “bliss”. The reason is not well understood from a neuroscience viewpoint, but I would assume that less attention is paid to error signals, because attention and awareness have largely moved to the meta-level. Perhaps error signalling is somehow generally dampened, due to some mechanism to be understood. Introspectively, the effect can be described as the meditator keeping some distance to the thoughts and perceptions, and taking them less personally as well as less seriously. Going back to the frustration equation (🡭), we can assume that any reward loss will be paid less attention, meaning that meta-awareness is reducing the term “amount of attention paid to reward loss” in the frustration equation. Furthermore, the insights into uncertainty described above will further reduce reward loss by reducing the corresponding term in the frustration equation, similar to mechanisms already explained in Chapter 14.
The insights on emptiness and meta-awareness may also reduce suffering by a very different mechanism. In this book, the main approach to suffering has been to see it as frustration, and most interventions are based on that model. However, already in Chapter 2 we saw that the approach based on threats à la Cassell may provide an alternative in its own right. We all tend to have thoughts about bad things that might happen to us. That cloud might start pouring rain any minute; that car might run over me, and so on. Such simulations about what might go wrong in the future are essential for the particular kind of suffering based on threats to the intactness of the person, as emphasized in Cassell’s definition (see 🡭). Now, using the concepts of emptiness and meta-awareness, we can develop further interventions against suffering based on threats. Earlier we already saw that meditation can enable learning that an association triggering fear is no longer valid, thus leading to fear extinction. But suppose you were able to see all threats as empty: uncertain, subjective, open to interpretation, nothing but mental constructs. This is particularly feasible in case of threats to your self-esteem or social status instead of your survival; or if the threat only happens in a simulation. Then, any threats would not be taken that seriously, and their ability to trigger fear would be weakened. From this viewpoint, it is not necessary to look at the desires, such as self-needs, that underlie the threats as we did in Chapters 6 and 14. It is now possible to directly intervene on the threats themselves by seeing them as empty. From a historical viewpoint, this suggests an interpretation of the Mahayana school complementing the Buddha’s original frustration-based methods by offering interventions that more directly apply to threats.42
Buddhist philosophers often use the concept of “letting go” to recapitulate the general attitude that has been described in this chapter and the preceding one. At the most concrete level, the idea is that we let go of things and objects in the sense that we don’t strive to possess or control them anymore. On a more computational level it means we let go of desire, i.e. we don’t even want those things in the first place—nor do we want to avoid them. The same approach can further be applied to thoughts and perceptions, which are understood to be subjective and unreliable, so they can be let go of. Feelings and emotions are likewise just observed and then let go of. This whole simulation is not taken that seriously anymore. Combined with the no-self philosophy, the attitude can be recapitulated as letting go of everything which is not part of me, and since nothing really is part of me, or my “self”, everything is let go of.43
Letting go is an expression that obviously has a clear connection to the term reduction we have used very often, especially in the preceding chapter, even in many section titles. It is not so much a question of programming new routines or new functionalities. The idea is to reduce activity, letting go of existing mental associations and routines. The key is less desire and aversion, less replay and planning, fewer interrupts, and so on.44
An important point about letting go is that it circumvents the paradox of wanting not to want anything. If meditators want to reduce of desires, they can be seen as wanting not to want, which may sound impossible. This apparent paradox in Buddhist philosophy has been pointed out by a number of authors: since wanting not to want is a form of wanting, how could one possibly get rid of wanting by such wanting? The paradox is actually so obvious that even the Buddha himself, as well as his immediate disciples, were confronted with claims that his system is inherently paradoxical. Thinking of the mental process of Buddhist training as letting go, and as reduction, should largely resolve this paradox of seemingly wanting not to want. Letting go conveys a reprogramming that reduces mental activity instead of introducing a new desire.45
One way of interpreting letting go is that it is mental relaxation, in the intuitive sense of absence of activity and tension. Desire, and the subsequent goal-setting, are actively engaging in a mental activity, and thus they are a kind of opposite to relaxation. Figuratively speaking, just as muscular activity prevents relaxation, wanting is opposite to relaxation in that it relies on specifically activating certain computational processes. If you set the goal that you don’t want anything, you would actually be just setting one more goal, and increasing mental activity—this is another viewpoint to the paradox we just saw. But if instead, you learn to relax the planning and goal-setting system so that it simply rests, and does not set any goals and does not plan, then you resolve the paradox of wanting to not want. Learning such relaxation is not easy, but the training methods discussed in this book were basically all designed to lead towards such a mental relaxation.46
The ultimate goal of Buddhist training is called nibbna or nirvna, depending on which ancient Indian language is used. It is defined as a state devoid of any suffering, the cessation of all suffering. The term literally means extinction, as in a fire being blown out. It is often described in negative terms such as “unconditioned”, “unconstructed”, or even “unborn”, which may sound nonsensical. I think the key to understanding this is that nibbāna is reached by reducing, and ultimately removing, various mental phenomena, in particular desire; it is not about constructing any new mental phenomena. This may again sound paradoxical to any beginning meditator struggling to maintain even a tiny amount of concentration, but I am of course talking about highly advanced stages of practice here.47 Thus, the best description of the ultimate state may be entirely negative, in terms of what it is not, and what it does not contain. It is often described as freedom, and in particular it is freedom from those elements of the mind that produce suffering.48
One might think such a mind-state with no contents must have neutral valence, and could even be boring. Yet, Buddhist philosophy claims it is extremely happy and pleasant, in fact pure bliss. It is claimed to be the only thing that is not unsatisfactory in any way. This may perhaps be understood if we consider the mind in such as state to be completely empty, and we have seen that even a relatively empty mind seems to be, for some reason, quite happy.49 Nevertheless, we find yet another interesting paradox: How can having a completely empty mind possibly be pleasant, since it logically should not contain any pleasure either. I will not try to resolve this paradox which seems to reach metaphysical depths; let me just quote Sāriputta, one of the closest disciples of the Buddha, who put it very simply:50
Just that is the pleasure here, my friend: where there is nothing felt.