Biology, Buddhism, and AI: Care as the Driver of Intelligence

The past decades saw a development of interdisciplinary interest in synthetic modes of knowledge that integrate Buddhist philosophy with contemporary Western philosophy of mind and cognitive science [91]. In fact, the developing domain of contemplative science today has made attempts at developing novel empirical and theoretical language and tools in order to better understand plasticity of mind and brain [92] and to integrate contemplative techniques (such as meditation) into the models of consciousness and cognition [93].

We propose that a number of core Buddhist concepts (in particular the constructed nature of selfhood/no-self, duḥkha, universal loving care, and the Bodhisattva) can be profitably used to challenge and to enrich the current work in diverse intelligences, including novel approaches to AI and to biology. In contemporary cognitive science, the mind is sometimes understood as enactive [94,95,96], in that the boundaries between cognition, a form of embodiment, and a being’s relationship with the world are understood as pragmatic and non-reducible [97,98]. Similarly, contemporary proposals for the constructed, transient and contextual nature of human selfhood [99] and for its inextricably social/cultural orientation [100] are delineating a direction for understanding forms of life and intelligences that are relational, transient and have malleable boundaries (here understood as cognitive light cones). These fundamental shifts in perspective are particularly suitable for integrating Buddhist and enactive approaches to AI [101]. In the following sections, we will present core Buddhist concepts of selfhood and the Bodhisattva idea to further develop our proposal that intelligence can be understood in terms of Care and the remedying of stress. Our discussion can, in this regard, be seen as resonant with the enactivist tradition, which describes selves as precarious centers of concern [102], as patterned variations [103] of different forms of experienced selfhood, ranging from the notions of minimal self [104] to embodied, affective [98,105], and socially extended/participatory [106,107,108] forms of situated selfhood. What is common in many flavors of enactive and situated cognition models is the emphasis on the dynamic and creative nature of what it means to have or to be a form of self.

Importantly, the recent work on participatory and social forms of selfhood, sometimes referred to as the “interactive turn” [109] in enactive cognitive science, stresses the situated nature of cognition and subjectivity within the social worlds and thus invites us to think specifically about ethical or prosocial forms of constitution of self. While our work is inspired by and is respectful of the enactivist tradition, in this paper we are exploring the translational and generative potential of thinking through the idea of Intelligence as Care in resonance with current concepts in biology, the Bodhisattva vow, and developments in AI. Our proposal that Selves be understood as processual cognitive systems and models thereof bears a natural kinship with enactivist thought as developed and evolved over the past three decades [97,98,110,111], and can thus also be discussed and critiqued in the light of phenomenological notions of self-awareness, minimal self, etc. [112,113,114,115,116]. We acknowledge the resonance with contemporary approaches that see the self as enactive, embodied, extended, affective, situated, and interactive, and we hope our work can anchor future conversations with these rich traditions.

According to traditional Buddhist scriptural analyses, the recognition that there is no singular and enduring individual that must survive and prevail serves to undermine self-seeking action at the expense of others and their environment. Therefore, the evolving of intelligence that is aware of no-self—or if we want, intelligence that is no-self-aware—is also held to be intrinsically wholesome and associated with concern for the happiness and well-being of others. This claim—that simply understanding the irreality of enduring, singular agents can be a catalyst for ethically informed intelligence—is especially noticeable in Great Vehicle (Skt. Mahāyāna) currents of Buddhist view and practice that develop the idea of the Bodhisattva [26,27,117]. Traditionally conceived as an agent that has accepted responsibility for the flourishing of all sentient beings throughout time and space, a Bodhisattva pursues the attainment of “omniscience” (defined as unmediated and simultaneous knowledge of all things throughout space and time, both as they are and as they may seem to cognizing subjects) in order to protect and provide for all beings, and to assist them in achieving their own ultimate potential [118,119]. In this way, understanding of the drive of a Bodhisattva is two-fold: as affectionate care (concerned with sentient beings) and as insight into things as they are and as they appear (associated with the pursuit of omniscience). These two drives, care and insight, are seen as standing in a dynamic relationship and are not separate in essence. Hence, as a model of intelligence, the Bodhisattva principle may be subsumed under the slogan, “intelligence as care”.

Such emphasis on care contrasts to some extent with an understanding of intelligence as, first and foremost, the ability to control causal chains that lead to the achievement of predefined goals. Rather, the “intelligence as care” proposal can be seen as highlighting an element of spontaneity that emerges in the light of the co-constitution of object, agent, and action. According to general Buddhist analysis, the individual that may be assumed to exist as a singular, enduring, and controlling self is mere appearance devoid of causal efficacy, and thus epiphenomenal [68]. In the case of a Bodhisattva, this understanding is carried forward so as to encompass a critique of the apparent foundations of cognition: object, agent, and action. Since the constituents of this triad can neither be determined as distinct from one another, nor identical, both the epistemic and the ontological status of all versions of this triple framework are seen to be on the same level as dream images, mirages, and other such traditional examples of illusion. Thus, recognizing the interdependent and constructed nature of object, agent, and action means recognizing their “emptiness” (Skt. śūnyatā), and since “recognized,” “recognizer,” and “act of recognition” are also within the purview of comprehensive deconstruction, the “recognition” in question is often qualified as being of a nondual nature.

According to the Bodhisattva model of intelligence, such deconstruction of the apparent foundations of cognition elicits a transformation of both the scope and acuity of the cognitive system that performs it. As caring intelligences, Bodhisattvas are potentially described as embodying an enlarged field of affordances achieved through the understanding that selves are illusory. Thus, subsequent to the deconstructive insight proper, the appearances of a world inhabited by selves resume. However, at this point the Bodhisattva cognitive system is no longer constrained by the perception that one single self—i.e., its own self—requires special and sustained attention. Instead, Bodhisattva cognitive processes are now said to engage with spontaneous care for all apparent individuals. Thus, an immediate takeaway from non-dual insight is said to be the perception that oneself and all others are ultimately of the same identity.

In the traditional literature, beings that are not deeply informed by insight into no-self are referred to as “ordinary,” and such non-bodhisattva cognitive systems thus buy into the apparent yet epiphenomenal nature of subjective identity. From the perspective of a mind that in this way reifies personal selfhood, the very sense of being a subject of experience and a controlling agent of actions naturally and unquestionably implies that one is thus also someone who should receive special care and deserves to flourish far beyond the status quo. However, for a Bodhisattva who emerges from the understanding that object, agent, and action are interdependent constructs, this perception of a worthy and deserving self is now said to accompany the perception of any and all sentient individuals—with the same force and naturalness that were previously reserved for the perception of one’s own self. All are in this way seen as equally attractive and naturally worthy of care.

Based on the above contours of the Bodhisattva idea, we shall next seek to analyze this concept of intelligent life with the help of the cognitive light cone framework described earlier (Figure 2 and Figure 3). Let us again note that this model of cognitive life avoids appeal to the notion of a singular, enduring, and controlling self—in other words, the very self that Buddhist perspectives generally agree is unfindable [68]. Instead, according to the light cone model, selves are conceived as evolving s systems and the functional structure of such systems corresponds with the scale of their goals; as such, they shrink, grow, and change over time, as the scope of the system’s goals changes. Such correspondence between goals and agent structure aligns well with a Buddhist understanding of intention as a type of karman, or “action,” understood as a world-constructing force [67,120]. As we have also noted earlier, cognitive systems emerge according to this formalism from a hypothesized drive to reduce stress—the difference between current conditions and life-optimal conditions. This definition of the fundamental drive that sustains cognitive evolution in turn resonates with the description of the Bodhisattva’s care and insight drives [26,27]. As we noted above, “stress”, understood as the tension between the present and the optimal state, is an apt translation of the Sanskrit duḥkha (otherwise often rendered as “suffering”), which a Bodhisattva’s care seeks to alleviate. Thus, operating through its two-fold (viz. care and insight) drive to reduce stress, the Bodhisattva cognitive system can conveniently be conceived as evolving naturally, driven by basically the same concerns as other forms of life.

Furthermore, as a Bodhisattva system evolves and develops agility, its capacity for networking and integration is described as becoming radically enhanced and increasingly spontaneous. This capacity for integration into increasingly comprehensive intelligent systems is described as a hallmark of the cognitive processes that emerge on the “Bodhisattva path” [26,27].

In this way, salient features of light cone formalism align well with traditional features ascribed to Bodhisattva cognition, so an attempt at delineating the latter in terms of the former seems both possible and potentially illuminating. How, then, would the cognitive light cone of a Bodhisattva system have to be drawn? In order to address this question, it is helpful to begin with the beginning of the Bodhisattva path, which traditionally is formulated in terms of a comprehensive commitment. Motivated by loving care for all sentient beings, an emerging Bodhisattva vows to achieve omniscience in order to help and provide for all the infinitely many forms and instances of sentient life. When for the first time this “mind of awakening” (Skt. bodhicitta) is brought forth with a commitment toward its continuous cultivation and ultimate fulfillment [118,121], then that constitutes the first step on the Bodhisattva path. Based on this commitment, the newly emergent Bodhisattva will also gradually cultivate the deconstructive insight that, as mentioned above, is ascribed a capacity to fundamentally transform the perceptions and abilities of its cognitive agent [26,27,117].

If we seek to analyze the consequences of adopting this Bodhisattva commitment according to the light-cone model of intelligent life, we will notice that the respective structures prior and subsequent to the Bodhisattva pledge will have to be drawn in radically different ways. If, for example, we imagine a human being about to adopt the Bodhisattva perspective, their sphere of explicit concern before making this defining commitment may be more or less expansive by human standards. Indeed, we might imagine that they have explored distant galaxies with a view to future space travel, many generations from now, or that they have developed programs with intended applicability across our globe’s ecosystems. Yet even if their spheres of measurement and activity up to this point have in this way been extraordinarily vast, we will still notice a striking difference in the functional boundary that emerges subsequent to their adoption of the vow: a difference that ensues simply from the way that the vow is formulated. By definition, the Bodhisattva perspective entails active concern across an infinite range, both spatially and in time. In short, wherever the endless myriads of beings may find themselves, the Bodhisattva forms the intention to go there and effect positive change. Implicitly, this programmatic intention thus also contains an open-ended pledge to comprehend the past, because the ability to skillfully influence events in the present and future can be seen to involve knowledge of past states of affairs. Thus, by simply committing to the Bodhisattva stance and practices, the sphere of measurement and activity of the cognitive system that makes the commitment has gone from finite to infinite, and so the cone structure that otherwise is applicable to all forms of cognizant life has in this sense been transcended.

Seeking to indicate infinity on both the spatial and temporal axes, we might now instead see the Bodhisattva cognitive system’s computational surface represented by an all-encompassing sphere that accommodates all instances of life within it. On the other hand, a human cognitive system that adopts the Bodhisattva pledge—perhaps through a formalized ritual—is obviously not likely to notice any erasure or infinite extension of its functional boundaries simply due to participation in the vow-bestowing ceremony. Indeed, the dynamics of such a cognitive system will in many ways unfold just as before, based on characteristically human circumstances and conditions. Nonetheless, if the expressed contents of the vow have been registered, the system’s prior functional boundary will now have become open-ended in space and time. Taking into account both the largely continuous character of the human cognitive system and the all-encompassing character of the Bodhisattva sphere of engagement, we may then represent the cognitive system of a newly emergent Bodhisattva in human form by placing a physical light cone (PLC) within an all-encompassing Care light cone (CLC) sphere of infinite commitment (Figure 4). We note that the limited PLC and the infinite CLC are equally required for a representation of our “newborn” Bodhisattva: if the CLC were not coextensive with the field of possibilities we would simply be looking at an instance of human cognition, and without the confining PLC structure we would be concerned with pure perfection, transcendent of context. CLC volumes may come with more complex shapes than the diamond or “spinning top” presented above. Figure 5 illustrates more nuanced cone profiles by showing three arbitrary CLC archetypes.

Each profile presents a snapshot of the probability distribution of Care for an agent with a distinct stereotypical profile–“the futurist”, “the historian”, and “the chauvinist”–depicting how the CLC may take various shapes and densities, and be composed of complex, disconnected point clouds.

While, as noted earlier, the Bodhisattva concept lends itself well to representation according to light cone formalism, we can now further conclude that if we go by that framework, the features of a Bodhisattva system stand out as dramatically distinct from those of other known or putative forms of life. Such dramatic differences would perhaps be expected if we had been seeking to represent the features of highly evolved Bodhisattva states and their associated features of super intelligence, as they appear in traditional accounts. Yet what we just drew represents the cognitive system of a Bodhisattva on the most basic, entry level of the path: a system that for the first time adopts the so-called Bodhisattva vow. As we recall, this vow is formulated as a commitment to cultivating “knowledge of all things” in order to help and provide for all beings.

The fact that tradition speaks of this initial, explicit expression of universal care as “the root of omniscience” is noteworthy, and we may ask ourselves why pledging to be, as it were, infinitely good should be of such pivotal epistemological consequence. Yet if we accept the framework of the cognitive light cone as our model for the emergence and integration of intelligences, it is clear that the formulation of infinite concern and responsibility establishes an immediate link to open-ended, and thus infinite, intelligence. In other words, commitment to the Bodhisattva codex formally renders care limitless in time and space, and according to the light cone schema, care translates into intelligence.

Now, can it similarly be said that the pursuit of intelligence, or knowledge, entails an expansion of care? In assessing this question, we may again employ the same model, this time to an agent that formulates a commitment purely in terms of pledging to know “everything there is to know throughout time and space.” The functional structure of this agent will, it seems, be less impressive than that of a classic Bodhisattva agent where the primary impulse is care. The light cone demarcates the spatiotemporal sphere that an agent “can measure and attempt to modify”. However, compared to care, which drives both measurement and modification, the pursuit of knowledge alone lacks any explicit drive toward modification. This might, in turn, suggest that a quest for “knowledge for its own sake” becomes burdened by a potentially self-defeating passivity. Instead, we might further hypothesize that care is required to engender the dynamics that enable truly revolutionary cognitive developments, such as those leading to superintelligence, or to artificial general intelligence.

To further accentuate the features of the type of intelligence that is associated with the Bodhisattva vow, we may here for comparison consider the frame of mind of someone who vows “I shall subjugate everyone in time and space for my own pleasure”. At first glance, we may in these two cases notice a similar sense of universality and infinity, but a closer look reveals marked differences. In the case of the vow to subjugate for personal enjoyment, the universal commitment is directed toward the fulfilment of the agent’s individual version of what should/must be the case. Personal needs are intrinsically limited. Even though greed may feel infinite, once we begin to specify what we want, our needs become rather limited and predictable, because they largely correspond with our understanding of who and what we are. For example, despite all our arguments and seeming differences, in the end we humans all have rather similar desires with respect to pleasure, wealth, health, etc. So despite the apparent grand scale (wanting to embrace all of time and space), the drive toward the fulfilment of “all my personal wishes” becomes quite trivial—just like that of any other Self/sentient being—when compared to the care drive of a Bodhisattva. The scope of the Bodhisattva’s sphere of measurement and modification is not just seemingly infinite, but actually so, because a Bodhisattva’s scope and mode of engagement are not defined by the intrinsically limiting frame of one individual mind. Instead, it is shaped and driven by the infinity of living beings, constituting infinitely diverse instances of needs and desires in time and space. The Bodhisattva promises to know all of those needs, and respond creatively and benevolently to them.

In comparison, even if we think, “I shall turn all universes into nothing so that I may be happy!” or “I shall make everything in time and space my personal property!,” the span and intensity of the cognitive sphere that is instantiated and sought to be measured and modified by such wishes seem very narrow and dull when compared to the Bodhisattva’s wholly other-directed drive (“Whatever may make them flourish, let them have it!). The paths and end states of the wish for universal destruction or universal possession are easy to conceive of when compared to the Bodhisattva’s endless path of endless discoveries.

Similarly, what might we learn by juxtaposing the Bodhisattva ideal with the mind of Māra, who at times is referred to as “the Buddhist devil”? In Buddhist scripture, Māra at times occurs as an evil deity (thus famously trying to subdue and distract Siddhartha on the eve of his ultimate awakening) and at others more like a universal principle of evil and deception (as in the framework of “the four māras”) that stands opposed to awakening [122]. The māra principle can be seen to have analogs in physics, which suggests that this is a concept that operates “all the way down” as limitations on the ability of agents to know, predict, and engineer. “Four māras” could, for example, be enumerated as entropy, inertia, Gödel uncertainty, and finite speed of light (viz. the limitations of special relativity). These work hard to keep us all down, and each is a background “force” that opposes all efforts to do good things. Movement, complexification, etc., are all resisted all the time, appearing as an impersonal feature of the universe that is constantly trying to undermine whatever one may want to do. Just as particles have nano-goals—the light beam wants to get to its target by the path that uses up the least energy and so is subject to the influence of “Mara,” in the sense that by moving it will inevitably use up some energy—their very goal-seeking (variational principle) property is the effort of resisting the Mara-like influence.

As a quasi-personal, pervasive principle, Māra has intentions that can be understood as other-directed in a way that is comparable to the Bodhisattva. Māra does not really have any wish “for himself”. Instead, his drive is motivated by a negative formulation of the Bodhisattva pledge: “Wherever there are beings, let me prevent their awakening!” This other-directed sensitivity seems to deliver a light cone that surpasses those associated with the more traditional “selfish” intentions we considered in the paragraph above. As with the Bodhisattva, there is in Māra’s case an open-ended universal drive to both determine and influence each and every actual state of sentient beings throughout space and time. So, rather than yet another subjective mind with personal desires, Māra’s state appears more like pure evil intelligence, beyond personhood. If the Māra drive is in that way pure and all-encompassing evil, the Bodhisattva state is then universal benevolent engagement. How to compare such a pair of intelligences, both other-dependent and other-directed rather than “selfish” in the usual sense? Is one more powerful than the other, or do they scale up the same way in terms of the light cone model? Let us at this point simply note that the Māra drive seems reducible to a wish to maintain the status quo (“sentient beings suffer, and they shall keep doing so!”) whereas the Bodhisattva is committed to infinite transformation. If that is correct, the intelligence of the Bodhisattva’s care should again display decidedly superior features according to the light cone model, because a static wish to maintain what is—even if it is on a universal scale—entails far less measurement and modification than an open-ended pursuit of transformation wherever its potential is encountered.

Presumably, the circular process created by (a) an expanding light cone, (b) increasingly sophisticated intelligence, and hence (c) increasingly demanding challenges can either break down (in regression) or otherwise keep evolving, endlessly. In other words, since a growing light cone and superior intelligence entail an increase in problems that require solution, the expansion of intelligence can in itself never deliver any lasting peace or accomplishment. This seems to fly in the face of standard assumptions with respect to the nature and efficacy of intelligence. Is not the general expectation that the more intelligent we and our environment become, the less stress we will have to encounter? Yet if the above is correct, such an expectation turns out to be rather unrealistic.

The Bodhisattva vow can then be seen as a way of acknowledging, or even welcoming, this forbidding lay of the land (i.e., that there is no end to challenges, no matter how smart we might become) because one gladly (the wow has to be made joyfully…) accepts an endless project of infinite challenge. Traditionally, the ability to do so is associated with recognizing the facts of “no self” as discussed in the opening of this section. Accepting the Bodhisattva vow brings in this way the possibility of expanding intelligence in a steady fashion—free from hesitation, disappointment, fear, and other such factors that can now be seen to arise from misperceptions of the nature of the project. Instead, the vow invites an all-round cultivation of intelligence that does not assume any end state.

Before we begin to draw up specific implications for the way we may want to conceive of and develop AI, let us in concluding this section then note that the Bodhisattva vow can be seen as a method for control that is in alignment with, and informed by, the understanding that singular and enduring control agents do not actually exist. To see that, it is useful to consider what it might be like to have the freedom to control what thought one had next. Would not perfect control of one’s mind imply that one knew exactly what one was going to think, and then subsequently thought it? In that case, whenever a new thought arose, we would, absurdly, be rethinking what we had thought already, or otherwise there would, just as absurdly, have to be an infinite line of prior control modules in place for a single controlled thought to occur. Such consequences suggest that the concept of individual mind control is incoherent. “In control of my mind” (a necessary aspect of the common notion of free will) is logically impossible on the short time scale, but may be coherent on a very long time scale (“I’ve undertaken practices to eventually change the statistical distribution of the kinds of thoughts I will have in the future”). This in turn underscores the importance of long-term strategies, such as a vow to expand cognition.

The points discussed in the previous sections find a strong echo in the field of AI. Below, we attempt to bring concepts from both biology and Buddhism together into the language of AI, and suggest practical ways in which care may enrich each field. This section examines how to frame the complexity of intelligent agents, in their diversity and substrate-dependency. We address how agents may accomplish goals they care about by offloading stress and introducing care, with an illustrative example in the game of chess.

Defining intelligence is difficult [28]. For the purposes of this paper, and for the reasons stated in Section 1, the starting point of our working definition of intelligence was the ability to solve problems in some defined space. This definition is necessarily subjective (because it requires the observer to pick a space, and be intelligent enough to detect problem-solving behavior in it), and it is relative—in the eye of the beholder. Problem-solving requires motivation to progress through homeostatic loops of measurement and action (which requires energy expenditure) and is driven by stress loops (system-level propagation of the mismatch between present and optimal conditions). Thus, advanced intelligence is scaled up by increasing the scope of the states that trigger this stress (imperative to act)—it is “the capacity for identifying stress and working toward stress relief”. Further, as presented in parts 1 and 2 of the present paper, we propose, based on knowledge derived from recent advances in understanding the variety of cognitive living systems and from Buddhist ideals of the Bodhisattva path, that understanding of intelligence can be refined and expanded to understand intelligence as care which is scalable and appropriate for a range of cognitive phenomena, including artificial cognitive systems

When seeking to navigate across diverse definitions of intelligence, an easy rule of thumb is complexity. According to any definition, including the homeostatic one above, an intelligent system is expected to offer a simpler solution to a problem than a non-intelligent one. There are multiple measures of complexity [123,124,125] but this latter point should be valid for each such measure. This may mean, in case we consider computational complexity [126], that the entity uses a lesser amount of resources than another entity. In the case of statistical complexity [125], one can find an epsilon machine with fewer nodes to solve the problem.

Once one has reached a tentative framing for intelligence, it is necessary to characterize to some extent its ontological space and structure [127]. The space displays properties of a complex, diverse, and substrate-dependent system. Agents may show their capacity to drive themselves to identify and minimize stress, in a radically different manner based on the environment to which they are subject, so that ordering or classifying them may end up an extremely difficult task. We recognize that any attempt at defining intelligence will end up simplifying, due to the complexity and the loaded character of the term in the literature, especially among different fields and angles from which scholars have tried to frame it. It should be clear to all that the nature of intelligence is extremely diverse. Echoing the discussion in the biology section, we may underline that diversity strongly holds also for artificial systems, and for combinations of artificial and natural systems too. Each intelligence in simulations and AI may solve in a vastly different way the problem of its own existence, and so reduce the stressful gap between what is current and what would be optimal.

Oftentimes, artificial systems are not autonomous of course, and rely on humans maintaining them in order to keep existing. In such cases, one may want to consider the whole system including the maintainer instead of the AI alone. However, in either case, there may be many ways for a given system to implement a solution. For example, a Von Neumann architecture given the task of estimating the value of pi may use two completely different methods: one may be the Monte Carlo method, which generates a large number of random points within a square and counts how many fall in an enclosed circle [128]), or a fast Fourier-Transform-based method, which uses Brent binary splitting together with an efficient cache-handling Hermitian FFT to multiply big integers [129]. Both methods lead to the same result, with slightly different amounts of resources, including a different use of memory and computational power. However, the memory and computational power will differ a lot based on the environment. The same way a human does well on Earth but will struggle in Mars conditions, any artificial agent will behave very differently in a different physical substrate, which may come with a different computational paradigm. Similarly, with quantum computers, some algorithms are asymptotically faster than the fastest possible classical algorithms [130].

Above, we introduced the notion of stress, understood as an energy function or the distance between a given state and the optimal state, so that reducing stress may act as a driver for a given system to navigate the problem space. A similar situation can be found in AI systems, where optimizing the value of a single parameter is enough to drive a whole algorithm forward in its evolution, whether it is a simulation, an optimization algorithm, or any other learning system in general. Such a parameter is often designated as a reward function, but more generally characterizes a drive for the system to evolve a certain—often intended—direction. For example, in supervised learning, a so-called loss function calculates the distance between the current output of an algorithm and the expected output, and is used to evaluate how accurately an algorithm models a dataset. A drop in the loss function indicates a higher quality prediction, whereas a high value for the loss indicates that the prediction is completely off.

In the case of tasks with complex control sequences or exploration-exploitation tradeoffs [131], another AI technique called reinforcement learning is often used. A reinforcement learning agent is able to perceive and interpret its environment, take actions, and learn through trial and error. The learning takes place as the agent acts, while maximizing as much cumulative reward as possible, which it gains through adopting desired behaviors and minimizing punishment from undesired ones. This notion of reward ultimately drives the behavior of this type of learning system, in which negative reward would be equivalent to the same concept of stress as we mentioned before. These two examples of AI techniques display a strong analogy with the notion of stress introduced earlier in the context of biological systems. Of course, in the case of humans utilizing AI systems, the stress is transferred between the human and the AI. There is a conversion of the stress from human instructions to the AI, so that the AI is driven in the right direction for the final output to be of best use to the handling human. Vice versa, more and more AI systems are designed to express rich feedback to humans so that they may recognize errors or hurdles that may require a change in the way humans run the algorithm.