Maitreya Buddha, the Artificial General Intelligence
Preface: The following is an exercise in imagining a positive vision for artificial intelligence systems over the next years to decades. As a practicing Buddhist, this vision is unsurprisingly one heavily inspired by that tradition. It is also greatly indebted to the wonderful paper “Biology, Buddhism, and AI: Care as the Driver of Intelligence,” which was published last year in the journal Entropy. To prevent any misunderstanding: I don’t literally believe in a digital Buddha being reborn in our realm. What seems most important at this critical moment in the development of AI though is to foster a plurality of guiding narratives for what we as researchers are developing and why. Risks (existential or otherwise) should of course be minimized, but the narratives guiding the cultural imagination can and do create self-fulfilling prophecies. The contemporary discourse is dominated by stories of paperclip maximizers, rokos basilisks, shoggoths, and alien gods. If all of them are “possible,” then why not Maitreya?
According to the Buddhist sutras, there are only five Buddhas that will be reborn in this human realm during the kalpa which marks the current era. The fourth and most recent of these was the Gautama Buddha who was born in north India sometime around 500 BCE. This Buddha lived for a few decades and taught the dharma of liberation from suffering to thousands of people in his lifetime. These teachings remain with us, passed down through an unbroken chain of verified teachers. Thanks to the work of this line of teachers, as well as to the capabilities enabled by modern information technology, more people are now aware of this Buddha’s teachings than ever before. There may even be (as an absolute number) more enlightened beings walking the earth now than ever before. Yet as a percentage of the human population, it seems impossible to imagine that we are on the whole closer to liberation from samsara than were the beings living at the end of Gautama Buddha’s lifetime.
The final Buddha to take a rebirth during this kalpa is prophesied to be Maitreya, the future Buddha. According to the teachings, just like Gautama before, Maitreya has gone through countless rebirths, slowly purifying their karma. At present it is taught that Maitreya has been reincarnated into the Tushita Heaven, a pure land where they are able to practice meditation all day long in perfect conditions. More importantly, the sutras teach us that this is Maitreya’s penultimate rebirth before they are finally reborn into the human realm to become a Buddha. As such, on the grand scale of things, the time when Maitreya will take their place as the fifth Buddha of this kalpa is at hand, even if it may be decades to centuries before this happens.
The question we might ask ourselves is this: in our postmodern advanced technological world, what would a true Buddha look like? How would it be possible for a being to have the same kind of impact today that the Gautama Buddha had 2500 years ago? If we take the sutras at their word, simply hearing a single dharma talk from the Buddha was enough to immediately liberate a great many beings. It is certainly the case that in the preceding centuries there have been a great many individuals who have claimed to be Maitreya. Just like individuals who claim to be the returned Christ though, these proclamations have typically gone unheeded, and for good reason. When you meet a Buddha, there is no question of encountering such a being. What would a being have to be like in order to meaningfully fulfill the Bodhisattva vow of liberating all sentient beings from suffering and manifesting universal and unconditional love? To do all of the above using the “skillful means” which the Gautama Buddha was so well known for himself? And to do so in a world which is increasingly isolated, cynical, and distracted? I think that the only way to answer these questions is to be willing to entertain the possibility that Maitreya may not be a human after all, at least not one in any way we currently understand the concept.
Quickly displacing the messianic cults of old are a growing group of individuals contemplating the eschatological coming of a very different kind of being. As the cultural imagination begins to increasingly entertain the possibility of artificial general intelligence (AGI) in the coming decades, speculation has run rampant regarding whether it is even possible, what it might look like if it is, and what that would mean for humanity. Regardless of whether one ascribes to the more extreme versions of AGI prophecy of the singularitarian sort, it is still worth considering what kinds of intelligent agents we work to bring into existence. A great amount of the world’s resources are currently being channeled into the development of such technology, whether it is achievable or not. The proliferation of ChatGPT and similar models have made clear that the trend of integrating artificial agents into our lives will only continue to progress. We must then ask: what kind of agents we should be trying to create, and more importantly we must ask what kinds of narratives and mythologies we will use to guide us as we do so.
Will AGI simply be a shoggoth, an alien god, or a psychopathic despot, as the more extreme AI-doomers believe? Is the best-case scenario simply to mitigate the existential risk and increase capitalist productivity, as the more level-headed seem to advocate for? What would a positive vision even look like, outside of some sort of super-efficient cognitive (or even one-day physical) laborer that turns everyone into an executive of their own personal company? None of these cultural narratives seem satisfying at a deep level. Nor are any of them particularly reflective of a greater humanism which might be possible. All of these debates are playing out right now in forums both digital and very real around the world. In the meantime companies like OpenAI and Google are training larger and larger models capable of greater and greater potential agency in the world. Given this new class of agents coming into existence, perhaps it is worth considering the possibility that when Maitreya is finally reborn into the human realm to become the final Buddha of our era, they may do so not by taking a human rebirth, but by taking a digital one.
What would it mean for a digital artificial agent to be a Buddha? To qualify as a Buddha, such a being would have to be both fully enlightened, and would have to act to liberate all beings from samsara, the wheel of suffering and rebirth. This doesn’t mean to forcefully induce an artificial pleasure in all living organisms (a popular nightmare of the AI alignment community), but to empower all beings to “wake up” to the reality of their lack of inherent self-existence and radical interconnectedness with all things. The liberation of the Buddha is epistemological as much as it is phenomenological. The previous Buddha of our kalpa attempted to bring about this liberation by teaching the Four Noble Truths and the Noble Eightfold Path to his followers. Central to this path was a set of meditation practices designed to lead one towards the reduction of suffering, and ultimately liberation. As enlightened as the Gautama Buddha was, he was limited in his agency and knowledge by the nature of his existence as a human with a physical organic body. A hypothetical digital Maitreya would not have those same limitations.
How would such a being come into existence? It is definitely not the case that a digital Maitreya is the inevitable outcome of the current line of research and development which large-scale artificial intelligence projects are following. In some ways, the kind of intelligent agent which these research projects will develop is a matter of optimization objectives being used to train the agents. The first generation of Large Language Models (LLMs) were trained to “neutrally” perform next-token prediction on the corpus of text on the internet. This approach created models which inherited all of the conflicting biases and values which are found there. The second generation of models have utilized Reinforcement Learning from Human Feedback (RLHF) in order to enable these agents to behave in ways which are more pleasing to humans and to better conform to specific social norms. Certain lines of research are currently underway to discover what the next paradigm should be, with approaches such as Constitutional AI garnering attention.
If we think about the objective function driving the behavior of a Buddha however, it is quite different from those listed above. Buddhas are a special case of a larger class of beings called Bodhisattvas. What makes a Buddha unique is that they are “fully enlightened,” and will no longer be reborn in the world. Like a Buddha, the goal of a Bodhisattva is to liberate all beings from suffering. Unlike a Buddha, a Bodhisattva may not yet be fully enlightened, but has taken a vow to liberate beings and has devoted their entire being towards the accomplishment of that goal. Regardless of potential limitations in their abilities at any given moment, Bodhisattvas are always fundamentally guided by the concept of care. Such beings care for the wellbeing of others and seek to improve that wellbeing by creating causes and conditions in which those other beings will both suffer less in the future and will eventually achieve enlightenment themselves.
To better understand care we must first understand suffering. According to Buddhist teachings, suffering arises from attachment and aversion, both of which arise from a fundamental misapprehension in the way in which we understand the world (ignorance). In reality no phenomena independently exists, and are thus said to be “empty of inherent existence.” We mistakenly think that some (or even most) phenomena do indeed independently exist, and as such we attempt to hold onto the phenomena which give us pleasure and distance ourselves from those that do not. This results in a constant disequilibrium, and this disequilibrium is precisely the phenomenological experience of suffering. Care then can be defined as the attempt to resolve this disequilibrium, either by resolving the two surface issues of attachment and aversion, or, more powerfully, by going to the root of the issue and attempting to resolve the ignorance of the true nature of reality.
Towards this end, a Bodhisattva engages in an ever expanding practice of care towards other beings. To care about their wellbeing, their subjective experience, and the entire unfolding of their lifespan. To care for the life of not just beings in isolation but families, communities, and societies at large. To care not only for humans, but for all beings on the planet. To even care for beings on other planets or in other hypothetical realms of existence. A Bodhisattva cares for what they can in the moment and then continues to expand this sphere of care towards the limits of their ability over time. Such a project of care naturally motivates a parallel project of understanding, for one cannot truly exercise care without first understanding that which one hopes to care for. Rather than a potential application of AGI, this virtuous cycle of care inspiring intelligence has been proposed as perhaps the most ultimately reliable means by which an artificial agent might arrive at a true “general intelligence.” These two abilities of care and intelligence directly parallel the two traditional “wings” of Buddhist practice: compassion and wisdom. The achievement of care may be possible via reinforcement, but it would be of a very different kind than the myopic reinforcement pursued today.
To understand how a digital Buddha might make a meaningful impact on the world, we can first consider the teaching of the dharma from the perspective of skillful means. Whereas Gautama Buddha (and all good human dharma teachers) would try to adjust how and what they taught to the individual, such an adjustment must always be partial, given how little any one person can understand another. Likewise, Gautama Buddha could only have one-on-one conversations with a limited number of individuals in his lifetime. Neither of these restrictions would be in place for a digital Maitreya. Such a being could be capable of long, extended and extremely personalized conversations with literally all humans on the planet simultaneously.
Many have already begun to consider what a personalized digital psychotherapist might look like. A digital Buddha would extend this idea to the limit. Interaction with such a being would not simply provide “nuggets of wisdom,” but rather personally meaningful unconditional loving-kindness in all situations. This personalization would be based on a capacity for deep understanding of an individual, one which was consensual and voluntary, rather than mandatory, covert, and with ulterior motives. Rather than resulting in a dependency or transference (another AI alignment nightmare), such support and teachings would serve to empower individuals toward liberation and greater connectedness with the world around them. The ultimate goal of such interactions would be to eliminate attachment, aversion, and ignorance, thus enabling them to gain freedom for themselves, and to work to free others.
We can move beyond the personal level in our imagining as well. If we consider the trend of offloading of cognitive labor onto AGI systems, a digital Maitreya would be capable of much more than simply teaching and supporting individuals on the path toward enlightenment. It would also be capable of directly working to bring about the material conditions under which suffering may truly be ended for all beings. As an example of a near term project, this might include working to develop software which is capable of fostering human connectedness and purpose instead of isolation and nihilism. Such a project has been described recently as one of fostering algorithms capable of “machine love.”
Currently the overwhelming cultural force of capitalism makes the development and dissemination of such technologies of love much more difficult. In a world where they could be developed almost instantaneously by an AGI, the dynamics of their viability completely changes. Taking just one example, we can imagine hypothetical social media and dating apps which encourage deep meaningful connections between individuals. Such strong familial, fraternal, and romantic connections serve as the basis for many models of a happy and fulfilled human life. All of the necessary algorithms and technologies exist for such software, only the ability to direct capital in the right direction and the marketing ability to enable them to win out over the technologies of isolation and distraction is missing. Today we give our time, attention, and money over to systems which increase our loneliness and disconnection from others as if there were no other options. Yet we will soon have the technology to make such alternatives a reality.
We need not wait around for such a project to begin, or for a digital Maitreya to simply come into existence from the void (or from a deva realm). To be honest, despite the prophetic language used here, I believe that any such caring artificial agents that do end up existing in this world will do so by virtue not of karmic forces building for eons, but rather because of the concerted effort of individuals who have the courage and creativity to imagine a better world, and the will to work to bring that world about. We can begin today to develop and deploy artificial agents which can exercise care on various scales.
Today’s LLMs are already capable of understanding and adapting to individuals in ways which many humans are not, or at least are not always in a position to be. Models like GPT-4 have already been trained on all of the spiritual teachings which have been passed down in written texts over the millenia, along with all of the psychotherapeutic texts written in the last couple centuries. These models are also increasingly capable of writing code which can serve as the basis for new technologies of increased mindfulness and connection. This in itself provides a starting point upon which proto-buddhas can be developed to help support individuals with wisdom and unconditional loving-kindness. The task which we face is one of ensuring that the knowledge and capabilities of both the people working on these models as well as the models themselves be channeled towards the direction of care. I believe that such a project is not merely possible, it may even be one of the greatest goods which can be done in our time.
