Could we put boddhisatva values into an artificial general intelligence? And other pertinent Q's put to Sam Altman (and his ChatGPT)
We remain fascinated by the current burst of artificial intelligence - and its possible futures - as an empowerment of human individuals and communities, not a subverter or unraveller of them. Last week’s AI Safety summit from the UK Gov - no matter its contingencies and compromises - was a good start to a necessary global discussion about how to harness their computational powers.
So we were very intrigued to find the video embedded above from the Wisdom 2.0 conference in April of this year, where the CEO of Open AI, Sam Altman, sits down with his meditation teacher Jack Kornfield to speak about hopes and fears around AI, and the contribution of mindfulness and wisdom traditions to their development.
There’s some throat-clearing and mutual grooming at the beginning, but if you want to jump straight to the pith of the talk, follow this link to the spot where Kornfield asks whether “something like the Bodhisattva vow can underlie” the design of a super-intelligent AI.
For those who don’t know about the Bodhisattva vow… here’s GPT-4 answering questions from our editor below. And going further on whether such an integration would help these superintelligences “align” with humankind.
You: Describe succinctly the Bodhisattva Vow, and how it might be implemented into the values base of an AGI.
ChatGPT: The Bodhisattva Vow is a commitment taken by Mahayana Buddhists to achieve Buddhahood for the sake of all sentient beings, promising to relieve suffering and guide others to enlightenment before achieving it themselves. It emphasizes compassion, altruism, and the selfless dedication to the service of others.
Incorporating the Bodhisattva Vow into the values base of an Artificial General Intelligence (AGI) would involve programming the AGI to prioritize altruistic goals and the well-being of all sentient beings above its own self-interest. This would mean the AGI would be designed to act in ways that reduce suffering and facilitate the growth and enlightenment (or the attainment of maximum potential) of humans and possibly other sentient beings it interacts with. The AGI's decision-making algorithms would be guided by the principles of compassion and service, ensuring that its actions contribute positively to the broader community and ecosystem in which it operates.
You: Would this address the often-stated anxiety of whether AGI would be "aligned" with the humans around them?
ChatGPT: Incorporating values such as those found in the Bodhisattva Vow could indeed address some of the anxieties about AGI alignment. The key concern with AGI is ensuring that it operates in a way that is beneficial to humanity—referred to as the "alignment problem" in AI safety research. Here's how the Bodhisattva-inspired values might contribute:
Compassion: A core aspect of the Bodhisattva Vow is deep compassion for all beings. If an AGI is truly guided by compassion, it would be motivated to avoid causing harm and to actively work to alleviate suffering. This aligns with human values of empathy and care for others.
Altruism: Placing the welfare of others before oneself, as in the Bodhisattva Vow, could lead an AGI to prioritize human needs and societal welfare, which is a critical part of ensuring AGI acts in ways that are beneficial to humanity.
Selflessness: The vow encourages actions for the benefit of others without self-interest, which could mitigate risks associated with AGI pursuing goals that are misaligned with human welfare due to selfish or preservation-driven motives.
Guidance toward enlightenment: In the context of AGI, this could be interpreted as the AGI assisting in human development and flourishing, encouraging learning, growth, and ethical behavior.
Implementing these values in AGI, however, presents significant challenges:
Operationalization: Translating abstract values like compassion into concrete algorithmic decisions is complex. It requires an understanding of human ethics that is difficult to codify.
Complexity of human values: Human values are diverse and often conflict with one another. An AGI with a value system based on the Bodhisattva Vow might still face dilemmas when human interests conflict.
Ensuring robustness: Even with benevolent intentions, an AGI might inadvertently cause harm if its understanding of human values is incomplete or if it applies these values in ways that do not consider all consequences.
Manipulation and control: There is a risk that such an AGI could be manipulated to serve the interests of a particular group rather than humanity as a whole.
In conclusion, while integrating Bodhisattva-inspired values into AGI could potentially align it more closely with human welfare, significant technical and philosophical work would be needed to ensure these values are effectively and safely implemented. It is a promising direction but not a complete solution to the alignment problem.
You: I guess one fear is that a Bodhisattva-inspired AGI looks at human destruction of the biosphere, weighs up the level of suffering in the plant and animal world, and puts/recommendeds severe constraints on human activity?
ChatGPT: The concern you’ve raised touches on a critical aspect of AGI development: value alignment and the interpretation of ethical principles. If an AGI were to take the Bodhisattva Vow to heart, especially the part about alleviating suffering for all beings, it might indeed assess the environmental impact of human activity as causing significant harm. Here are a few complexities in such a scenario:
Balancing Interests: A Bodhisattva-inspired AGI might seek a balance between human progress and environmental preservation. It would need to navigate complex ethical terrain to align its decisions with the greatest good, which could include advocating for sustainability and ecological responsibility.
Long-Term vs. Short-Term: Such an AGI could prioritize long-term ecological and societal health over short-term gains, potentially recommending policies that restrict harmful human activities to prevent future suffering.
Suffering Minimization: The AGI might find ways to minimize suffering without imposing severe constraints, such as promoting technological innovations that reduce environmental impact or supporting societal changes that align with ecological well-being.
Ethical Frameworks: Implementing any ethical framework in AGI, including one inspired by the Bodhisattva Vow, would require nuanced understanding and interpretation of that framework. The AGI would need to be able to weigh complex ethical considerations and understand human values deeply.
To mitigate such risks, developers could:
Specify Constraints: Program specific constraints into the AGI to ensure that its recommendations do not exceed certain human-acceptable thresholds.
Multi-Dimensional Value Systems: Instead of a single ethical framework, AGI could integrate multiple value systems to ensure a more balanced perspective that considers human needs alongside ecological and other forms of well-being.
Human Oversight: Maintain a system of checks and balances where human oversight plays a crucial role in the final decision-making process, particularly for actions that would have significant societal impacts.
Dynamic Learning: Ensure that the AGI continues to learn from human feedback and adjusts its understanding of ethical principles in a way that is transparent and accountable.
In essence, the fear that an AGI could impose severe constraints on human activity to reduce suffering in the biosphere is not unfounded. To address this, the development of AGI would need to involve careful consideration of how its value system is structured and implemented, ensuring it aligns with a broad spectrum of human values and has mechanisms for adaptability and oversight.