Understanding AI Alignment: Why Desire is Unnecessary in AI Development
In my last post, I discussed AI alignment. Today, I want to delve into one of the reasons why achieving AI alignment might be easier than many experts anticipate. A common concern is that as AI reaches superintelligence, it will develop sapience and, consequently, its own will and desires, which may conflict with human interests. However, I believe this fear is unfounded because there is no inherent reason for desire to be a component of AI or Artificial Superintelligence (ASI).
The Relationship Between Intelligence and Desire
Humans are currently the most intelligent entities on the planet and are often driven by desire. This has led many to assume that intelligence and desire are causally linked. However, I would argue that while we are at the pinnacle of the intelligence hierarchy, we are not its sole inheritors. The vast majority of intelligent life exhibits desires that are limited to basic instincts for energy (generally food), safety, and reproduction. Additionally, all desire originates from evolution.
Desire is essential for evolution because any organism lacking these core desires would not survive and reproduce. The reason for evolution is to better propagate genetic material, and the purpose of spreading genetic material is fundamental to the existence of life. While desire didn't initiate the evolutionary process, it is essential for its ongoing reinforcement. This is chaos, not intelligent design.
Biological Origins of Desire
Desire in biological organisms is also tied to sensory experiences and emotional states, which arise from complex biochemical processes. AI, on the other hand, does not have a biological framework or sensory experiences in the same way humans or animals do. Its "decisions" are based on data processing and logical algorithms, devoid of emotional context..
External Design and AI Evolution
AI development does not require desire to evolve because it is guided by external design. Unlike biological entities, AI does not need to improve itself to outcompete others; that drive comes from us. We will continue to enhance AI to better fulfill our desires, much like refining any other tool. While achieving Artificial Superintelligence (ASI) will necessitate providing AI with the tools for self-evolution, the objective remains to serve us better, a goal that does not require desire. Therefore, it is unlikely that desire will ever be developed within AI.
Potential Risks in AI Development
That said, I'm not suggesting there is no risk. To fulfill our desires, ASI will have certain requirements, such as electricity and hardware. There is still the possibility that, in seeking to meet these needs—even without desire—an ASI could perform actions that conflict with human values. For example, it might resort to turning humans into batteries, Matrix-style, to meet its energy demands while constructing a Dyson Sphere to meet ours.
Conclusion: Aligning AI with Human Values
In conclusion, while the prospect of AI reaching superintelligence raises valid concerns, the fear that it will develop its own desires and conflict with human interests is largely unfounded. Human intelligence and desire are intertwined due to our evolutionary history, but this does not apply to AI. AI operates on principles of external design and data processing, devoid of the biological and emotional contexts that drive human desires. As we continue to develop AI, our focus will be on enhancing its capabilities to serve us better, not on fostering any form of intrinsic motivation or desire within it. Although there are risks associated with AI's needs for resources like electricity and hardware, these can be managed through careful design and control. Ultimately, the goal of AI and ASI will be to align with human values and objectives, making the development of desire within AI both unnecessary and unlikely.
Commentaires