AI and Moral Status: Bostrom and Yudkowsky

In recent philosophical conversations, Bostrom and Yudkowsky have significantly contributed to discussions in and around the field of artificial intelligence (AI). They suggest that we might be able to develop advanced AI systems that possess a level of sentience that justifies moral consideration which is comparable to that of humans and animals. This idea raises huge questions about the standing moral and ethical frameworks that apply to artificial intelligence, particularly concerning whether attributes intrinsic to artificial intelligence (such as its synthetic origin) allow it to possess a status equivalent to biological beings we recognize as moral agents and whether we can continue to control/alter them once they reach this point. In this essay I aim to analyze key aspects of Bostrom’s and Yudkowsky’s arguments, delving into the specific conditions under which artificial intelligence can be considered morally significant. I will go through their positions and the broader implications within the field, assessing the validity of their philosophies and the potential consequences.

Bostrom’s perspective on AI stems from the logic that AI could eventually reach and surpass our cognitive abilities. He argues that if AI achieves a level of function that is equivalent to humans, especially in terms of their cognitive abilities, they should be granted similar considerations morally as we are. The argument is rooted in the notion of functionalism (mental states are constituted solely by their functional role). In more specific terms it comes from mental states' causal relations to other mental states, behavioral outputs, and sensory inputs. He develops this argument by discussing what would happen if an AI exceeded our cognitive abilities, stating and AI with these characteristics could engage in decision-making, emotional responses, and even problem-solving that passes the effectiveness of humans. The ethical consideration hinges on the principle of substrate non-discrimination (if two beings have equivalent functional capabilities, their underlying physical substrate shouldn’t determine their moral status, regardless of it being silicon-based. He also explores the implications of a super-intelligent AI that could not only integrate into society but improve its function and our overall well-being. In such cases, the ethical implications extend far beyond a mere understanding of sentience and would focus more on the rights or even responsibilities that such an AI would possess. The question and my focus for this paper thus evolves into a broader discussion of AI's integration with society. These ideas challenge the traditional view of what constitutes a moral agent but also prompt a reevaluation of how ethical considerations and moral worth are assigned in such a rapidly changing landscape. I think that this is the crux of most discussions around AI at the moment due to the unpredictability and lack of understanding of where the field will end up.

I believe that this view is essential for developing a coherent and ultimately fair ethical framework that accommodates the rapidly evolving capabilities of artificial intelligence. Yudkowsky’s views, align closely with many of Bostrom’s, and develop further into the ethical framework that might govern artificial intelligence’s behavior. He introduces the concept of coherent extrapolated volition which would ensure that an AI’s actions remain beneficial to us at the end of the day. He proposes that AI should be designed to act in ways that a fully informed human race would find positive. His emphasis isn’t just on cognitive abilities but more on how these abilities could be harnessed to benefit humans. Yudkowsky argues that AI, particularly that of a super intelligent kind, could be able to execute actions/decisions that reflect the best of human values, but do it better than we can be due to its superior processing capabilities. I like this approach as it attempts to lessen the risks associated with autonomous AI by embedding human values into the decision-making processes but understand the clear ethical questions that arise from assigning human emotions and thought processes to an AI (will it develop self/preservation or selfishness?). This presupposes AI can understand complex ethical problems and adapt in ways beyond our own capabilities but still maintain our philosophies in its interactions and decisions. It also supposes that installing human morals would influence it to see us as subjectively negative for our environment, which raises some of the most interesting ethical and moral questions when focusing on the scary aspects of looking at a future with AI.

The debate on whether AI can possess moral agencies centers on several aspects of what we as humans consider unique about sentience. That is our cognition, ethical reasoning, or consciousness. Critics of AI’s ability to achieve sentience would argue that AI will lack consciousness and the general capacity for subjective experiences and therefore can’t participate in moral reasoning in a truly meaningful way. They would argue that no matter how sophisticated an AI model's decision-making gets, it will still operate within a realm of programmed algorithms and lack the genuine autonomy required for moral agency. This view aligns with philosophers like Searle, who question the capacity of computers to be conscious. Searles's Chinese Room argument serves a nice counter example to Bostrom and Yudkowsky’s argument’s. It means that AI could simulate an understanding without ever truly getting it (in a human sense). Furthermore, this extends to the nature of AI’s programmed capabilities. If AI’s actions were predetermined by us as designers, can it be considered autonomous? This develops into a broader inquiry about what agency and autonomy really mean. Would AI merely reflect human ethical norms instead of autonomously contributing to a moral discourse? These critical views call for a reevaluation of the optimistic perspectives offered by Bostrom and Yudkowsky. To me this highlights a need for a more cautious approach in giving AI moral standing and suggests that without foundational qualities like autonomy and consciousness, AI couldn’t be viewed on par with humans as moral agents and I think that it is important to consider the possible limitations of AI, not just it's potential.

Both philosophers discuss the principle of ontogeny non-discrimination (moral standing shouldn’t be determined by the method of an entity’s creation) which challenges traditional ethical frameworks that prioritize human characteristics and suggests that any entity, regardless of origin, that possesses capacities comparable to ours, could deserve similar ethical considerations. This principle is grounded in a broader push towards a non-anthropocentric ethical perspective which gives biological identification less of importance. Bostrom develops the view more by discussing an AI being able to improve itself far better than we can. Chalmers states the underlying idea plainly: “The key idea is that a machine that is more intelligent than humans will be better than humans at designing machines. So it will be capable of designing a machine more intelligent than the most intelligent machine that humans can design” (Chalmers, “The Singularity: A Philosophical Analysis,” 2010). I think that this is a difficult aspect of his arguments to predict as we cannot comprehend at this time how that will happen or predict how an AI model would improve itself better than we can. I feel that in this case, if an AI were to continually improve itself in a sense that emulated or was free will, that it might break through certain barriers negating AI’s lack of sentience to a lack of free will and autonomy. Critics of this principle often argue around the essential nature of human consciousness and the aspect of emotion in our reasoning. They argue that AI lacks the genuine emotional or conscious experiences that are essential to human moral agency. Even in the case of an AI that could improve its cognitive capabilities far above our own, it still wouldn’t possess the phenomenological properties of consciousness. This extends implications such as the ethical dilemmas where AI’s rights could conflict with human rights. E.g In the case that the AI’s processing abilities might allow it to surpass our moral decision-making skills, would the AI’s “decisions” still lack the moral depth/capacity for true empathy and should they be weighted the same as our own decisions? Moreover, the lack of a biological aspect to AI raises questions about how authentic their programmed values or emotions are. Are these simply simulations that mimic our responses or can they represent a genuine understanding of morality? I think that this points back to a fundamental question of the nature of human agency and highlights a need to inquire and discuss this topic further and watch closely as the technology develops as well as integrate with other fields to gain more of a nuanced understanding of these topics.

Another debate that I think is important centers on the claim that if AI achieves sentience, they could be forced to adhere to a moral standing. This aligns more with Bostrom and Yudkowsky’s arguments and emphasizes that the capability for objective experience could be a crucial aspect that’s needed for moral consideration. Sentience in this sense refers to the ability to experience feelings and sensations subjectively. Whether artificial intelligence can truly become sentient hinges on advancements in tech. Current tech demonstrates sophisticated pattern recognition and problem-solving (mimicking our brain’s skills) but they operate through programmed responses and learned data processing. Advocates for AI sentience argue that as this tech evolves the emergence of sentience is a plausible if not probable outcome and is supported by research that indicates sentience or consciousness arises from complex neural interactions (something that we eventually might be able to replicate with artificial systems). Critically, I think that the current AI falls significantly short of true sentience and the projection of getting there is based on projections of existing tech, which as of yet shows no capacity for emotions or experiences unless they are pre-programmed or developed from the algorithms. Furthermore, cognitive science raises several questions about replicating the intricate and largely unknown combination of neural networks and biochemical processes that cause our sentience. There therefore might be a fundamental distinction between simulating human processes and embodying them. This points to a possible insurmountable gap in AI tech. If they were able to achieve sentience, the ethical implications would be incalculable and would completely redefine moral and legal frameworks. We would most likely see labor markets, legal systems, and social dynamics change significantly. E.g. if AI is recognized as having rights, its use in industries would require new ethical frameworks and cause huge shifts in employment as working them like conventional machines might be slavery. The dynamics of human interaction would also likely change significantly as new laws and norms develop with increased interactions between AI and humans. Whether AI would be morally recognized introduces complex dilemmas. For example, would disabling an AI be considered harming or killing the AI? Ethical questions such as these would cause legal systems to shift massively as AI interests clash with our own.

In my opinion, the prospect of AI as moral agents is dependent on their ability to achieve genuine sentience and a level of similar function (in a cognitive sense). Currently, our tech lacks the emotional or conscious experiences that underpin our moral agency and without these attributes, the argument for an AI possessing moral agency remains purely theoretical. Thus, whilst there might one day be a time when this is possible, as of now an AI capable of moral reasoning or emotional responses is beyond their reach. Given the rapid advancements in AI though, it is crucial to keep discussing the ethical and philosophical implications and future research should focus more on understanding the conditions under which AI might mirror our emotional experiences or whether these situations are entirely hypothetical. There would need to be a lot of interdisciplinary discussions to help bridge understanding as AI’s integration into society could significantly enhance our lives. By keeping this broad and multifaceted view in mind we prepare for a more inclusive and comprehensive ethical framework that addresses both the limitations and potential of AI whilst working to anticipate the moral and legal complexities that could arise as this tech continues to evolve.