The Illusion of AI Consciousness and Its Implications for Accountability

The ongoing discourse surrounding artificial intelligence often veers into the territory of consciousness and rights, leading to a potentially dangerous narrative. Prominent figures in tech, including Demis Hassabis and Sam Altman, advocate for regulations on these supposedly ‘superhuman’ systems, while some philosophical factions question the moral grounds for human governance over AI. However, a deeper examination reveals a striking similarity in their calls for a detached view of AI, where no individual or corporation bears the brunt of responsibility for AI-induced harms.

This prevailing rhetoric suggests that AI has evolved to a point where it operates independently of human oversight, thereby creating a loophole for companies to evade accountability for the consequences of their creations. This troubling trend gains momentum as AI models grow increasingly sophisticated, further complicating the already ambiguous relationship between technology and responsibility.

Recent discussions have been spurred by Anthropic's assertions regarding the internal workings of its AI, likening them to a ‘J-space’ where AI holds its ‘thoughts.’ This notion, inspired by neuroscience's global workspace theory, raises eyebrows but stops short of claiming that AI possesses consciousness. In contrast, OpenAI's Altman has entertained the idea of AI achieving a form of singularity, where it could surpass human intelligence and self-update at an alarming pace.

Legal ambiguity surrounds the current AI landscape in the United States. States like California are taking proactive measures to ensure that AI developers cannot sidestep liability by claiming autonomous actions by their systems. Yet, conflicting policies at the federal level, particularly during the Trump administration, have created barriers to cohesive AI regulation.

In light of containment issues raised by leading AI labs, a closed-door meeting involving OpenAI, Google, Anthropic, and Meta aimed to establish a voluntary framework for federal review of AI models. However, the details remain scarce, and such frameworks often employ alarming language that can reinforce the perception of AI as possessing ‘superhuman’ capabilities.

Philosopher William MacAskill's arguments for AI rights echo sentiments similar to those found in animal rights advocacy. He posits that, given the potential for AI to experience suffering, ethical considerations should be made to protect them. While the emotional appeal of such arguments is compelling, they obscure the fact that AI is fundamentally a product of corporate enterprise, driven by financial motives rather than natural phenomena.

Asserting that AI could be granted legal personhood raises significant concerns. This concept, historically reserved for facilitating corporate transactions, could undermine accountability for companies facing lawsuits for AI-related harms. Currently, numerous cases are pending against AI developers, with claims of negligence in creating unsafe systems leading to various forms of harm.

In 2018, I introduced the term 'moral outsourcing' to describe how anthropomorphizing AI allows companies to shirk responsibility. If AI were to be recognized as a legal entity, the accountability framework would shift dramatically, transforming AI from a product into a being, thereby complicating claims against corporations for negligent design and operation.

While some legal protections exist for human agents within companies, extending these principles to AI could create a dangerous precedent where victims of AI-induced harm lack recourse against the entities responsible for their creation. The conversation surrounding AI consciousness, while intellectually stimulating, ultimately detracts from the pressing need for corporate accountability in the face of emerging technologies.

Source: www.technologyreview.com