Should AI Be Your Alibi for Spousal Murder?

Consider a hypothetical: Should artificial intelligence be developed to a degree where it could assist individuals in planning heinous acts, such as the murder of a spouse? Intuitively, the vast majority would likely respond with a resounding no, perceiving this as a straightforward ethical boundary.

However, the landscape of American thought encompasses a wide array of perspectives, and a notable counterpoint to this conventional view was articulated recently by George Hotz, the founder of Comma AI and a prominent figure known for his work as a jailbreaker.

Hotz's commentary emerges amidst broader discussions on AI alignment strategies, specifically in reaction to the "AI 2040: Plan A" policy paper published by the AI Futures Institute. This influential paper proposes a global collective effort by researchers to deliberately decelerate AI development for 14 years, prioritizing humanity's long-term benefit. Naturally, its premises and conclusions have not garnered universal agreement.

Hotz aligns himself with those who dispute these proposals. In his published remarks, he contends that the "fast-takeoff" scenario—the rapid attainment of superhuman capabilities by AI—lacks practical plausibility. He advocates instead for an approach to AI alignment and safety centered on locally controlled AI models, designed to be intimately aligned with the specific interests of their individual users.

This concept presents an intriguing alternative, especially when considering the prevalent model of current AI services, which are largely built around centrally managed platforms such as Claude and ChatGPT. While infrastructure costs and the infrequent continuous use by most individuals currently justify the centralized hosting of extremely large, state-of-the-art models, these constraints are expected to diminish as technology advances. The experimental, DIY ethos exemplified by projects like OpenClaw highlights the potential value of decentralization, suggesting a desirable direction for future AI product development.

True to his reputation as a provocateur, Hotz extends his argument to more extreme examples. He controversially likens a user-aligned AI to a firearm, asserting that it would not object if used for illicit purposes, such as harming a stepparent. He further postulates that a truly aligned AI, responsive solely to user intent, would facilitate actions like ordering illicit materials (e.g., meth-lab equipment from Amazon Prime) and providing instructions for their use, should a user request it. Hotz emphatically states his willingness to "die to defend this principle," though the practical scenario leading to such an outcome remains difficult to envision.

Hotz encapsulates his philosophy with the declaration, "We either live in a world with freedom or we don’t." While the concept of a "freedom world" inherently appeals, the implications of such unrestricted AI warrant deeper consideration.

However, the discourse surrounding advanced technology cannot be solely confined to an absolutist definition of freedom. Any complex human construct—be it societies, marketplaces, or corporations—necessitates a delicate balance of equities, weaving individual needs into an intricate network of interdependent preferences and robust systems of accountability. Developers of mass-market technological products bear a responsibility to consider this broader societal network, which inherently includes safeguarding the interests of all stakeholders, including potential victims of misuse.

The very "freedom" that Hotz champions is fundamentally a product of collective enterprise and shared societal structures. Such potential futures would rapidly erode if widespread adoption of unchecked, user-aligned AI led to individualistic, "AI-powered Napoleons." As the popular adage reminds us, we operate within a society.

While the prospect of a localized AI capable of advocating for individual interests against corporate entities holds undeniable appeal, the broader societal implications of such unrestrained technological autonomy demand rigorous ethical scrutiny.