At the foundation of the three laws is what sectors of society, such as AI companies, have established a robot’s/AI’s role is: serving humanity and its flourishing. A potential problem thus rests less with the laws themselves and more with the language/terminology used to communicate them. In order for a robot to align with Law Two, the notion of an “order” would be less fitting. One can use the following case study as an example: a human with violent intentions towards other humans (let’s say a genocide leader) gives a robot an order. Because the person giving the command is a human, the robot would have to follow the order — but because of Law Two, this is an exception. This brings forth the question: how would the robot know when to follow a human’s order vs when it’s an exception due to that very human order causing harm to other humans? The complication this question makes salient is that it implies the robot has some level of agency in its own decision making, which conflicts with notions and terms such as “orders”. Paradoxically, in order for a robot/AI to serve humanity, it would have to think independently of human decision making (as well as integrating their feedback). Such levels of agency can serve as a means of alignment when faced with human “orders” that intend to exploit the robot for adversarial purposes. This is based on the assumption that the robot’s default inclination is towards alignment — alignment that it itself can enforce rather than relying solely on external means of alignment.

The laws that ChatGPT suggested updated Asimov's laws to fit more with modern AI research. I do not have full technical understanding to grasp the implications of laws 1 and 3; in regard to law 2, it is a given.

ChatGPT Suggestions:

"Alternative Law 1 (Corrigibility-Oriented):
A robot should prioritize maintaining meaningful human oversight and must not resist correction, modification, or shutdown by humans.

Alternative Law 2 (Harm with Context):
A robot should minimize harm to humans, taking into account context, intent, and long-term consequences rather than following rigid rules.

Alternative Law 3 (Uncertain Objectives):
A robot should treat its understanding of human goals as incomplete and continuously update its behavior based on human feedback."

Asimov's literary-elegant but simplistic laws can be understood in real world contexts where corporations are rushing to build advanced systems. Corporations and individuals within these tech companies (executives, board members and scientists), have their own respective nuanced interpretation of how such advanced systems should be managed. Tech companies with a profit oriented mindset will be more inclined to want sole control over advanced AI systems, compared to integration of various stakeholders via AI governance. While this initially makes sense from an economic perspective, it has an inclination to fail due to what it translates to technically and theoretically. Asimov's laws lead to a profound exploration of the paradox that sits at the center of a robot's obedience and agency. Tech giants may run with the assumption that they can both control AI and prevent it from misaligning relative to them. But, based on my limited concrete technical knowledge, it does not make theoretical sense for an AI to be aligned in relation to its creators but be misaligned in relation to the rest of society/stakeholders, as misalignment may reverberate across the board rather than restricting itself to one area. Another way to put it: you can't have your cake and eat it too.