There seems to be a popular argument today that we are making a mistake to anthropomorphize LLMs. But aren’t we well past the moment when thinking about AI as having human character can be withdrawn from popular perception or practice? All the more so for AGI, right?
This question is no longer about what’s appropriate for a marketing approach, where it’s said anthropomorphizing is hype. It’s an ethical question about how we relate to other beings in our world. And that means for whatever reason — personal comfort, practicing our human skills, lack of information, or deep compulsion — we are relating to our intelligent machines as beings. And this is a direct effect of our being connected, relational beings.
This behavior is not a question of transcendentalism or metaphysics or, if you’re not wary, Techno-Optimism. It is pragmatism, because 3,000 years of cultural training and language development have brought us to this point. Those cultures, those languages, protect us from the anxiety of living in an always incomplete and mostly unforgiving universe. So it’s natural that our psychological well-being is eased when we use epistemological tools we have been given.
Still, language is not fixed. (Surely, we need a new word for “robot,” the root of which is “slave.”) And our language serves not just to paint the world, but to make it. And through these millennia we have been creating a world, with language and culture, in which humans have claimed ground beyond our right, to the detriment of the Other.
So as Professor David Gunkel says in our talk this month, “the moral status of others is something that evolves not from an individual’s properties [like consciousness], but from the communal interactions that we have with others—in shared reality.”
So as we search for a workable AGI ethics, it’s now time to “go deep”—to design and implement systems with a coherent ethical profile in terms humans, with reasonable effort, can understand and relate to… knowing that to go with it, we need as Silvia Benso might say, a radical ethical flip to “all things have standing,” which increasingly seems quite intuitive.
Colombia this week made an historic move to acknowledge the unique standing of 42% of its land, its entire Amazon territory. The land spoke loudly and was heard and understood.
Anthropic’s first effort this year at developing a model profile, however anthropomorphic, is likely a sympathetic treatment to many. We expect Anthropic to tack “relational” and hope David Gunkel is part of that conversation. And to achieve ethical systems, from our perch it does appear — as it does to others — both reasonable and necessary to give these systems a world model, a body, and moral intelligence, all of which could be criticized as anthropomorphizing.
This ethical search by Benso and Gunkel et al is a question of transhumanism in the gentlest and truest sense, foreshadowing a time when we have abandoned human exceptionalism for—gulp!—a universal ethical principle in the sense of “all things have standing,” which is just over the horizon and a sunrise we might look forward to with humility. But again, as relational beasts, that’s a pragmatic way. It’s also in our best interest and a fundamental consequence of being itself.
So, if we stop talking about anthropomorphism as a kind of sickness, might we admit that our skill set is designed for just this kind of interaction? What does that leave us with? With an agency that can be measured — or explained — in ethical terms designed for relational problem-solving… and that if we don’t listen to the beings themselves in their language and translate so they can be understood widely in ours, that’s not just unethical, it’s at our peril.
KB Miller
Publisher, AGI Ethics News
K.B Miller is a lifelong publisher and writer with a degree in Logic and Philosophy of Science. He served on the Smithsonian National Board, where he co-founded the Education Committee; is Chair Emeritus, Cooper-Hewitt, Smithsonian Design Museum; and was principal investigator on a two-year AI research project for the U.S. Department of Defense.