The Technium

Pets, Aliens, Spirits, and Slaves

There are roughly five different models we can adopt for our relationships with AIs. In the future our lives will be packed with thousands of different species of AIs, so we’ll relate in multiple ways depending on the type of AI were are using at the time.

Pets. Many, if not most, of the AIs we encounter will feel like pets or working animals to us. They will be engineered to be smart in specific, useful ways. Just like some of the animals we have selectively bred for our pets, these AIs will be optimized to respond to our emotions and gestures. They will have limited autonomy, not straying too far from their purpose. While bred to excel in certain jobs, they will lack awareness and intelligence in other dimensions outside their expertise. For example they might be really good at translation, but not navigation. This will summon our most common complaint about them: they are dumbschmarten. They are geniuses in some things, and idiots in others. We’ll swear at them, “how can you be so dumbschmarten?” MIT researcher Kate Darling has written an insightful book about this model, The New Breed. She argues that we already have a well-proven extensive framework in law and cultural norms dealing with the behavior of pets and work animals owned by humans which can be used for AIs and robots. And like the domesticated animals in our lives, these AIs will have recognizable variations and personalities. We’ll prefer one particular individual over another of the same type.

Aliens. More advanced AIs might be thought of as artificial aliens. Their smartness can be encased in degrees of self-awareness and consciousness. Like Spock or Data in Star Trek, their intelligence will be significant, but also non-human-like, or alien. This alienness is actually their chief attraction, because they think differently. Even if their cognition doesn’t wow us, the fact that they have an alien — different — approach makes them a good collaborators. These AIs need a sufficient human interface in order to keep us comfortable working with them. They will be conversational and will mirror our behavior. As we elevate the level of self-awareness and consciousness in these alien AIs, they become more difficult to work with. They can exhibit more attributes of mental dysfunction caused by self-reflection. The price of an alien AI collaborator will be their unfathomable thought process; always surprising us (good and bad) by their reactions. At the same time, some humans will find an affinity for their alienness and will develop careers as AI Whisperers. They will be comfortable working with them, and will be able to get the best from them, which other people will find magical.

Spirits. Many religions around the world perceive spirits or gods that inhabit places, buildings, and objects. This spirit animation is understood to be energetic and lively, but not necessarily highly intelligent. Many of the objects, structures, and systems we make in the future will have mild AI in them that can animate them in a lively way. They can adapt, learn, probe, respond as a living system might. Their intelligence is slow and diffuse, but persistent. It is distributed in the system, and therefore hard to eliminate or even damage. In this way, this kind of AI is like a spirit god inhabiting an old tree, or even a lamp. It has an enduring presence, and can respond to new things, adapt to injuries and assaults, and will persist for a long time, but otherwise it is obscure and vague. The primary purpose of these spirit AIs is to protect the system, and it will collaborate with us to the degree that we cooperate with it. We can imagine an AI spirit god in say the security system of the power grid, or in submarine, or even a piano. Keiichi Matsuda wrote a short essay that explores the idea of using the god metaphor for AIs.

Slaves. A well-worn model for relating to AIs will be to treat them as servants and slaves. To the extant that they are obedient and capable, they act as servants, doing the dirty, hard, unsavory jobs that need to be done. For some humans, these AIs are the perfect servant or slave because they are unambiguously not human, and therefore not deserving of our kindred compasion. They can be considered disposable, interchangeable, and without feelings. They are machines, and so we can excuse any harsh treatment of them. Whatever human-like characters they may gain — such as some levels of self-awareness — these will be cast as mechanical. No matter how much they grow or learn, they will be considered to be made not born, so will always be a machine, and a servant or slave to us. The problem with the master/slave relationship is that it is toxic to us. The AIs might not be bothered, but our human souls are corrupted when we treat sentient beings like slaves. People who mistreat animals will mistreat people, and people who mistreat AIs will also mistreat people. For this reason, we should educate and try to minimize the slave/servant mode of relating to AIs.

Overlord. In science fiction movies the sole model for powerful AI is to become our overlords. In these scenarios the AIs keep getting smarter until they exceed us in brainpower, then they take over and kill us. This mode is often called the Superhuman mode, or the Singularity. While a few prominent tech leaders believe superintelligence is a possibility, I think this mode is highly unlikely and the fear of it is totally misguided. I wrote a long essay on why superintelligent overlords won’t happen (see The Myth of Superhuman AI); the gist of the argument is that superintelligence (as defined in this context) is a distortion of what intelligence in humans actually is. The major human accomplishment of making many varieties of AI will be overcoming our ignorance of how our own minds work. We will discover that the Superhuman Overlord breaks the engineering rule that you can’t optimize everything in all directions and that any real creation — including cognition — must entail tradeoffs. But just as Superman is impossible in real life, yet the myth and archetype of a Superman endures, so even if the Superhuman AI is impossible, its mythic role will endure. There might be times when we feel we are subservient to an AI overlord, even if we aren’t, so this mode is important to recognize.

The types of relationships I don’t expect us to carry on with AIs are the classical human-to-human ones of friends, parent/child, co-workers, or lovers. I believe we’ll reserve these for other humans. I suspect the first attempts to replace humans with “friendly” humanish AIs won’t last. For instance, telephone call centers or retail stores that replace human operators with AIs that sound exactly like humans, will find that ordinary people will prefer to talk to a really intelligent alien rather than an almost-human. The small gap between real and fake feels weird in someone trying to pass as a human, while it is almost charming in an alien. Over time, designers will make AIs more alien rather than trying to fool us into thinking they are 100% human, in part because we are hard to fool this way. At the same time, as we fill the world with myriad new kinds of AIs, I can certainly imagine other models of relationships emerging beyond the five outlined here. If I have missed one that is already latent, please make a comment.