Revisiting ‘Beyond Anthropomorphism’ Tuesday, Feb 16 2010
friendly ai 6:52 pm
My understanding of the concept of anthropomorphism really “clicked” when I first read “Beyond anthropomorphism”, part of Creating Friendly AI, an early (2000) Singularity Institute document. I strongly recommend it for those who are interested in better understanding the concept of non-anthropomorphic artificial intelligence. Here is the opening:
If you punch a human in the nose, he or she will punch back. If the human doesn’t punch back, it’s an admirable act of self-restraint, something worthy of note.
Imagine, for a moment, that you walk up and punch an AI in the nose. Does the AI punch back? Perhaps and perhaps not, but punching back will not be instinctive. A sufficiently young AI might stand there and think: “Hm. Someone’s fist just bumped into my nose.” In a punched human, blood races, adrenaline pumps, the hands form fists, the stance changes, all without conscious attention. For a young AI, focus of attention shifts in response to an unexpected negative event – and that’s all.
As the AI thinks about the fist that bumped into vis nose, it may occur to the AI that this experience may be a repeatable event rather than a one-time event, and since a punch is a negative event, it may be worth thinking about how to prevent future punches, or soften the negativity. An infant AI – one that hasn’t learned about social concepts yet – will probably think something like: “Hm. A fist just hit my nose. I’d better not stand here next time.”
The more I study nature and biology, the more I see that anthropomorphism gets in the way of understanding animals as well. Certain birds, cats, dogs, and even rodents are intelligent, but thinking of their intelligence merely as inferior to humans is not the whole story. Different forms of intelligence have to be understood on their own terms — not through starting with an archetype of human intelligence and making incremental modifications to that archetype. That sort of thinking can lead to anchoring.
You know, punching back is only one of many possible reactions. It comes out of a specific cultural context. Your statement assumes a subject with traditional masculine conditioning. Other responses include but are not limited to the following: running away, crying, falling over, cowering, calling for help, cursing, or asking why. None of these necessarily comes from self-restraint. Humans aren’t genetically programed to punch back when punched. We’re far more complex than that. I find it troubling that you introduce the subject of anthropomorphism with such a simplistic and demonstrably false description of human behavior. In order to meaningfully discuss the potentially limitless flexibility of artificial intelligence in relation to human intelligence you must understand and acknowledge the considerable flexibility of the latter.
@Summerspeaker – Don’t be naive. It’s an analogy, to point out that the instinctive response of a human will be to react to a negative event with pre-programmed responses.
Which an AI will not have.
Harping on the specifics of the analogy is missing the forest for the trees.
I perceive Michael’s presentation here as reflecting a pattern of oversimplifying human behavior. It’s typical evo-psych reasoning. Describing our behavior as crudely mechanistic and AI’s as thoughtful significantly distorts the issue. I’m not merely quibbling. The reality of human pre-programmed responses does not resemble Michael punch-equals-punch-back equation. This matters.
I on the other hand perceive that human behavior is often mystified in academic and philosophical discussion way beyond that which is observable in real life. I fail to see complex behavior with deep, mysterious motives in real life. The patterns, relationships, and attractors that govern animal, including human, behavior are simple.
It seems to me that people, even the most complexly rich and deep ones, all play by the same simple rules. What they do in their profession may be otherworldly in its depth and complexity (say, the greatest mental achievements in science and arts) – truly reaching the ineffable – but not their behavior. Where do you see complexity in day-to-day human behavior?
@Elementary
You beat me to the punch as it were. Clicked through from my reader to post a similar opinion, that humans are behaviorally much simpler than most people tend to believe.
@Michael
Behavior and intelligence are not the same, and the analogy of understanding animals as humans is backwards: you should try to understand humans as animals and see where that takes you.
Anthropomorphism keeps us unaware of the truly alien reality outside of ourselves. It also keeps us unaware of our true being, as science defines it. Our species is but a tiny island in the universe but we mentally extend humanness to a vast number of things and beings. It keeps things seeming more familiar than they actually are.
Elementary: exactly.
Summerspeaker: Yudkowsky could have used any number of qualitatively similar examples. We are not describing AIs as “thoughtful” here, just lacking the inbuilt complexity that humans come with. Any number of examples could have been used instead, such as:
1. When AIs see food, they don’t start watering at the mouth and wanting it.
2. When AIs see “sexy” humans, they don’t see what the fuss is about.
3. When AIs see “oppression”, they don’t automatically consider its meaning in a hominid hierarchical context.
4. AIs would lack almost all of Brown’s Human Universals.
5. AIs would lack the dozens of human-universal facial expressions.
AIs would not have human-like qualities unless they are built in intentionally. That’s the point. I am actually arguing here that the human interpretative milieu is complex and that a baby AI would not be. So I’m not sure why you are accusing me of oversimplifying humans. The metaphor used here is merely illustrative. Even if a human doesn’t immediately want to punch back upon being punched, a whole cascade of mental and physical changes occurs upon being punched, all of which would not exist by default in AI.
While the diversity (this, not complexity, was the original point under discussion) of human behavior isn’t apparently relevant to the point about non-anthropomorphic AI, letting oneself get away with underestimating it in this context risks underestimating it in contexts where it does matter.
@Elementary
Everywhere. Just consider what we’re doing right now. We’re punching keys (or using voice-recognition software) in order to debate the nature of human behavior as it relates to the prospect of artificial intelligence. Our opinions and expression of them come out of intricate histories of personal experience and exposure to myriad intellectual currents. It’s difficult to say exactly why I’m arguing for complexity while you’re arguing against it. A mind-boggling array of interactions between genes and environment lead to this point. If you don’t see that as impressive complexity, you’re probably a lot smarter than I am and I want whatever intelligence-enhancement technology you’ve gotten your hands on.
@Michael
That’s not what the example suggests. Only the AI gets to think; the human reacts like one of Pavlov’s dogs. Humans have one predetermined response while AIs have many possibilities. As such, AIs come out looking more complex.
As a side note, the example both oversimplifies human behavior and understates the potential flexibility of designed intelligence. It does the latter assuming a specific approach to AI and a guiding set of assumptions. An AI wouldn’t automatically consider getting punched in the nose something to be avoided. Given the specific nature of the section (Reinventing retaliation), I can understand framing the issue in this way. Part of my reaction to the piece comes from thinking about it the way you introduced it, as a general introduction to the anthropomorphism debate.
So why not put it that way? Why did Yudkowsky describe human behavior as simply mechanistic when it actually isn’t? (My apologies for erroneously attributing the document to you.)
Summerspeaker: I don’t see that as impressive complexity. I see it as a fairly high (for humans to grasp) number of parameters and variables that influence each other, steps that follow other steps, but not as complexity.
You may have many parts and many interactions, but those interactions can be simple, and those parts can have relationships that don’t produce, require, or exhibit complexity. A high number of parts and interactions doesn’t necessarily mean complexity.
No need for brain boosters: when I’m confronted by apparent complexity I try to find the top level in the hierarchy of patterns of patterns of patterns of patterns of… You may be looking at some lower level. At the top level the underlying complexity is hidden, abstracted away. Looking at it at the top level, human behavior is simple. You don’t have to solve hard problems to deal with humans (though one could argue that we have a special human-behavior processor that makes this hard problem easy for us).
From the perspective and abstraction level you look at things, all systems are seem, and indeed are, complex, there’s no simplicity to be found anywhere. Obviously there’s enormous, humanly incomprehensible complexity and an astronomical number of interactions and information processing occurring per human-scale time units, in the machinery underlying everything alive and even some human-built systems, but that doesn’t mean the resulting behavior is complex; it’s obviously not very complex compared to the machinery itself. The reasons for our discussion and our arguments can be traced back to simple origins and random yet simple interactions. Many steps, sure. Complexity? Nah.
@Elementary
I’m curious as to what you would call complex or complicated. I’d also like to know those simple origins of our argument. While we all abstract away some the complexity of human behavior to produce guidelines we can grasp, I don’t consider this evidence of simplicity. I’m hardly satisfied with my current social ability; I feel certain superhuman intelligences will far surpass us in this area because they’ll have the processing power to better navigate the maze. Both predicting and influencing human behavior are hard problems.
I think claiming that the human response is to punch back is wrong, at least in most cultures. The reality is that pouching someone can result in anything from them committing suicide to wanting to have sex with you depending, on the culture and circumstance. In the past there have been a lot of women who have been abused, and many of samurai warriors who committed suicide when their lord looked at them the wrong way. You are so tried to you own culture you fail to see how different the minds of people in other culture can be.