I just stumbled upon this rather intriguing anecdote about AI research gone amusingly awry. What amazes me the most about it, is the fact that I would expect it to see it being a plot to a science fiction story, and not a news blurb from 2009. And yet here we are, reading about a robot programmed to love, who ended up loving “too much”.
Kenji was part of an experiment involving several robots loaded with custom software designed to let them react emotionally to external stimuli. After some limited environmental conditioning, Kenji first demonstrated love by bonding with a a stuffed doll in his enclosure, which he would embrace for hours at a time. He would then make simple, but insistent, inquiries about the doll if it were out of sight. Researchers attributed this behavior to his programmed qualities of devotion and empathy and called the experiment a success. (…)
The trouble all started when a young female intern began to spend several hours each day with Kenji, testing his systems and loading new software routines. When it came time to leave one evening, however, Kenji refused to let her out of his lab enclosure and used his bulky mechanical body to block her exit and hug her repeatedly. The intern was only able to escape after she had frantically phoned two senior staff members to come and temporarily de-activate Kenji.
If I didn’t know this was real, I would immediately assume this was a synopsis of a short story by Asimov, or someone who was inspired by his robot series. It absolutely blows my mind that researches are seriously working on this sort of thing right now.
On the other hand, I am a bit skeptical. After all, what is love? (baby don’t hurt me, don’t hurt me, no more) I don’t we all agree on what exactly does love entail. Most certainly, we don’t think there is any empirical way to test for it. How can we program a robot to love if we don’t fully understand this emotion ourselves?
If we assume love is somewhat mechanical process where biochemical stimuli combine with social cues, interaction patterns, physical closeness and intimacy then it is easy to see how we would end up with creepily over-eager love bots like poor Kenji. But everything I have ever read on the subject suggests that love, as it is is more complex than that. It is something that requires the impressive abstraction linking quantum machinery that is a human brain to develop and act upon. That it stems not from a baser impulses, but from our slow, self aware frontal lobe area. As far as we can tell we are the only animals capable of love, because we invented it. It is purely a human abstraction and as such it is immensely dense, complex and difficult to untangle into an algorithmic representation or a procedural subroutine.
I guess what I’m saying is that before we can teach artificial intelligences about love we must first give them human level self-awareness. Getting them to fake their way through a Turing test is not enough. They need a genuine sense of self, and identify as a person before they can actually “love” another in a way that would meet any of our hundreds definitions of love. Otherwise, all we can ever hope to get is a poor imitation of love based on instinctive and learned behavior.
To put it in perspective, think about this: you can love your dog, but does it love you back? Surely what the dog “feels” is much simpler, baser impulses like pack loyalty, and borne in instincts to nurture and defend their kin. We like to personify our pets, but deep down inside we know they are not people. If they were, they would be almost comically sociopaths, selfish, rude and impulsive. Because they are fuzzy, cute and dumb we find those qualities endearing and funny – we know dogs don’t operate at the same high level of social interaction as we do.
Think about this – what does your dog do when you leave your house? It snivels, it whines, and it wants to go with you. How is this behavior different from that of Kenji? The researches who made him succeeded in making him “love” people at the same level as a domesticated pet does – a dumb, animal attachment devoid of complexity and nuance. I submit that that’s all we could ever hope for, until we figure out how to build self aware intelligences. And I do believe we are going to do that soon – or at least I hope we will.