Much has been written about Spike Jonze's Her, the Oscar-nominated tale of love between man and operating system. It's an allegory about relationships in a digital age, a Rorschach test for technology. It's also premised on a particular vision of artificial intelligence as capable of experiencing love.
Poetic license aside, is that really possible?
Not anytime soon, though not for lack of processing speed or algorithmic finesse. What computers lack are bodies. The thoughts and feelings and emotions we call "love" are not abstract experiences; they're intertwined with senses and hormones. An AI — a computer hooked to video cameras, a microphone and a screen — would not experience flesh-and-blood love.
"You can't make a computer without a body feel love," said David Havas, director of the Laboratory for Language and Emotion at the University of Wisconsin-Whitewater. Though trying to replicate it "may produce wonderful gadgets, and potentially life-saving achievements, it can never achieve the same result."
'In a sense, the body is the computational engine that makes emotion out of emotionless parts.' Havas isn't simply skeptical because modern AIs are unsophisticated. The opposite is true: AIs sort our mail, defeat our Jeopardy! champions and recommend medical treatments. From behind a screen, it can be difficult to distinguish chatbots from people.
Indeed, with some clever coding and a sufficiently nuanced grasp of human experience, it might be possible to build an AI that gives the appearance of loving. This wouldn't be easy: As philosopher and cognitive scientist Daniel Dennett wrote in "Why You Can't Make a Computer That Feels Pain," some states of being are simply too messy to code. When Siri says, "I did have strong feelings for a cloud-based app once," she's probably faking it.
*Her'*s Samantha is different, though. She's not going through the motions or running predetermined subroutines. Her love wasn't programmed; it grows. She falls in love. She experiences infatuation and fascination, passion and care, a sense of giving and taking and sharing. The breadth and depth of her feelings evolve.
That capacity for growth is difficult to program, said cognitive scientist Benjamin Bergen of the University of California, San Diego. Many mid-20th century AI researchers thought it could be replicated in code alone, imagining human faculties as a mental software suite that would work the same in silicon as in a body. That paradigm underlies Her's essential premise, and it no longer holds.
Instead, researchers in the field of embodied cognition have found close links between body and thought. In experiments, this has been demonstrated in fairly simple ways — the effects of postures and facial expression on emotion, how different textures influence perception — but they suggest a basic principle.
"Emotional comprehension requires a body," said Havas. "In a sense, the body is the computational engine that makes emotion out of emotionless parts."
That's evident in child development, said Bergen. Infants bond with caregivers by being picked up and held, physically consoled, feeling warmth. "If they weren't there, if we didn't have that capacity to sense warmth, if we hadn't been small and able to be picked up, we wouldn't develop those same emotional bonds," he said.
In the experience of adult, romantic love, tactility is likewise important, both in obvious ways and in endless small gestures and touches. Also integral is the endocrine system, which releases hormones that interact with our brain and nervous system. Their totality is what we experience; to develop a truly human sort of love, said Havas, an artificial intelligence would require them. Otherwise it just couldn't feel what we do.
But what if those systems and interactions could be encoded, too? Some theorists, said philosopher Matthew Fulkerson, Bergen's UCSD colleague and author of The First Sense: A Philosophical Study of Human Touch, think an internal, virtual representation of a body -- an endocrine emulator, if not the actual glands -- could suffice. A simple example comes from programs in which physics engines allow virtual bodies to "feel" the pull of gravity on their limbs as they learn to walk.
Something similar, a sort of biology engine, could theoretically be used in AIs, Bergen said. Writing it would, however, require a deep, fine-grained understanding of how biological systems work. Presently that doesn't exist, and might not survive translation into a machine form. Cells and tissues refined by evolution remain far more sophisticated than human hardware. "The details matter," said Bergen. "Silicon is not neurons."
Yet if computers aren't quite ready to feel love, it's worth remembering that love comes in different forms. Human versions encompass only part of the spectrum. If love is, as Bergen says, a species-specific experience, perhaps what artificial intelligences could someday feel -- even in the absence of biology -- would simply be their own particular version. It might not work for Her, but it could still be meaningful.
"A machine won’t actually have to possess every characteristic of emotional love in order to be treated as though it had actual thoughts and feelings and desires," said Fulkerson. He drew a comparison to the way people think of pets.
"Their emotions and feelings are certainly different from ours, and recognized as such, but they are often deeply loved," he said. "It almost doesn’t matter whether they really feel love. They feel enough. I’m guessing something similar will eventually be true of machines."
