Science & Technology

Could Artificial Intelligence Experience Depression?

Imagine, if you will, a future where wars are fought by military robots. Drones installed with artificial intelligence systems find and fire upon enemy targets day in and day out. At some point, a few of these drones break down. All of their internal components are in working order, but when engineers query the systems, they report a lack of energy (despite full power reserves), a dislike of their day-to-day activities, and a desire to self-destruct. If that was a human, you'd say they were suffering from depression. But a robot? Some experts think the idea isn't so farfetched — especially when you consider what it takes to think like a human.

AI, Interrupted

The main suspect behind depression in humans is the neurotransmitter serotonin — the lethargy, apathy, mood swings, and sadness of the disorder are linked to a deficiency of this particular brain chemical. While it's sometimes called the "happy hormone," that's a pretty big oversimplification. Serotonin, like most neurotransmitters (including the equally misunderstood chemical dopamine), actually plays a role in a whole slew of brain functions. It helps out with sleeping, eating, motor activity — the list goes on. But one of its starring roles is in the learning process.

Artificial intelligence, of course, is essentially a learning machine. Engineers feed these systems huge swathes of data illustrating what they want the system to be able to do — for example, pictures of dogs labeled with their breeds to teach an AI to identify an unknown dog's breed — and the AI learns by example.

That's where serotonin comes in, according to neuroscientist Zachary Mainen. In March, Mainen gave a talk at an NYU symposium in which he pondered the possibility of depression in artificial intelligence. In an interview with Science, he explained, "Serotonin is a neuromodulator, a special kind of neurotransmitter that quickly broadcasts its message to a large fraction of the brain ... Computational approaches to neuroscience see other neuromodulators as other sorts of 'control knobs' similar to those used in AI. One important 'knob' from AI is the learning rate."

There are plenty of occasions when increasing your learning rate would come in handy. Mainen uses traveling abroad as an example. "In these situations, your old knowledge — your usual model of the world — is out of date and needs to be sidelined or reworked to adapt to the new situation." Similar occasions routinely come up for artificial intelligence, too.

It's not that we would one day program AI with a sort of e-serotonin — it's more that whatever it is that's helping AI learn might be so similar to serotonin that they might experience its side effects just like we do. It's possible that serotonin is just a "biological quirk," Mainen says, but if it turns out to be an essential part of the learning process in any intelligent system, then robots could certainly get the blues.

Android Dreams

There's a philosophical angle to this, too. Some say that to truly feel emotions like happiness and sadness, you have to be self-aware and conscious. We hardly understand consciousness in humans, let alone robots, so whether or not AI is or will ever be conscious is an open question.

But Hutan Ashrafian, a lecturer at College London in the UK who has written a lot on ethics in AI, says that if an AI system does experience depression, that by definition means the system is conscious. And if these conscious "beings" can experience mental illness, Ashrafian argues, "then it would be incumbent on mankind to perform the humane action of treating them." But how would we treat them? Reprogramming or switching out hardware could irreparably change them, which carries its own ethical problems. Medical care also requires informed consent — would AI be able to grapple with the uncertainty of the risks involved?

We're just at the beginning stages of the artificial intelligence revolution, where our AI systems do great things like search for exoplanets, but also hilariously clumsy things like come up with this list of ideas for Halloween costumes. There will come a time, however, when these questions aren't just hypothetical. And that time may come sooner than we think.

Probably the most influential book about AI ethics is a Philip K. Dick novel written in the 1960s. It's called "Do Androids Dream of Electric Sheep?", but you probably know it by the title of the movie it spawned: "Blade Runner." That's an affiliate link; if you use it to make a purchase, Curiosity will get a share of the sale.

Is Developing Artificial Intelligence (AI) Ethical?

Written by Ashley Hamer May 7, 2018

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.