Last week, Microsoft introduced an artificially intelligent chatbot called Tay upon twitter. Designed to mimic the speech patterns of a teenage girl and learn based on the way human twitter users’ interacted with her, Tay was described by Microsoft as “an experiment in conversational understanding”. Unfortunately, less than 24 hours of exposure to twitter taught the bot to be racist, Hitler-loving, and sexually explicit.
We’ve become so accustomed to bad behavior on the internet, it seems surprising that Microsoft didn’t anticipate an outcome like this one. But lets take a moment to suspend our resigned acceptance and ask why, when given the opportunity to interact with Tay, did so many twitter users choose to teach her vitriol?
The answer likely has to do with dopamine.
Dopamine is the brain’s positive reinforcement, a neurochemical intended to reward us for behaviors that promote survival. However, our neurobiology is easily hacked by entertainment technology. Digital media reliably elicits the release of dopamine, especially media that’s violent, absurd, or sexually explicit.
Dopamine produces a euphoric rush similar to opioid drugs. It’s addictive, and we develop a tolerance to the stimuli that elicits it. This means we need ever increasing amounts of the dopamine-producing stimulus to achieve the same dopamine rush.
Many internet users seek out violent, absurd, shocking, and sexually explicit material, bored by “mundane” content. Combine that with the fact that a dopamine high cuts you off from areas of the brain associated with critical analysis, long-term planning, and empathy, and it’s no wonder twitter treated Tay the way it did.
This raises some serious questions about human nature and the future of AI. As James Vincent, writing for The Verge, aptly put it, “how are we going to teach AI using public data without incorporating the worst traits of humanity? If we create bots that mirror their users, do we care if their users are human trash? There are plenty of examples of technology embodying – either accidentally or on purpose – the prejudices of society”.
If we don’t want AI to replicate the problems of our own fallible human minds, we need a mass revolution in understanding and managing dopamine.
by Devon McNaughton