If the recent Microsoft Tay debacle tell us anything, it’s that we’re very far away from a robot takeover, no matter what the movies say. Standalone artificial intelligence (AI) is merely a teenager in the technological landscape, and no story proves this more than the 16-hour transformation of Microsoft’s teenybopper persona into a feminist-hating Nazi. AI may be taking over the popular consciousness, but in reality it’s a technology that just wants to fit in.
Here are four typical teenage traits we can expect from bots over the next year or so:
1. Growth spurts. The onslaught of bots appearing in the market has contributed to a series of growth spurts in artificial intelligence that have brought us intelligent personal assistants like Siri and, soon, self-driving cars. With each spurt, the technology advances a bit further and human acceptance increases. But just like with human growth spurts, this process can be painful and will inevitably result in mishaps and spasms along the way, as we saw with Tay and Google Photos before that.
2. Awkwardness. No matter how many maps of the English language AI may review, a computer will never be quite as much of a smooth talker as its older, human counterparts. Putting together a slick sentence with nuance takes a lot more than a mere grasp on language and an arsenal of fancy words. Today’s bots will confuse and confound most humans if left to construct sentences on their own. We’ll need to keep smoothing the rough edges and hand-holding our bots until they mature.
3. Lack of empathy. AI has yet to master a key life skill: understanding others’ emotions. As it stands, machine learning is more like a child: It’s capable of processing basic commands like “yes” and “no.” Until a program can comprehend intangible concepts like “love” or “sorrow,” humans still need to play a role in bot training. Otherwise bots may become gaffe-prone.
“People tend to forget that empathy is a learned skill for humans, and the same is true for AI,” notes John C. Havens, author of Heartificial Intelligence: Embracing Our Humanity to Maximize Machines. “Thankfully, along with great strides in the field of affective computing (focused on emotion), there is a growing desire to implement ethical design methodologies that provably align AI outputs with end-users’ values.”
4. The need to fit in. AI is going through a stage of wanting to please its peers. Most robots at this point will do whatever humans tell them to do. There are roughly 30 technologies that fall under the umbrella term “artificial intelligence.” These include machine learning, robotics, and computer vision. Sabri Sansoy, the founder of WhizRobotics and a former MIT rocket scientist for USAF, believes Microsoft’s Tay bot was most likely an experiment in Recurrent Neural Networks (RNN) and/or a variation of inverse reinforcement learning (IRL), a type of machine learning where an AI system learns both good and bad behavior through observations.
“The danger with RNNs and IRLs is that habits are being formed based on external pressures,” he said. “You wouldn’t want your behavior dictated by a group of Internet trolls. At this point, technologists need to be mindful of whom they want to be influencing their personas.”
When bots begin to actually counter people and offer differing opinions, that’s when we’ll begin to see AI’s full potential. Technologists will need to develop this area further, as well as give bots the ability to communicate in different languages and within different cultural contexts, in order to push artificial intelligence toward adulthood.
In spite of recent bad behavior, AI, like any teenager, holds the promise of something more. We’re seeing rapid progress in shifting the ratio of human versus machine messaging, but the last few areas — those related to human preference and empathy — will be disproportionately difficult to automate. Many experts in the field believe the next AI bot game-changer will be the development of software that mimics mirror neurons, which causes humans to reflect certain actions they see, like a yawn.
Many startups are entering the AI space, but the success of each startup and even AI as a whole will largely depend on careful execution that earns user trust — and that possibly even coaxes us into being better humans.
Swapnil Shinde is CEO and Cofounder at Mezi. He a product entrepreneur with more than 15 years of experience in the tech industry. He previously cofounded Dhingana, a music startup that was acquired by Rdio. At Rdio, he served as the VP of Product for international markets. Prior to working at Rdio and Dhingana, he spent four years at Yahoo as a Senior Product Manager, leading Yahoo’s web performance optimization products.