Technology 3 min read

MIT CSAIL Teaches AI to Anticipate the Future

Charles Taylor | Shutterstock.com

Charles Taylor | Shutterstock.com

For human beings, predicting the future based on the context of events is something that is refined by experience. One of the biggest challenges facing AI technology is a lack of accurate foresight, but researchers at MIT have made a breakthrough exposing AI to an unlikely source of experience: TV shows.

AI are Awful at Pattern Recognition

MIT’s Computer Science and Artificial Intelligence Lab (MIT CSAIL) has been studying AI technology for years. Their goal is to make intelligent software that can converse, learn, and solve problems by itself. Despite making significant advances toward that goal, there is one problem that has proven exceptionally hard to crack. AI are terrible at predicting future events.

The ability to predict events is something that many people take for granted, but it represents a critical juncture in creating an artificial intelligence.

When people interact, they have a wealth of experience to draw from to understand social cues. For example, when someone raises their hand, your knowledge of the situation is automatically cross-referenced with your experience. That’s how you recognize whether to high five them or answer their question.

This interaction may seem simple to humans, but that’s because we’re efficient at pattern recognition. Until now, AI hasn’t had the necessary depth of experience or a way to acquire it efficiently. With deep learning, the team at MIT CSAIL is changing that.

“CSAIL developed a new algorithm and used it in conjunction with a novel source of experience to feed their AI: television.”

MIT CSAIL Teaches AI with TV

Deep learning allows an AI to make better predictions by identifying and analyzing patterns. As an algorithm encounters data, it can compare it with previously acquired data. This cross-referencing allows AI to make more human-like predictions and assessments.

Despite scientists having created a learning AI, the next step is giving them significant experience that allows their reactions to seem natural to humans.

Humans interact with one another in the physical world, but the burgeoning AI is like a virtual, younger, less socially active person. Whereas people have a life with which to gather experience, AI has traditionally forgotten one problem entirely before moving on to the next. It does not have the capacity to learn like we do.

Fear not, though, because CSAIL developed a new algorithm and used it in conjunction with a novel source of experience to feed their AI: television. By watching YouTube videos and television shows like “The Office,” the AI were able to predict the kinds of interactions that it was going to see between people in the videos with an accuracy never before achieved in the field of AI research.

During the shows, the AI tried to anticipate human interactions such as hugging, kissing, high fives or handshakes. The algorithm was able to make correct predictions 43 percent of the time, which beats existing algorithms that can only get it right 36 percent of the time.

And with deep learning, more data might mean more accuracy. According to Carl Vondrick, a Ph.D. student with CSAIL, “We might see some significant improvements that would get us closer to using predictive-vision in real-world situations.”

Creating intelligence is a highly complicated proposition, but if the scientists at CSAIL have anything to say about it, we may have an entirely sociable AI in our near future. Or, at least, one that won’t leave you hanging when you raise your hand for a high five.

First AI Web Content Optimization Platform Just for Writers

Found this article interesting?

Let William McKinney know how much you appreciate this article by clicking the heart icon and by sharing this article on social media.


Profile Image

William McKinney

William is an English teacher, a card carrying nerd, And he may run for president in 2020. #truefact #voteforedgy

Comments (0)
Most Recent most recent
You
share Scroll to top

Link Copied Successfully

Sign in

Sign in to access your personalized homepage, follow authors and topics you love, and clap for stories that matter to you.

Sign in with Google Sign in with Facebook

By using our site you agree to our privacy policy.