An approach to artificial intelligence that embraces uncertainty and ambiguity could paradoxically help make future virtual assistants less confused.
Gamalon, an AI startup based in Cambridge, Massachusetts, developed the new technique for teaching machines to handle language, and several businesses are now testing a chatbot platform that uses it.
The approach lets a computer hold a more meaningful and coherent conversation by providing a way to deal with the multiple meanings that an utterance might convey. If a person says or types something ambiguous, the system makes a judgment about what was most likely meant.
Today’s virtual assistants and chatbots typically follow simple rules in order to respond to questions. Recent advances in statistical machine learning can add some flexibility by, for example, letting a machine find an answer to a question by searching through large amounts of text. However, both these approaches can fall victim to the vast complexity and ambiguity of meaning often encoded in language (see “AI’s language problem”).
Gamalon’s founder and CEO, Ben Vigoda, told MIT Technology Review his company’s approach also relies on rules and machine learning, but it adds a probabilistic technique to the mix, synthesizing programs that handle probabilities automatically (see “AI software juggles probabilities to learn from less data”). In practice, this means the system can deal with uncertainty by making its best guess about what someone means. It also provides a conversational memory: you could ask “What about tomorrow?” after previously asking what the weather is like today.
Vigoda says the approach lets a machine learn from a smaller amount of data and reduce the rate of errors. It can also show why the machine responded the way it did. “Language isn’t really like a decision tree,” Vigoda says. “This is trying to be more like a person.”
Gamalon has also created an interface that lets ordinary users interact with the system. They can build a powerful chatbot by defining a tree of options for a conversation, letting the underlying system deal with the various different ways the dialogue might unfold. The technology is currently being tested by several companies.
Gamalon is unusual among AI companies in the way it is training machines to perform useful tasks. However, a growing number of experts believe that new techniques may well be needed to achieve significant further progress (see “Is AI riding a one-trick pony?”).
Any advances in natural-language processing could have a big commercial and practical impact. Voice assistants like Alexa or Siri represent very convenient new way to interact with computers, but they are extremely limited in how they use language. Unless you talk carefully, using voice assistants and chatbots can be a pretty infuriating experience.
David Blei, a professor at Columbia University, says Gamalon’s approach brings together several important emerging themes in machine learning. He says the idea of making AI systems more interactive and explainable is especially exciting. “Interactive machine learning is about bringing the human into the loop,” he says. “This is a very realistic way to imagine augmented intelligence could work.”
Meta has built a massive new language AI—and it’s giving it away for free
Facebook’s parent company is inviting researchers to pore over and pick apart the flaws in its version of GPT-3
The dark secret behind those cute AI-generated animal images
Google Brain has revealed its own image-making AI, called Imagen. But don't expect to see anything that isn't wholesome.
Yann LeCun has a bold new vision for the future of AI
One of the godfathers of deep learning pulls together old ideas to sketch out a fresh path for AI, but raises as many questions as he answers.
The hype around DeepMind’s new AI model misses what’s actually cool about it
Some worry that the chatter about these tools is doing the whole field a disservice.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.