A chatbot trained to engage its partner on personal topics can learn to predict information about the other participant.
Background: Even with AI, chatbots are brittle systems that typically can’t talk about anything outside of what they’ve been trained on. They also tend not to have a personality or long-term memory, so most development for chat agents is around task-specific goals, like airline bookings.
Once more, with feeling: A new paper by Facebook AI Research demonstrates a way to create a bot that is good at general conversation, a.k.a. chit chat. Trained on a new data set consisting of 164,356 utterances between crowdsourced workers who were asked to chat, the system can store a persona in a memory-augmented neural network and produce “more personal, specific, consistent and engaging responses.”
Getting to know you: Using the same setup, the chatbot asks its dialogue partner questions about personal topics, which it can then use to build a model of the partner. Human evaluators still ranked human partners much higher across board for things like persona and consistency, but this model outscored one trained on movie dialogue—one of the largest dialogue data sets out there.
Why Meta’s latest large language model survived only three days online
Galactica was supposed to help scientists. Instead, it mindlessly spat out biased and incorrect nonsense.
DeepMind’s game-playing AI has beaten a 50-year-old record in computer science
The new version of AlphaZero discovered a faster way to do matrix multiplication, a core problem in computing that affects thousands of everyday computer tasks.
A bot that watched 70,000 hours of Minecraft could unlock AI’s next big thing
Online videos are a vast and untapped source of training data—and OpenAI says it has a new way to use it.
The White House just unveiled a new AI Bill of Rights
It's the first big step to hold AI to account.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.