“What we’re talking about here is more of a symbiotic team between an AI agent and a human,” he said. “They solve the problems together, it’s not that one of them tells the other what to do; they go back and forth. They can formulate the problem, they can build on each other’s ideas. It’s really important because we’re seeing significant advancements and penetration of AI technologies in almost all industries.”
You can listen to Recode Decode on Apple Podcasts, Google Play Music, Spotify, TuneIn, Stitcher and SoundCloud.
Kurtoglu believes that both in our personal lives and in the office, every individual will be surrounded by virtual helpers that can process data and make recommendations. But before artificial intelligence reaches that level of omnipresence, it will need to get a lot better at explaining itself.
“At some point, there is going to be a huge issue with people really taking the answers that the computers are suggesting to them without questioning them,” he said. “So this notion of trust between the AI agents and humans is at the heart of the technology we’re working on. We’re trying to build trustable AI systems.”
“So, imagine an AI system that explains itself,” he added. “If you’re using an AI to do medical diagnostics and it comes up with a seemingly unintuitive answer, then the doctor might want to know, ‘Why? Why did you come up with that answer as opposed to something else?’ And today, these systems are pretty much black boxes: You put in the input, it just spits out what the answer is.”
So, rather than just spitting out an answer, Kurtoglu says virtual agents will explain what assumptions they made and how they used those assumptions to reach a conclusion: “Here are the paths I’ve considered, here are the paths I’ve ruled out and here’s why.”