The future of AI is developing the capability of interacting with the surrounding world.
Can artificial intelligence understand human humor? According to Fei-Fei Li, professor in the Computer Science Department at Stanford University and co-director of Stanford’s Human-Centered AI Institute, the answer is, not yet. “Today’s technology is not there yet,” she said during an online event organized by the Kibbutz Shefayim-based company Zebra Medical Vision on Tuesday. “What is humor? What kind of sentiment does it carry? Humor requires a deep and nuanced reasoning which is not a strength of current AI.” A former Google VP and one of the world’s expert in the field of computer vision, in the talk Li highlighted how many Israeli researchers have impacted her over the course of her career. “I was very much looking forward to visiting Israel in person for this event, but the coronavirus has prevented me from doing so. It will need to happen in the future,” she said.
In the lecture, the professor focused on different projects to shape the future of artificial intelligence guaranteeing a more ethical approach, a goal that Zebra, a healthcare company proving AI-based medical image diagnosis, also shares. Together with tremendous opportunities, Li acknowledged how the new technologies developed risk to enhance problems such as a wider gap between generations in interacting with machines, but also job displacement, bias and privacy infringements. “For this reason, we believe in a different approach to AI, a human-centered approach,” she pointed out, explaining that the goal is to carry out research with a concern for its human impact, with the idea of augmenting people’s capabilities rather than replacing them, as well as by drawing inspiration from human intelligence. Among the projects illustrated by the computer scientist was the work to reduce bias in AI facial recognition.
“Today’s state of the art facial recognition algorithm is biased in recognizing people from different races, genders and backgrounds. How do we mitigate it from machine learning bias to machine learning fairness? It turns out that there is a whole slew of solutions, starting from datasets and algorithms,” she explained. Li said that AI can also have a significant impact in healthcare, from improving its delivery to help reimagine its policies. The future of AI, the scientist argued, is developing the capability of interacting with the surrounding world. “Is today’s deep learning good enough for AI to form an understanding of human behavior and interfacing with humans? The short answer is no,” she said. “Current AI is powerful but is static, driven by simple reward functions, whereas human intelligence is dynamic, multisensory, complex, uncertain and interactive. The next wave of AI research is going to focus on this much more active perception and interaction with the real world.”