Fei-Fei Li at Google I/O: Humans Overestimate AI in the Short-Term, Underestimate Its Long-Term Potential

Google Cloud Chief Scientist Fei-Fei Li is one of the most popular and influential AI figures today. The woman behind the large-scale image dataset ImageNet is a visionary and an authority on AI’s development. And so it came as no surprise that Li’s panel session today at the Google Developer Conference I/O was packed to capacity.

Li was joined by Google Principal Scientist Greg Corrado and Google Cloud CEO Diane Greene on the “Building the future of artificial intelligence for everyone” panel. The trio shared their views on the future of AI from a scientist’s perspective.

Speaking on artificial general intelligence (AGI) — the long-range, human-intelligence-level target of contemporary AI technology — Li stressed that “humans have a tendency to overestimate the short-term promise while underestimating long-term promise.”

Li gave the example of her two-year-old daughter figuring out how to safely climb out of her crib, a task that state-of-the-art AI would have a tough time with. “Even a cat has things it can do that AI cannot,” said Li.

Corrado, who pioneered the Google Brain project in 2011, followed on Li’s thoughts: “We can build a toy system that imitates a cat’s “meow”, but we don’t have a system that be dumped in a forest and walk out by itself. Machine learning can do pattern recognition, but that is only one small sliver of what goes into something like intelligence. To me, [AGI] is really a leap too far from that.”

One of the biggest barriers to accelerating AI development is its technical complexity, which can exclude outside talents from participating. Li spoke highly of Google Cloud’s newly-released AutoML, a cloud-based tool that automates the custom-coding of machine learning models, thus removing a level of technical difficulty from AI development. A Japan-based chain restaurant is now using AutoML to customize models that recognize different types of ramen.

Another issue facing the AI community today is the risk of bias seeping into models if there is a lack of diversity among the researchers developing a particular technique. Li has been pushing for greater diversity in AI research, and last year launched AI4ALL, a non-profit organization that aims to nurture the next generation of AI technologists with representation from a broad range of ethnic backgrounds. She even sported an AI4ALL shirt at I/O.

Said Li, “I believe in the future of AI changing the world. The question is: who is changing AI? It is really important to bring diverse groups of students and future leaders into the development of AI.”

Journalist: Tony Peng| Editor: Michael Sarazen