Advances in AI are increasingly framed around automation and job loss. But Google executive Yossi Mathias believes the technology will become most important as a tool to support human scientific progress.
As head of Google Research, he has worked on many projects throughout his illustrious career, including Google Trends, Autocomplete, and Duplex. Matthias’ interest has always been in exploring what will happen next and applying it to reality. He calls this a “magic cycle” and sees AI accelerating this process.
Recently, he has been working on two ambitious AI systems aimed at accelerating scientific discovery: Co-Scientist and ERA, which stands for Empirical Research Assistant. Co-Scientist is designed to help researchers generate and rank new scientific hypotheses. ERA helps automate the tedious process of building computational models and testing those ideas.
This system has already produced interesting results. Collaborators have identified a potential drug repurposing candidate for acute myeloid leukemia and helped elucidate mechanisms associated with antimicrobial resistance, according to a new paper in Nature.
I sat down with Matthias at Google’s recent I/O conference to discuss why he believes AI will help humans, not hurt them. This conversation has been edited for clarity and length.
What excites you most about these AI science systems?
“The idea that we can actually use AI to build tools that help scientists in their research processes, accelerate science, ask bigger questions, and make bigger advances is really exciting to me.”
He said Co-Scientists can sift through vast amounts of scientific literature, generate hypotheses, rank them, and help researchers decide what to test next.
“Imagine a future in which every junior, every scientist (even students) has their own virtual laboratory that helps them sift through endless literature,” he said. “It’s like having a polymath in your pocket.”
Why are hypotheses so important in science?
“When you ask questions like, ‘Find a new drug,’ or repurpose a drug for a different condition, the way you really address this is to form a hypothesis about something that might actually work, and then refine and test it,” he said.
He pointed out that the danger is in spending years pursuing weak ideas, so researchers should be pointed in the right direction sooner if their collaborators can help them create and rank the right hypotheses.
“Ultimately, it will probably lead to new drugs and treatments,” Mathias added.
Could this ultimately help treat diseases such as cancer?
Mattias agrees, but stressed that the process will take time.
“Cancer and other treatments, rare diseases, ALS. Now that we have systems that allow us to look at more global data, there will be even more opportunities,” he told me.
Mathias cited research conducted by Google in collaboration with the UK’s NHS to use AI as a “second reader” in mammograms to improve breast cancer detection.
“In that study, we found that AI can actually identify 25% of errors and give 40% of the time back to the physician,” he said.
And this was AI technology five years ago now. “This was research. The challenge for society and health systems is how to incorporate these learnings. It takes time and effort.”
“I think the potential for accelerating scientific research is huge, and as AI becomes more powerful, we’ll be in a better position to actually continue to reduce, continue to address, and perhaps eventually eradicate. I’m very optimistic that we’ll get to a state where we can actually identify and find solutions to everything that needs to be identified about the disease.”
Will AI replace scientists?
“I think AI is an amplification of human ingenuity because it allows research scientists to ask bigger questions, pursue bigger impacts, and do it earlier in their scientific careers,” he said.
He likened future scientists to lab leaders managing teams of AI collaborators. It’s just as more junior software engineers are taking on broader architectural and strategic roles, while coordinating coding agencies and doing much of the menial work.
Traditionally, it would take a scientist years to run a team of researchers and their own lab, but AI could make this opportunity available to more scientists, he explains.
“I believe that something that was previously very well-established and really only available to a small number of scientists will become available to virtually all scientists,” he said. “The power and effectiveness of all researchers will double, but we are not even close to answering the important questions we need to answer.”
Sign up for BI’s Tech Memo newsletter here. Please contact us by email. abarr@businessinsider.com.