When robots become researchers
How can the rate of scientific discovery be scaled up? Human beings as scientists are limited in several ways—we’re biased towards certain hypothesis, we get bored and tired easily, which impedes our ability to do large scale analyses, and our brains struggle to comprehend some concepts. These limitations greatly reduce the number of scientific breakthroughs that we’re capable of as a global society. And, given the challenges we face, from climate change, to an aging population, to new diseases, it is imperative that our rate of scientific discovery increases.
Prof. Hiroaki Kitano, who leads the Integrated Open Systems Unit at the Okinawa Institute of Science and Technology Graduate University (OIST), thinks that we’ll move beyond these cognitive and behavioral limitations if we utilize the power of artificial intelligence (AI). His recent perspective article, published in npj Systems Biology and Applications, describes how he wants to create an AI scientist that can win a Nobel prize (without the judges realizing that it isn’t human). He foresees that, in around half a century, this will increase the rate of scientific discovery by more than ten-fold.
Science communicator, Lucy Dickie, recently spoke to Prof. Kitano about this research and why he hopes OIST will be at the center of it.
“In a way, our cognitive ability is rather limited in terms of the amount of data we can process per day,” he explained. “We get tired, we get bored. We can’t maintain the same level of attention. We can do high precision analyses perhaps ten or twenty times a day. But if you want to do a thousand data analyses per day, for three years, that’s beyond our capability. But a machine would be able to do that. They don’t complain, they just do it. Machines can also check every hypothesis and determine which ones are more likely, in a way that isn’t influenced by human biases.”
An example that Prof. Kitano gives that an AI scientist would be able to aid with relates to his own research interest in aging. Currently, researchers have a few hypotheses on aging.
“It would be very interesting to have an exhausted research plan on how we can control aging,” he explained. “This is something that we could use an AI scientist for.”
At the same time, an AI scientist could look at which regulatory mechanisms (such as cells and homeostasis) are influencing aging and other diseases and how they be cured by doing an exhaustive experiment at the cellular level. This could lead to drug discovery.
“We can really do everything,” he continued. “We can use this system to understand what is the best way of re-programming a cell—for stem cell, regenerative medicine, or reproductive medicine. That kind of thing, we need relatively exhaustive experiments to understand how we can control the nature of the cell.”
But is this the start of the robot takeover of science?
“The robots aren’t going to take over,” assured Prof. Kitano. “The reality is this will be extremely competitive, but it will also be a reasonably expensive system. At the same time once people understand that this machine can automatically discover, there will be international regulations on the ethical standard on how we can run this system so that people can’t misuse the system to discover something extremely harmful.”
You can hear more about Prof Kitano’s ideas about AI in research and his history with OIST in his presentation in the OIST Forum 2021 (Japanese with English captions):
Title: Nobel Turing Challenge: creating the engine for scientific discovery
Research journal: npj Systems Biology and Applications
Author: Hiroaki Kitano
Date: 18 June 2021