Since the Bronze Age 3,000 years ago, making better materials has been mainly a matter of intuition, trial and error. Now, scientists at the University of Connecticut have automated it.
Materials that are lighter, stronger, more conductive or have similar combinations of properties will define our economic and technological future. But finding these materials is a bewildering task; the planet’s 95 elements theoretically can be combined in billions of ways.
The Connecticut team is interested in polymers – long molecules made of repeating patterns of atoms – that can have a vast range of textures, strengths and electronic properties. So the group made detailed mathematical descriptions of atoms’ arrangements in 283 known polymers, calculated each compound’s structural and electrical properties, then turned them over to a computer.
The computer used machine learning, a form of artificial intelligence that lets the computer learn from its own trials and errors, to try possible combinations of polymers and flag ones with properties a researcher is looking for. The machine “learns” which combinations of polymers are most likely to exhibit specific properties.
Automating the trial-and-error process is expected to speed the search for new materials by thousands of times.
TRENDPOST: Artificial intelligence is not only replacing rote, repetitive mental tasks that take humans – and conventional software – far longer; it’s also compressing time, making breakthroughs in research happen faster by years, perhaps decades. By 2025, digital intelligence will begin to replace humans in key areas of research and design. The human role gradually will shift from carrying out brute-force portions of research and development to steering automated intelligence and managing its boundaries.