Hannah Montana once said that nobody’s perfect (a little something we like to call “human error”), but if anyone in the world is close to perfection we expect it to be someone incredibly intelligent, like a doctor or scientist. However, an artificial intelligence algorithm developed by Lawrence Berkeley National Laboratory has exposed connections and made discoveries from a multitude of science writings that had gone unnoticed by researchers. So, how does artificial intelligence make these connections, and what does it mean for the future of science?
The artificial intelligence used in this instance falls under one of the most complex categories of learning in the AI field: deep learning. A machine with “deep learning” thinks and learns to the same degree that humans do, and the machinery that used these old scientific documents to make new discoveries that went beyond the scope of human connection. Researchers from the Lawrence Berkeley National Laboratory were focused on making discoveries in thermoelectricity, the practice of turning heat into electricity, based on previous documentation about which materials would be the most safe and efficient to create it.
To make these discoveries, researchers fed the algorithm, a program called Word2Vec, over 3 million abstracts on materials sciences that allowed for the algorithm to make mathematical connections between the research that humans had previously left untouched. Through linking words often found close together, the AI created segments of related words to create related concepts. This literature processed by the AI concluded which material had the most efficient properties for generating thermoelectricity, but it extended beyond the expectations of researchers by predicting what materials would appear in later studies.
To see how accurate the predictions made by the AI could be, researchers took older materials with conclusions that were already proven true to see if it could make predictions about the discoveries before they happened, and it did.
You’re probably wondering how anyone working in science, a $148 billion industry, could ever leave a stone unturned. The answer? Science is churning out new discoveries and publishings everyday on top of hundreds of prior postings from the last 100 years. So, no matter how many scientists we educate and employ and how many hours we spend analyzing old and new research, there will never be enough time and human brain power to go over the millions of documented scientific discoveries to make the same connections that artificial intelligence can.
So, what does this discovery mean for the future of artificial intelligence? Lawrence Berkeley National Library released the algorithm so that others could use it to research materials. So, expect to see discoveries in materials science and even medicine in the near future. In the meantime, keep hoping that artificial intelligence doesn’t take over the world.