AI Can Now Predict Future Scientific Discoveries By Analyzing Past Data

Seventy years ago, Alan Turing, the well-known code breaker, mathematician, and computer scientist, asked if a machine could think. Nowadays, several experts believe that the answer is yes. And soon, AI will have something really close to human’s general intelligence. However, others argue that while machines can do certain tasks faster and more accurately than us, they can never have our creativity.

For so long, our intuition and passion have driven us to come up with several spectacular inventions. Some of the examples are fundamental particles and vaccines. However, it is possible that AI can someday compete with that ability of ours.

Recently, an article on Nature argues that AI can predict future inventions by merely extracting information from research publications.

Languages play a more critical part in shaping our modern society than any of us can imagine. It has helped forge relationships, communities, and finally, our intelligence. So it comes as no surprise that the ultimate goal of AI research is to understand human languages. Machine learning, or more accurately, NLP (Natural Language Processing), as we speak, is extracting and evaluating information from written data.

Children often learn how to speak by trails and errors, meaning they have to interact with the outside world. So to have new skills and gain more knowledge, we make mistakes. Machine learning operates in exactly this way, sometimes with educational input.

To train AI to recognize an object, scientists must feed it a large number of examples either showing the said object or not. The guess accuracy is then used to adjust the software’s its statistical model accordingly.

However, we can leave the AI to learn by itself. Either way, its computer program must find how wrong it is and try to minimize errors.

If we want to understand the properties of some material, the manual way now is to find articles online, books, and other specialize literature to find relevant information. AI can help us with this by identifying mutual relationships and concepts from a larger database, in a shorter amount of time.

In the study, the AI learns by itself, meaning the human labor work will be minimized. The system can identify and extract information without human involvement.

Next, the software classifies words into categories like “binders,” elements,” and “gas.” This method can help identify complex relationships and several information layers, which would otherwise be an impossible task to carry out by human.

In this way, the AI can even predict materials many years before they are discovered. So far, we have had five of such inventions, all of them based on scientific papers published before 2009.

A prime example is a thermoelectric substance called CsAgGa2Se4as, which was found out in 2012. As it turns out, the AI can predict its existence with only materials known to science in 2009.

The software was able to do so by making the compound connection with words like “chalcogenide,” materials that contain chalcogen elements, for example, selenium and sulfur; Photovoltaic applications; and optoelectronic (electronic devices that can control, source, and detect light.

The AI’s ability suggests that to accelerate new findings, we could find clues from the past publications and the fact that AI is becoming more and more independent. While this might sound like a fearful prospect at first, it is actually not. It can be of great help for us to navigate and extract information from a vast amount of data that is being created every minute by us.

While the concerns about security and privacy are backed by solid evidence, it is no denying that AI is changing our lives for the better, and ultimately, someday, make us smarter.