Artificial intelligence (AI) can already perform many of the tasks that humans take pride in, such as playing chess and trading stocks. Now, a new study from the U.S. Department of Energy’s Lawrence Berkeley National Laboratory revealed that AI is capable of reading old scientific papers to make a discovery that people missed. What does this mean for the future or research?

At the Lawrence Berkeley National Laboratory, researchers put together 3.3 million abstracts from scientific papers that were originally published from 1922 to 2018. They created an algorithm called Word2vec to analyze the abstracts from 1,000 different journals. It seems that even artificial intelligence doesn't have time to read the complete papers.

Word2vec evaluated 500,000 words from the papers about materials science. The AI used machine learning, which is an application that allows it to learn and make improvements without specific programming, to turn words into numbers and find connections among them.

Researchers point out that the AI had "no training in materials science" but was able to use mathematical models and machine learning to find connections among the papers. Word2vec was able to understand the meaning of the words to find hidden knowledge that humans missed.

The papers were about thermoelectric materials, which can produce electricity because of a difference in temperature. For example, they can turn heat into electricity. Silicon-germanium alloys are an example of thermoelectric materials.

Word2vec figured out what would make the best thermoelectric materials and made accurate predictions about future discoveries when researchers stopped the abstracts at 2008. This means that the AI was able to use previous knowledge to predict what scientists found in later years. In addition, Word2vec figured out the structure of the periodic table without researchers having to program it.

Scientists think that if this AI existed in the past, it could have accelerated materials science research in a significant way. So far, researchers have made the AI's list of the best thermoelectric materials available to the public. They also plan to make the algorithm behind Word2vec public, so others can use it, and they want to create a better search engine for abstracts.

AI's ability to scan previously published work and make new discoveries is a powerful feature. It's estimated that from 1665 to 2009, 50 million journal articles have been published. Today, about 2.5 million articles are published every year, and there are more than 20,000 peer-reviewed journals.

When you combine intense competition to publish more work with a growing number of scientists around the world, you get an explosion of information that is almost impossible for any human to analyze. A study by James Evans reveals another concern: Scientists are ignoring older research and citing fewer studies in general. This creates the possibility of them missing or duplicating previous work without realizing it.

AI can help by combing through older research to find relevant sources and better citations. It can also help make connections between different studies that people can miss.

What do the growth of AI and the expansion of its abilities mean for research? Some scientists are welcoming the changes and are embracing new technology. They think that artificial intelligence will be able to make discoveries that improve people's lives.

Others worry that AI will replace people and eliminate jobs. Critics of AI are concerned that it will make humans lazy because machines will be able to do most tasks. Whichever side of the AI debate you're on, it’s clear that there aren't any easy solutions.