AI

The Periodic Table of Elements. / Photo by: lukas505 via 123RF

 

When scientists account for all known chemical elements, they took almost 100 years to organize each of them in a single chart – the periodic table of elements. But an AI system developed at Stanford University was able to do this in a few hours. So, the researchers behind the AI plan to create a new benchmark for AI systems.

Atom2Vec is the program successfully used by the research team that allowed an AI to recreate the periodic table of elements. The AI has been able to distinguish the differences of atoms in each element by analyzing a list of chemical compound names from an online database. Although the AI was unsupervised for the task, the researchers employed the natural language processing on the system so it can accomplish its goal.

“We wanted to know whether an AI can be smart enough to discover the periodic table on its own, and our team showed that it can,” said Shoucheng Zhang, the lead author of the study and a professor of physics at Stanford.

The results from the AI system in recreating the periodic table opened an ambitious goal – to develop a new benchmark test for AI. Currently, AI systems must pass the Turing test by responding to written questions, patterned to something humans cannot distinguish. However, the test is flawed due to its subjective approach.

Zhang said that humans are a product of evolution and possess a mind with irrationalities, which is very difficult to replicate in an AI system. So, he proposed a new standard to test AI systems that can surpass humans in discovering new laws of nature.

“But in order to do that, we first have to test whether our AI can make some of the greatest discoveries already made by humans,” Zhang added.

The Atom2Vec based on Google’s Word2Vec may be able to lead the development of the new benchmark. Word2Vec is a language used by AI to convert words into numerical codes and vectors. Thus, Atom2Vec converts chemical compound names into recognizable codes to recreate the period table. If an AI can recreate or reinvent scenarios using that language, there is a possibility it can discover new things humans failed to figure out.

Zhang and colleagues are now working on the second version of Atom2Vec and its focus is on solving a medical puzzle – a design for a correct antibody that attacks an antigen to induce an immune response.