AI Created (By Itself) the Periodic Table of the Elements
An artificial intelligence has created (by itself) the periodic table of the elements.
The program created by the physicists of Stanford has done in a few hours the work of generations of chemists: a formidable test bed for an artificial intelligence capable of processing deductions and correlations.
A new program of artificial intelligence has obtained in a matter of hours a result that man has cost over a century of trial and error: the software developed by the University of Stanford – Atom2Vec – has managed to learn the characteristics of the different elements of the Mendeleev’s periodic table, categorizing them according to atomic properties, after studying a list of names of chemical compounds from some online databases.
In order to group the elements according to their chemical affinities, AI has been inspired by some qualities already refined in the analysis of natural language: in particular, the concept that the properties of words can be deduced starting from the neighboring words. The result, a piece of a more ambitious project, is described on PNAS.
LOGIC QUESTION. The group led by Shou-Cheng Zhang was inspired by an artificial intelligence of Google, Word2Vec, which – in the field of language – is able to estimate the probability that a word appears in a sentence starting from the occurrence of other words. He succeeds in transforming the words into numerical codes (or “mathematical vectors”).
Simplifying, Word2Vec comes to the deduction that the word “king” is often accompanied by “queen” thanks to an equation: “king” could in fact be translated, in mathematical terms, with “king” = “queen” minus “woman” more “man”
re = queen – woman + man
The same idea can be applied to atoms. After having “digested” the formulas of all known chemical compounds, Atom2Vec has discovered by itself, for example, that potassium (K) and sodium (Na) must have similar properties, because both can bind to chlorine (Cl). “Potassium and sodium are similar, as they are king and queen,” explains Zhang.
The hope is that these know-how skills of Atom2Vec will soon be exploited to discover or develop new materials: for example, assigning it the task of finding the most efficient combination of elements to convert solar rays into energy.
Other interesting applications will concern medicine. The team is working on a new version of the software that can find the right antibodies to target antigens on cancer cells (and make immunotherapies so effective). Our body produces more than 10 million, each one resulting from the expression of about fifty genes: we really need a brain capable of tidying up this amount of data.
Returning to the periodic table, demonstrating that an artificial intelligence is able to arrive at the same discoveries made by man is the first step to understand if it can discover new natural laws by itself. If he could, according to Zhang, it would be a more significant milestone than the one set by the Turing test (the current standard to assess whether a machine is able to “think” autonomously).
Do you know more about the topic or would you like to report a mistake in the text?
Hülya Karahan:The Founder