There is an old A joke physicists love to tell: Everything was already discovered and reported in a Russian magazine in the 1960s, we know nothing about it. Although the joke is hyperbolic, it accurately captures the current state of things. The volume of knowledge is large and rapidly growing: the number of scholarly articles published on arXiv (the largest and most popular prepress server) is expected in 2021. to reach 190,000—And this is just a subset of the scientific literature produced this year.
We obviously don’t really know what we know, because no one can read the entire literature even in its narrow field (which includes, in addition to journal articles, doctoral theses, lab notes, slides, white papers, technical notes, and reports). In fact, it is quite possible that in this mountain of papers the answers to many questions are hidden, important discoveries have been overlooked or forgotten, and the links remain hidden.
Artificial intelligence is one potential solution. Algorithms can actually analyze text without human supervision to find relationships between words that aid in detection Knowledge. But much more can be achieved if we move away from writing traditional scholarly articles whose style and structure have barely changed in the past 100 years.
Text mining comes with a number of limitations, including access to full text papers and Legal Concerns. But most importantly, AI isn’t really like that Understanding the concepts and the relationships between them, and are sensitive to biases in the data set, such as the choice of papers to analyze. It is difficult for AI – and indeed, even for a non-expert human reader – to understand scientific papers in part because the use of terms differs from one discipline to another and the same term can be used with completely different meanings in different fields. The increasing interdisciplinarity of research means that it is often difficult to precisely define a topic using a combination of keywords in order to discover all relevant papers. Making connections and rediscovering similar concepts is challenging even for the brightest of minds.
As long as this is the case, AI cannot be trusted and humans will need to re-check everything the AI outputs after text mining, a daunting task that challenges the primary purpose of using AI. To solve this problem, we need to make scientific papers not only machine readable but machine-Concept By (re)writing it in a special kind of programming language. In other words: the science of machines in the language they understand.
Writing scientific knowledge in a programming-like language would be dry, but it would be sustainable, because new concepts would be added directly to the library of science understood by machines. In addition, as machines are taught more scientific facts, they will be able to help scientists simplify their logical arguments; detect errors, inconsistencies, plagiarism and duplication; and highlight communication. Artificial intelligence with an understanding of physical laws More powerful than AI trained on data alone, so science-savvy machines will be able to aid future discoveries. Machines with great knowledge of science can help rather than replace human scientists.
Mathematicians have already begun this translation process. They teach mathematics to computers by writing theorems and proofs in languages such as Lean. Lean is a proof helper and programming language in which mathematical concepts can be presented in the form of objects. By using known objects, Lean can interpret whether a statement is true or false, thus helping mathematicians to check proofs and identify places where their reasoning is not sufficiently accurate. The more Lean knew about mathematics, the more she could do. The Zena Project Imperial College London aims to bring the entire undergraduate mathematics curriculum into the Lean curriculum. One day, proof assistants may help mathematicians conduct research by checking their reasoning and researching the vast mathematical knowledge they possess.