Progress in science can be lumpy. Ground-breaking advances, such as the emergence of quantum theory, the development of spectroscopic techniques including IR and NMR, or the widespread adoption of computers in chemistry, catalyse research, and open new fields. Artificial intelligence (AI) is the latest significant advance.
Perhaps, this assessment of AI seems too generous. Can a computational technique really be transformational? After all, in every university chemistry department, there are computers and spectrometers, but in only some is AI used. However, there are signs that AI may have as great an impact in science as it promises to do in society more broadly; this book explores that impact.
We begin with an introduction to some of the core concepts in AI and machine learning, at a level suitable for those who are new to the field (Allen, Chapters 1 and 2). The tutorial that follows outlines some of the most widely used AI methods at the interface of machine learning and medicine (Lawrence, Chapter 3). Subsequent chapters deal with a wide range of topics, including fundamental studies in catalysis (Liu, Chapter 19) and prediction of the properties of materials (Winkler, Chapter 9; Jelfs, Chapter 12; Brgoch, Chapter 13); studies of synthesis design (Stukenbroeker, Chapter 6; Hirst, Chapter 7; Brgoch, Chapter 13); drug design (Hudson, Chapter 11; Speck-Planche, Chapter 16); industrial applications (Curteanu, Chapter 10; Clough, Chapter 14); theoretical areas of chemistry (Marquetand, Chapter 4; Mizoguchi, Chapter 17) and some more technical aspects of the use of AI (Staker, Chapter 15); autonomous chemistry (Stukenbroeker, Chapter 6; Simpson, Chapter 18); and chemical astronomy (Viti, Chapter 8). Several chapters (among them Stukenbroeker, Chapter 6; Hirst, Chapter 7; Brgoch, Chapter 13; Simpson, Chapter 18, and Shankar, Chapter 20) touch upon some of the difficulties that may complicate the use of AI in science and consider how one might circumvent them, while one further chapter (Cartwright, Chapter 5), which also discusses challenges and solutions when using AI in science, is aimed principally at newcomers to the field.
Chemistry research that takes advantage of AI is burgeoning: from a modest number of published papers in 2000, the publication rate had, by the end of 2019, risen 100-fold. Several factors have contributed to this growth:
access to increasingly large volumes of data;
a continuing rise in computer speed;
improvements in the efficiency of AI software;
the availability of innovative specialised chips, such as custom neural network chips;
the re-purposing of graphics chips as fast surrogate CPUs for AI applications;
the refinement of existing AI methods and the development of new techniques;
recognition that some types of problems can be more successfully tackled using AI than with more conventional methods of analysis; and
the availability of upgraded tools for automatic data extraction from sources of non-standard structure, such as journals, Powerpoint presentations, or laboratory notebooks.
To that list, we add (cautiously) the future availability of quantum computers. The current view is that such computers will be particularly well-suited to optimisation problems, which are widespread in chemistry (What does the lowest energy conformer look like? Which potential drug binds most strongly to the protein? Which synthetic route provides the highest yield?). Chemists can expect to be among the greatest beneficiaries of this new technology.
Chapters within this book, written by leading groups in the field, provide intriguing examples of how AI can be applied in chemistry. They provide a fascinating glimpse into the future of AI in science.