ABSTRACT
Artificial intelligence technologies are revolutionising discovery science and leading a transition into a more data-driven era. However, no discussion of the benefits of artificial intelligence technologies is complete without also addressing their limitations and risks. Artificial intelligence technologies generalise from patterns and regularities in data. As a result, they may also learn biases and “fit” to shortcomings in data, with associated ethical implications. Moreover, there remain many problems for which current artificial intelligence technologies do not provide any good options. In this chapter, the limitations of artificial intelligence technologies for scientific research will be discussed, cutting through the hype that can be associated with these technologies. A key focus is the difference between theoretical advances and implementations in practice in real-world contexts.
