Against generalisation: Data-driven decisions need context to be human-compatible

During the past two decades, there have been a number of breakthroughs in the fields of data science and artificial intelligence, made possible by advanced machine learning algorithms trained through access to massive volumes of data. However, their adoption and use in real-world applications remains a challenge. This paper posits that a key limitation in making AI applicable has been a failure to modernise the theoretical frameworks needed to evaluate and adopt outcomes. Such a need was anticipated with the arrival of the digital computer in the 1950s but has remained unrealised. This paper reviews how the field of data science emerged and led to rapid breakthroughs in algorithms underpinning research into artificial intelligence. It then discusses the contextual framework now needed to advance the use of AI in real-world decisions that impact human lives and livelihoods.