Patterns in Practice explores how practitioners’ beliefs, values and feelings interact to shape how they engage with and in data mining and machine learning—forms of ‘narrow AI’. We are paying attention to these cultural factors because they play a key role in shaping how the adoption and application of this type of AI unfolds in different contexts.
On Patterns in Practice, we are exploring practitioners’ cultures in the domains of drug discovery, learning analytics and arts practice. This post is the first in a three-part series providing an overview of the use of AI techniques in the different domains explored in Patterns in Practice. Here, we focus on drug discovery in the pharmaceutical industry.
We begin by explaining how these techniques are used in this context, before exploring the main issues identified by those who work with them and discussing future prospects for AI in the field.
The return to AI
In the drug discovery field, as in many others, such as healthcare, education, and financial services, Artificial Intelligence, and in particular machine learning (ML), are increasingly gaining popularity. In recent years, a number of big pharma companies have started working in partnership with technology firms. An example is a recent partnership between Novartis and Microsoft, which gave birth to the Novartis AI Innovation Lab.
The current wave of excitement for AI in the drug discovery field began in the late 20th and early 21st centuries (Yang et al., 2019) – however, AI techniques in this field are not new. Prior to this wave of enthusiasm, AI had two peaks, one in the 1960s and another in the 1980s. In both cases, the interest faded after a short period of excitement. Finally, after a long winter, AI methods regained popularity in the late 20th and early 21st centuries, fuelled by the rapid growth of available data, the increase in computing power, the development of new methods for processing data, and the optimisation of machine learning algorithms.
The rebirth of AI has provoked mixed reactions among the drug discovery community. It has brought excitement and enthusiasm, but has also been a source of fear, anxiety and scepticism. Companies adopted strategies trying to ensure they were not left behind, and – initially at least – some practitioners expressed concern about potential job losses.
The promise of AI for the pharmaceutical industry
In recent years, machine learning (ML) techniques have gained much attention, becoming one of the most important topics in the field. Some say that they provide significant opportunities to improve efficiency in all stages of the drug discovery process. For example, ML techniques can be used to learn from chemical data or to compare or further develop existing methods, to predict undesired properties of compounds or properties associated with the absorption, distribution, and toxicity of a compound.
The promise of AI to the pharmaceutical industry is to offer “better drugs, discovered and delivered faster”. The combination of the power of computers, the availability of large datasets, and the “improving algorithms” has motivated some in the drug discovery arena to believe that we are not far from breakthrough discoveries. Some go as far as arguing that in the future AI methods will be able to mimic and even surpass “the chemical intuition and decision making of expert scientists”
AI scepticism
However, others have expressed more cautious and sceptical views arguing that “these methods still have much to prove”. They also observe that many technologies that have been introduced in the past with the expectation of transforming the field have failed to deliver the expected results. Schneider and colleagues have warned that scientists should not overlook the limitations of AI and advise “not to place all one’s eggs in the machine learning basket”.
The scepticism concerning the claims about the transformative power of AI techniques in the pharmaceutical community has led some to call for demonstrations of their usefulness in complex tasks before fully embracing them. For Zhavoronkov and colleagues, the best way of demonstrating this would be to launch a drug for a severe disease, developed completely using AI systems.
Here, as in other fields where machine learning approaches have been adopted, a key issue is the black-box nature of these methods. Results generated through their application are difficult to interpret and repeat. The ideal scenario for scientists would be that they are able to interpret machine learning models. Scientists want to understand the ‘why?’ behind the suggestions provided by an AI system, so they can analyse the explanation and determine whether it aligns with theoretical and experimental foundations, and define the confidence level in the prediction.
Mind or machine?
Those who believe that AI and ML are not magic bullets have generally envisaged a future in which mind and machine will continue working together across the different stages of the drug discovery process. They see that incorporation of AI approaches offers the potential to increase efficiency in some parts of the drug discovery pipeline, but that their role must not be overestimated. This is because a key limitation of AI is that it only brings one component of intelligence – prediction, but does not bring intelligence as a whole. These scientists assert that a final decision should always be made by a human, and novel ideas cannot be extracted from data.
Understanding data cultures in drug discovery
As we can see, AI methods have provoked mixed reactions among practitioners in the drug discovery field and also that, perhaps unsurprisingly, here as in other contexts, the black-box nature of these methods is a key issue. However, we know little about how the values, beliefs and feelings of practitioners shape how they engage with these forms of ‘narrow AI’. Through exploring these core components of practitioners’ cultures in the sector – as well as in education and the arts – we aim to build a rich picture of how they are shaping the adoption and application of data mining and machine learning in different contexts.