On scientific understanding with artificial intelligence

Introduction. Imagine an oracle that correctly predicts the outcome of every particle physics experiment, the products of every chemical reaction, or the function of every protein. Such an oracle would revolutionize science and technology as we know them. However, as scientists, we would not be satisfied with the oracle itself. We want more. We want to comprehend how the oracle conceived these predictions. This feat, denoted as scientific understanding, has frequently been recognized as the essential aim of science. Now, the ever-growing power of computers and artificial intelligence poses one ultimate question: How can advanced artificial systems contribute to scientific understanding or achieve it autonomously? We are convinced that this is not a mere technical question but lies at the core of science. Therefore, here we set out to answer where we are and where we can go from here. We first seek advice from the philosophy of science to understand scientific understanding. Then we review the current state of the art, both from literature and by collecting dozens of anecdotes from scientists about how they acquired new conceptual understanding with the help of computers. Those combined insights help us to define three dimensions of android-assisted scientific understanding: The android as a I) computational microscope, II) resource of inspiration and the ultimate, not yet existent III) agent of understanding. For each dimension, we explain new avenues to push beyond the status quo and unleash the full power of artificial intelligence’s contribution to the central aim of science. We hope our perspective inspires and focuses research towards androids that get new scientific understanding and ultimately bring us closer to true artificial scientists.

Artificial Intelligence (A.I.) has recently been called a “new tool in the box for scientists”[1] and that “machine learning with artificial networks is revolutionizing science“[2]. Additionally, it has been conjectured “that machines could have a significantly more creative role in future research.” [3]. For instance, it has even been postulated that “[t]he new goal of theoretical chemistry should be that of providing access to a chemical ’oracle’: an A.I. environment which can help humans solve problems, associated with the fundamental chemical questions of the fourth industrial revolution [...], in a way such that the human cannot distinguish between this and communicating with a human expert” [4]. However, this excitement has not been shared among all scientists. Specifically, it has been questioned whether advanced computational approaches can go beyond numerics [5–9] and contribute fundamentally to one of the essential aims of science, that is, gaining of new scientific understanding [10–12]. In this work, we address how artificial systems can contribute to scientific understanding – specifically, what is the state-of-the-art and how we can push further. Besides a thorough literature review, we surveyed dozens of scientists at the interface of biology, chemistry or physics on the one hand, and artificial intelligence and advanced computational methods. These personal narratives focus on the concrete discovery process of ideas and are a vital augmentation to the scientific literature. We put the literature and personal accounts in the context of a philosophi-

Related work. Figure 1. How can Androids contribute to new scientific understanding? In addition to scientific literature, we take inspiration from the philosophy of science and from dozens of stories provided by active computational natural scientists. Thereby we identify three fundamental dimensions of computer-assisted scientific understanding. From there, we look into the future and develop a roadmap on how to develop Androids that can contribute to understanding – the essential aim of science. cal theory of Scientific Understanding recently developed by Dennis Dieks and Henk de Regt [12, 13], who was awarded the Lakatos Award in 2019 for the development of this theory. We thereby introduce three fundamental dimensions for scientific androids1 contribution towards new scientific understanding:

Method. I) Androids acting as a microscope in the responses, i.e., akin to an instrument revealing properties of a physical system that are otherwise difficult or even impossible to probe. Humans then lift these insights to scientific understanding.

II) Androids acting as muses, i.e., sources of inspiration for new concepts and ideas that are subsequently understood and generalized by human scientists.

III) Lastly, in an ultimate dimension of androidassisted scientific understanding, computers are the agents of understanding. While we have not found any evidence of computers acting as true agents of understanding in science yet, we outline important characteristics of such an artificial system of the future and potential ways to achieve it.

In the first two dimensions, the android enables humans to gain new scientific understanding while in the last one the machine gains understanding itself. These classes enable us to layout a vibrant and mostly unexplored field of research, which will hopefully manifest itself as a guiding star for future developments of artificial intelligence in the natural sciences. The goal of this perspective is to put Scientific Understanding back to the limelight – where we are convinced it belongs. We hope to inspire physicists, chemists and biologists and A.I. researchers to go beyond the status quo, focus on these central aims of science, and revolutionize computer-assisted scientific understanding. In that way, we believe that androids will become true agents of understanding that contribute to science in a fundamental and creative way.

Let us imagine an oracle providing non-trivial predictions that are always true. While such a hypothetical system would have a very significant scientific impact, scientists would not be satisfied. We want “to be able to grasp how the predictions are generated, and to develop a feeling for the consequences in concrete situations” [13]. Colloquially, we refer to this goal as “understanding” – But what does that really mean? Can we find criteria for scientific understanding? To do that, we seek guidance from the field of philosophy of science. Notably, while hardly any scientist would argue against “understanding” as an essential aim of science (next to explanation, description and prediction [14]), this view was not always accepted by philosophers. Specifically, Carl Hempel, who made foundational contributions clarifying the meaning of scientific explanation, argued that “understanding” is subjective and merely a psychological by-product of scientific activity and is therefore not relevant for the philosophy of science [15]. Other philosophers criticized these rather unsatisfying conclusions, and they tried to formalize what scientific understanding means. Proposals include that understanding is connected to the ability to build causal models (Lord Kelvin said “It seems to me that the test of ’Do we or not understand a particular subject in physics?’ is, ’Can we make a mechanical model of it?’ ”[13]), connected to providing visualizations (or Anschaulichkeit, as its strong proponent Erwin Schr ̈odinger called it[16, 17]) or that understanding corresponds to providing unification [18, 19]. In recent years, Henk de Regt and Dennis Dieks have developed a new theory of scientific understanding, which is both contextual and pragmatic [12–14]. Importantly, they find that techniques such as visualization or unification are “tools for understanding”, thereby unifying previous ideas in one general framework. Their theory is agnostic to the specific “tool” being used, making it particularly useful for application in scientific disciplines. They extend crucial insights by Werner Heisenberg [20] and rather than introducing mere theoretical or hypothetical ideas, the main motivation behind their theory is that a “satisfactory conception of scientific understanding should reflect the actual (contemporary and historical) practice of Science”. Put simply, they argue that:

A phenomenon P can be understood if there exists an intelligible theory T of P such that scientists can recognise qualitatively characteristic consequences of T without performing exact calculations [12, 13].

Discussion. Second, representing the information in a more interpretable way will help to lift the indications from computers to true scientific understanding.

Conclusion. V. CONCLUSION