Milena Ivanova is a philosopher of science and lecturer at the University of Cambridge. She is interested in the relationship between science and art, the role of aesthetic values and creativity in scientific discovery, and how AI can enhance human creativity. She has been the recipient of multiple awards from the British Society for Philosophy of Science, the Royal Institute for Philosophy as well as the British Society for Aesthetics. Her recent article ‘The Beautiful Experiment’ won the American Philosophical Association Op Ed competition. She is the author of ‘Duhem and Holism’, published by Cambridge University Press, and the co-editor of two collections: The Aesthetics of Science: Beauty, Imagination and Understanding, and The Aesthetics of Scientific Experiments.
A post by Milena Ivanova
When Matthew Meselson and Franklin Stahl performed their famous experiment, the results of which were published in 1958, their experiment was widely celebrated for its beauty, clarity and significance. The experiment came only 5 years after the discovery of the double helix structure of DNA. This experiment aimed to offer a decisive answer to the question James Watson and Francis Crick posited in their paper ‘Molecular Structure of Nucleic Acids’, published in Nature in 1953: how does DNA replicate? Three hypotheses about DNA replication were offered at the time: conservative, semi-conservative, and dispersive. Meselson and Stahl offered what many consider to be a crucial experiment that decisively answers this question in favour of the semi conservative replicon. The experiment is celebrated for producing important, clear and decisive results and to have definitively settled the question on DNA replication. Beyond producing these results, the very design of this experiment is considered elegant, original and beautiful, making it, according to many, the most beautiful experiment in biology. In ‘What makes a beautiful experiment?’ I have argued that what makes this experiment particularly aesthetically valuable is the relationship between its design and its results, and the original ways in which Meselson and Stahl decided to label DNA in the experiment, using density-gradient centrifugation to study the weight of the different strands of DNA they obtained during the experiment, rather than the standard techniques used at the time. This element of their experiment was original and creative.
Creative thinking and use of the imagination are central elements to celebrating many scientific discoveries as well as works of art. I believe the imaginative solution that Messelson and Stahl came up with in labelling DNA in their elegant experiment is constitutive in valuing its design. Consider another example: the experiments carried out in the late 19th century by Albert Michelson to detect ether drift. In the words of historian of science Gerard Holton, “nobody before Michelson was able to imagine and construct an apparatus to measure the second-order effect of the presumed ether drift. The interferometer was a lovely thing” (1969, 135). Michelson not only created a highly elegant and simple way to measure the presumed effect but constructed one of the most precise instruments ever created in science, used in many discoveries since.
One question that has recently emerged when we think about the role of the imagination in scientific discovery is how the aesthetic value of scientific products is influenced by the increasing automation of the design process. We might wonder whether such automated scientific experiments, models or representations bear the same aesthetic value as those we have traditionally celebrated in the history of science, seeing them as the pinnacle of creative genius, use of the imagination and aesthetic sensibility. One might also wonder whether any of these aesthetic dimensions even matter in science. If we are making more progress than ever using, for instance, AI for scientific discovery, what does it matter if we do not deem the process or the product as aesthetically valuable or imaginative? In this post, I offer two points to motivate why the aesthetic dimension of science matters. First, celebrating the aesthetic value of discoveries, the imaginative and creative thinking of scientists, seem to be a dimension of appreciation throughout the history of science. Second, the aesthetic dimension of daily practice seem to be correlated in an important way to epistemic goals and achievements, but also with an overall sense of satisfaction, well being and flourishing in science. In fact, a very recent sociological study suggests that being creatively involved as a scientist in the process of discovery is a major motivator and reward in itself for scientists, seen as a valued aesthetic experience in the lab.
While imagination and creativity seem to be an important feature celebrated in many experiments of the past, a question emerges as to whether experiments designed today exhibit similar features, given the increased use of AI not only in the more mundane processes of collecting and interpreting experimental results but the more creative part of the discovery process: designing experiments. Take a recent case in quantum mechanics. When Mario Krenn and his team were looking to create highly complex entangled states, Krenn trained a machine learning algorithm MELVIN to explore how such states can be experimentally created. Within a few hours came the solution for which Krenn and his team have been looking for months. The solution has been described by Krenn as surprising and unintuitive. Upon studying the solution Krenn realised that similar solutions had been devised previously in the 1990s but in much more simpler configurations, while his algorithm MELVIN produced what we could qualify as a highly apt experimental set up which has allowed quantum physicists to perform new experiments to study quantum phenomena.
A number of interesting questions emerge when we consider the creative process in such discoveries and the role of the imagination in them. What was MELVIN’s role in this process? One way of describing the contribution of AI is to see it as automating the imagination. I draw inspiration from the mathematician Henri Poincaré, who in his book Science and Method (1908) describes the creative process as involving four distinct phases: preparation, incubation, insight, and revision. During preparation the subject has an explicit problematisation with which they are occupied. During incubation, the subject’s mind freely explores possibilities without being conscious about it; Poincaré aptly calls this stage as delegated to ‘the unconscious machine’, that comes up with moments of sudden illuminations. While problematisation is a conscious process, the process of incubation is subconscious, not requiring the agent to explicitly be working on the problem. After the moment of illumination comes revision where the mind explores the tenability of the ideas produced by the unconscious mind, followed by critical conscious reflection after the inspiration in order for the ideas to be evaluated and assessed. Interestingly, for Poincaré such revisions are driven by our aesthetic sensibility, which acts as a ‘delicate sieve’ in selecting the elegant ideas.
Using the above framework, I see the contributions of AI in creative processes such as designing experiments in terms of Poincaré’s ‘unconscious machine’, an automated process of exploration of possible combinations of ideas. We can call this process prosthetic imagining. Just like microscopes and telescopes extend our capacities and allow us to experience ‘prosthetic perception’, AI can allow the exploration of ideas and extend our imaginative capacities in bringing to our attention combinations we have not yet conceived. But the place for human’s aesthetic sensibility and judgement remains crucial. As Krenn notes, it was his input that streamlined the solution his AIs had proposed and it was human input that generalised the solution of the system and implemented them in different contexts.
While prosthetic imagining can be used to accelerate the conception of new ideas, we also want to be realistic about the kinds of possibilities that AI can help us imagine. Prosthetic imagination can certainly enhance our combinational creativity, but one could question whether it can truly transform a field and overcome the constraints on which algorithms train on. While we may be reminded of Ada Lovelace’s original observation that ‘the analytical engine has no pretensions whatever to originate anything,’ this does not in any way preclude us from considering prosthetic imagination as one way in which AI can enable us to enhance and redefine the creative process.
References
Ananthaswamy, A. (2021) AI Designs Quantum Physics Experiments beyond What Any Human Has Conceived, Scientific American
Holmes, F. L. (2008). Meselson, Stahl, and the replication of DNA (pp. 83–101). Yale University Press.
Holton, G. (1969) “Einstein, Michelson, and the ‘crucial’ experiment”, Isis, 60, pp. 133-197
Jacobi, Christopher J., Peter J. Varga, and Brandon Vaidyanathan. "Aesthetic experiences and flourishing in science: A four-country study." Frontiers in Psychology 13 (2022): 923940.
Ivanova, M. (2023) What is a Beautiful Experiment?. Erkenntnis 88, 3419–3437
https://doi.org/10.1007/s10670-021-00509-3
Ivanova, M. (2021) Duhem and Holism, Cambridge University Press
Ivanova, M and Murphy, A. (2023) The Aesthetics of Scientific Experiments, Routledge
Ivanova, M. and French, S. (2020) The Aesthetics of Science: Beauty, Imagination, Understanding, Routledge
Ivanova, M., Ritz, B., Duque, M.and Vaidyanathan, B. (forthcoming) Beauty in Experiment: A qualitative analysis of aesthetic experiences in scientific practice, Studies in the History and Philosophy of Science
Poincaré, H. (1908/1914) Science and method, Mineola, N.Y.: Dover Publications. Edited by Francis Maitland (1914)