Siyu Yao

Since the confirmed discovery of planets outside the Solar System in the 1990s, exoplanet and planet science have experienced exponential growth in data, research practices and knowledge. While the field is interested in ambitious topics like the historical evolution of planetary systems and the potential habitability of other worlds, what scientists can observe is limited. This tension between epistemic goals and challenges invites clarification and analysis from the philosophy of science. What special difficulties do exoplanet and planet science face in knowledge acquisition? How do they affect the construction and confirmation of the field’s hypotheses? What methods and strategies provide opportunities to counter those challenges? Based on ongoing philosophical discussions on the epistemology and methodology of general geosciences, astronomy, and the historical sciences, I show that exoplanet and planet science present epistemic situations alike, but with special challenges and opportunities. Scrutinizing such commonalities and specificities enriches our understanding of how knowledge can be reliably obtained in difficult epistemic situations. I will also discuss how some philosophical topics featured in those general contexts unfold in our specific domain of focus: the realism of unobservable objects, the methodological complementarity between studies of different systems, and the classification of natural objects.

Machine learning models based on artificial neural networks have been increasingly used in scientific research and the public domain. These models are notoriously opaque and may conceal caveats in important tasks. Explainable artificial intelligence (XAI) develops network interpretation strategies that reveal how these models work. However, the situation of XAI does not meet its bright expectations: algorithms proliferate, but there are no agreed-upon standards. I suggest that the notion of the experimenters’ regress (ER) in scientific instruments is useful to understand this situation and yield sobering lessons for XAI. I identify a new variant of ER in the design and application of XAI algorithms: the validity of XAI algorithms and the workings of black-box models cannot be secured without assuming the other. This variant shows two distinct features: strategies for breaking the regress are context-specific, and metrics for evaluating XAI algorithms face a second-order regress. They defy XAI’s current approaches of self-justification and search for universal evaluation standards. ER provides two lessons for XAI. First, as various practical, conventional and communal factors can play a role in decision-making, one should be aware of the considerations involved in designing and applying XAI algorithms. Philosophers can contribute to this by analyzing XAI’s “living standards”, the non-universal and unjustified rationales practitioners use. Second, instead of searching for universal standards, those using XAI should reach pragmatic agreements in each specific context of its application based on empirical test results. Therefore, XAI’s support in the trust of AI through transparency also requires empirical validation.

Machine learning (ML) shows strong performance in making accurate inferences from massive, high-dimensional data. Many scientists turn to this new tool when traditional inferential procedures cannot deal with overly messy data and complex target phenomena. One example is galactic archaeology, a branch of astronomy that aims to unravel the epic history of the Milky Way using the present snapshot of stars with only a handful of physical parameters. Historical inference in galactic archaeology is difficult due to the uncertainty of what counts as evidence of past events. While ML is expected to help address such challenging situations, it comes with its own epistemological limitations. Two issues, theory-ladenness and non-uniqueness, can undermine the evidential power of ML outcomes in scientific inferences. ML’s success in aiding Galactic archaeology teaches lessons about how it can serve key epistemic functions for the historical sciences, unaffected by those two issues. I show that ML can empower the evidence-gathering procedure in science, not by outputting data to be directly taken as evidence, but as evidential constraints: ML partitions and structures data for it to be subsequently interpreted as evidence with the help of domain knowledge and inferential procedures. Used in this way, ML not only provides a solution to the ambiguity of historical evidence but also avoids its own epistemic limitations. This use of ML can be generalized to other historical sciences that engage with big data and automated inference.

Narratives, or constructing storylines, serve important cognitive functions in life and historical studies. A growing interest lies in their roles in generating and structuring frontier scientific knowledge. Philosophers of science characterize narratives as a “technology of sense-making,” as they connect diverse scientific elements from different sources to create a coherent understanding. Distinctive features of narratives lead to both appreciation and criticism. Because narratives can be loose in organization and connect gappy materials, they empower the study of complex phenomena. However, they are also criticized for generating “just-so” stories, which, without being verified with other established scientific methods, are not useful to science. I defend the central role of narratives in science, focusing on cases in astronomy. Astronomy has achieved remarkable progress in the last century despite theoretical and empirical challenges. Philosophy of science needs to explain how knowledge about the universe’s unobservable past can be reliably obtained. I argue that narrative construction is a key solution. Specifically, I argue that unverified narratives can still provide a special kind of knowledge for science, the knowledge about how to use certain scientific tools. I demonstrate this by showing how constructing coherent narratives makes astronomers develop three key tools for productive research. First, narratives enable the establishment of time measurement methods in the universe. Second, in astronomers’ first account of the exoplanet Hot Jupiter, constructing a narrative for its orbital migration suggested the novel use of existing models and produced a toolkit for studying similar phenomena. Third, when astronomers integrate a data-driven method, machine learning, into research, narratives mitigate its uncertainties and adapt it for solving problems about the Galaxy’s past. To conclude, my dissertation demonstrates narratives’ power as a meta-method for science. This also helps to understand important discoveries in 20th-21st-century astronomy and provides lessons for its continuing success.

At the commencement of the universe and in the deep past of the observable realm, the first three minutes is a topic both scientifically challenging and philosophically intriguing. While the universe is believed to have undergone drastic changes over this short period, scientists seem to have essential difficulties with gaining observational evidence and conceiving physics in high-energy conditions. This essay delves into philosophical issues concerning evidence, inference, methodology, and the standard for legitimate scientific knowledge about the early universe. Focusing on three central scientific topics, the Big Bang nucleosynthesis, cosmic inflation, and multiverse, I analyze how scientists employ various forms of evidence, methods, and inferences to achieve remarkable success in obtaining detailed knowledge. It also turns out that those topics face different epistemic challenges and deserve different degrees of credibility. A feature of the early universe is the philosophical root of debates in science. I exemplify this by showing that debates about inflation and the multiverse stem from philosophical disagreements about standards for scientific explanation and the significance of hypothesis testing. I conclude by situating the philosophy of the early universe in the context of big history and pinpointing how it shapes our picture of “the beginning of everything”.

Referential communication is a complex form of social interaction that communicates a spatially or temporally distant referent. Previous modeling practices have studied how artificial agents manage to communicate locations that directly determine foraging behaviors. In our study, we introduce conceptual referential communication. In this mode of referential communication, communicated information can lead to behaviors that change flexibly to suit the environment. Instead of giving specific behavioral instructions, this mode only communicates a label of the desired referent, the location of which is unknown to both the sender and receiver. This requires the signal receiver to adjust its foraging behavior based on its own exploration of the environment. We evolve artificial dynamical agents that can communicate 2 and 3 different labels and successfully forage the target label in changing environments. We found that a typical strategy to communicate and differentiate labels in our experiments is by varying the numbers and lengths of contacts between the agents. We also identify several ways in which the receiver develops inter-neurons that differentiate and store information both from communication and the environment.

The philosophy of historical sciences investigates their distinct objects of study, epistemic challenges, and methodological solutions. Rethinking astronomy in this light offers a contribution. First, the methodology of historical sciences adds to a more adequate description of how astronomers study and utilize token events. Second, astronomy faces a typical difficulty in identifying traces of some past events and has developed a delicate solution. This enriches the idea of trace and suggests a methodology that relies on iterations between data-driven approaches and theory-driven approaches, together with the cross-validation between multiple relevant historical events or datasets.

Artificial neural networks and supervised learning have become an essential part of science. Beyond using them for accurate input-output mapping, there is growing attention to a new feature-oriented approach. Under the assumption that networks optimised for a task may have learned to represent and utilise important features of the target system for that task, scientists examine how those networks manipulate inputs and employ the features networks capture for scientific discovery. We analyse this approach, show its hidden caveats, and suggest its legitimate use. We distinguish three things that scientists call a ‘feature’: parametric, diagnostic, and real-world features. The feature-oriented approach aims for real-world features by interpreting the former two, which also partially rely on the network. We argue that this approach faces a problem of non-uniqueness: there are numerous discordant parametric and diagnostic features and ways to interpret them. When the approach aims at novel discovery, scientists often need to choose between those options, but they lack the background knowledge to justify their choices. Consequentially, features thus identified are not promised to be real. We argue that they should not be used as evidence but only used instrumentally. We also suggest transparency in feature selection and the plurality of choices.