Charles Rathkopf

• Doctor of Philosophy
• Research Associate at Forschungszentrum Jülich

Generative AI is increasingly used in scientific domains, from protein folding to climate modeling. But these models produce distinctive errors known as hallucinations - outputs that are incorrect yet superficially plausible. Worse, some arguments suggest that hallucinations are an inevitable consequence of the mechanisms underlying generative infe...

In humans, the reuse of neural structure isparticularly pronounced at short, task-relevant timescales. Here, anargument is developed for the claim that facts about neural reuse attask-relevant timescales conflict with at least one characterization ofneural reuse at an evolutionary timescale. It is then argued that, inorder to resolve the conflict,...

Artificial intelligence holds promise for individualized medicine. Yet, predictive models in the neurobiomedical domain suffer from a lack of generalizability and replicability so that transitioning models from prototyping to clinical applications still poses challenges. Key contributors to these challenges are confounding effects; in particular th...

Evaluating the cognitive capacities of large language models (LLMs) requires overcoming not only anthropomorphic but also anthropocentric biases. This article identifies two types of anthropocentric bias that have been neglected: overlooking how auxiliary factors can impede LLM performance despite competence (Type-I), and dismissing LLM mechanistic...

When actions are mediated by means of a brain-computer interface, it seems that we cannot assess whether the user is culpable for the action without determining whether the brain-computer interface correctly decoded the intentions of the user. Here I argue that this requirement is confused. I also argue that, at least for the purposes of assessing...

Argument maps represent some arguments more effectively than others. The goal of this article is to account for that variability, so that those who wish to use argument maps can do so with more foresight. I begin by identifying four properties of argument maps that make them useful tools for evaluating arguments. Then, I discuss four types of argum...

Position papers on artificial intelligence (AI) ethics are often framed as attempts to work out technical and regulatory strategies for attaining what is commonly called trustworthy AI. In such papers, the technical and regulatory strategies are frequently analyzed in detail, but the concept of trustworthy AI is not. As a result, it remains unclear...

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is v...

Will brain imaging technology soon enable neuroscientists to read minds? We cannot answer this question without some understanding of the state of the art in neuroimaging. But neither can we answer this question without some understanding of the concept invoked by the term “mind reading.” This article is an attempt to develop such understanding. Ou...

In humans, the reuse of neural structure is particularly pronounced at short, task-relevant timescales. Here, an argument is developed for the claim that facts about neural reuse at task-relevant timescales conflict with at least one characterization of neural reuse at an evolutionary timescale. It is then argued that, in order to resolve the confl...

Nervous systems process information. This platitude contains an interesting ambiguity between multiple senses of the term " information. " According to a popular thought, the ambiguity is best resolved by reserving semantic concepts of information for the explication of neural activity at a high level of organization, and quantitative concepts of i...

We argue that Brette's arguments, or some variation on them, work only against the immodest codes imputed by neuroscientists to the signals they study; they do not tell against “modest” codes, which may be learned by neurons themselves. Still, caution is warranted: modest neural codes likely lead to only modest explanatory gains.

Heyes suggests that selective social learning comes in two varieties. One is common, domain general, and associative. The other is rare, domain specific, and metacognitive. We argue that this binary distinction cannot quite do the work she assigns it and sketch a framework in which additional strategies for selective social learning might be accomm...

From Bacteria To Bach and Back is an ambitious book that attempts to integrate a theory about the evolution of the human mind with another theory about the evolution of human culture. It is advertised as a defense of memes, but conceptualizes memes more liberally than has been done before. It is also advertised as a defense of the proposal that nat...

A fascinating research program in neurophysiology attempts to quantify the amount of information transmitted by single neurons. The claims that emerge from this research raise new philosophical questions about the nature of information. What kind of information is being quantified? Do the resulting quantities describe empirical magnitudes like thos...

In this article, network science is discussed from a methodological perspective, and two central theses are defended. The first is that network science exploits the very properties that make a system complex. Rather than using idealization techniques to strip those properties away, as is standard practice in other areas of science, network science...

This article describes one style of functional analysis commonly used in the neurosciences called “task-bound functional analysis.” The concept of function invoked by this style of analysis is distinctive in virtue of the dependence relations it bears to transient environmental properties. It is argued that task-bound functional analysis cannot exp...