Workshop on Imprecision in Perceptual Representations

This interdisciplinary workshop will bring together philosophers and cognitive neuroscientists to discuss different results about, models of, and approaches to the topic of imprecision in perceptual knowledge and representations. 

Saturday: 

11-11:15: Opening remarks

11:15-12:45: Timothy Williamson "The KK Principle and Rotational Symmetry"

1-3: lunch break

3-4:30: Rachel Denison "How does attention affect perceptual precision?"

4:45-6:15: Michael Rescorla "The Perception/Belief Interface: A Bayesian Perspective"

Sunday: 

9:30-11: John Morrison "Probabilistic Representations: Subpersonal and Personal"

11:15-12:45: Megan Peters "The next generation of psychophysics"

-----

Abstracts:

* Denison: Attention is a cognitive process that improves sensory processing. Whereas a large body of evidence shows that attention improves an observer’s perceptual sensitivity measured via performance across trials, we would also like to understand how attention affects an observer’s perceptual precision—which, I will suggest, should be characterized on a single trial and reflect how the stimulus appears to the observer. I will discuss two studies in which we address these aspects of precision. In the first study, we asked whether human perceptual decisions adjust for how attention affects sensory uncertainty on a trial-by-trial basis. We found that human categorization and confidence decisions took attention-dependent uncertainty into account in an approximately Bayesian fashion, indicating that the observer’s attentional state on each trial contributed probabilistically to the decision computation. In the second study, we asked how attention affects the appearance of visual features by asking observers to compare simultaneously presented stimuli. Using these similarity judgments, we estimated the representational geometry of the perceptual feature space and found that attention expanded the representational space around attended feature values. Together these studies support the idea that attention enhances perceptual precision, with implications for how we incorporate precisional changes induced by our cognitive states into downstream computations and behavior.

* Morrison: As animals interact with their environments, they must infer properties of their surroundings. Some animals, including humans, can represent uncertainty about those properties. But when, if ever, do they use probability distributions to represent that uncertainty? Most philosophers focus on uncertainty at the level of belief. In previous work, I considered uncertainty at the level of perceptual experience. In this talk, I will describe an ongoing collaboration with Sam Lippl, Raphael Gerraty, and Niko Kriegeskorte about uncertainty at a subpersonal level, in particular in biological and artificial neural networks. We propose that these networks use probability distributions to represent uncertainty when their representations are: 1) invariant to the source of uncertainty and 2) are used in distinctively probabilistic ways by downstream computations across multiple tasks. I will end by considering how to apply this definition to perceptual experiences.

* Peters: Since the late 1800s, the field of psychophysics has painstakingly described the relationship between physical properties of the world and an observer’s experiences of those properties. Fechner (1860), Weber, Stevens — these are names we all know from our introductory textbooks as pioneers in laying the groundwork for explaining how our minds represent the physical world and act upon those representations. Shortly after the first psychophysical work, others also (namely, Pierce & Jastrow, 1884) became fascinated by the apparent disconnect between what an observer could objectively discriminate and what they could introspectively feel confident in discriminating. However, thanks to the rejection of introspection as a viable target of scientific study due to Freud’s pseudoscience, we all collectively ignored this fascinating topic for at least 70 years. Now, we are tasked with rebuilding the psychophysical study of introspection out of the ashes of behaviorism and the cognitive revolution. But it is working: in the past 20 years, we have seen a resurgence of interest in the quantitative study of introspection (Peters, 2022; Kamerer & Frankish, 2023). In this talk I will describe what I see as the path forward towards "introspective psychophysics": the process relating introspective evaluation to objective (perhaps unconscious) representations/behaviors, including progress we have made and analytical approaches we are building. I will present data and paradigms we have designed and tested, and place these in context with theoretical work linking the quantitative study of introspection to phenomenology in general.

*Rescorla: Bayesian models play a large role in statistics, cognitive science, philosophy, and numerous other fields. The basic idea behind these models is that an agent begins with credence P(H) in a proposition H, then receives evidence E and comes to assign credence P(H | E) to H. The agent is said to have conditionalized on E. Many Bayesian models feature conditional probabilities P(H | E) where P(E)  = 0. Models of this kind figure crucially in Bayesian statistics, Bayesian cognitive science, and many other scientific applications of the Bayesian framework. I explore whether such models can help us analyze epistemic defeat as it arises in perception-based belief. The core idea is that, when an agent conditionalizes on a proposition E such that P(E) = 0, she can dislodge certainties gained through previous exercises of conditionalization.

* Williamson: full paper here https://philpapers.org/rec/WILTKP-3