blog idea perceptual skills research

Perceptual skills is a tangible concept. You ask people: tilt forward while keeping your spin straight, they try, they fail, and when you ask, “can you feel that your spine is not straight?” they admit they can’t. And we say: if we specifically train these “perceptual skills,” we get amazing outcomes in motor control, skill acquisition, well-being etc. But in all these years, nobody EVER came to us saying “I want to improve my sensory discrimination capacity.” They come with “I have tension here, pain here” and “I want to be stronger and more flexible”. Or, we ask people whether they have any sensations in their muscles when they are not contracted, and only about 50% say they can. The rest don’t even understand the question. It is interesting from the perspective of fundamental science, but also super valuable in education, wellness management, rehabilitation etc.

Comments on key papers.

Shmuelof (2012)

“Overall, while the term skill is frequently used to praise sports players and musicians, the scientiﬁc community still struggles to understand the fundamental processes underlying skill acquisi-tion, or to even agree on what the term should encompass.”

Yes!

“Surprisingly,comparisons between perceptual and motor learning are largely lacking in the

literature (Censor etal.,2012).”

Yes!

“…a fundamental difference between motor and perceptual skills—while motor skills require intention and knowledge of facts, perceptual skills can be learned implicitly outside of conscious awareness.”

No!

This is where we diverge. My argument is that we both OVERestimate how much we actually consciously “know” about a task when we learn a motor task. The experience of “volition” is necessary in a motor control architecture to distinguish between self-generated and externally inflicted forces. But overall, I follow Grossberg on his idea that “movement” (and whatever happens in the Dorsal perceptual stream” is “unconscious”, because it is based on negative matching (constructing a vector and trying to bring it to zero), and therefore cannot support conscious states.

And also I argue that we UNDERestimate how much of our perception can be conscious/don’t have frameworks to easily talk about different levels of processing. For example, the term “proprioceptive awareness” is often used to refer to the knowledge of position in space. Whereas, in our frameworks in Baseworks, we find it relevant to talk about “proprioceptive awareness” as a localizable sensations in muscles/tendons (stemming either from mechanoreceptors or GTOs/muscle spindles). Our questionnaire shows that people differ drastically in their ability to ve aware of these sensations (while, all our respondents were aware of their position in space).

Schorn (2022)

“Humans and other animals possess impressive abilities to learn new motor and perceptual skills through practice and experience. Traditionally, the literature has failed to integrate experimental work on these two forms of skill learning <…> In addition, perceptual learning likely contributes to several motor skills or procedures.”

Yes!

“Perceptual learning can be defined as “an increase in the ability to extract information from the environment, as a result of experience and practice with stimulation coming from it” (Gibson, 1969).”

Yes!

“Perceptual learning, like motor skill learning, occurs over the lifespan and in various domains. Both medical students learning how to detect small but critical abnormalities in an X- ray and sommeliers discriminating complex flavor profiles in wine are able to do so because of perceptual learning (Bende & Nordin, 1997; Sowden et al., 2000).”

Yes!

“Learning new motor skills is not just important for babies learning to walk or pick up objects, but is an important aspect throughout the lifespan.”

Yes!

“Historically, perceptual learning was thought to be due to changes in early visual areas of the brain (Seitz & Watanabe, 2005). However, recent research has challenged this hypothesis and instead supports the idea that perceptual learning might also occur outside of these early visual cortices (Zhang & Li, 2010).”

“In perceptual learning, it is often the case that improvements are specific to the trained stimulus and do not tend to generalize. Early studies of transfer of perceptual learning found it to be highly specific for retinal position (Fahle et al., 1995; Karni & Sagi, 1991), orientation (Fahle & Edelman, 1993; Vogels & Orban, 1985), and trained eye (Karni & Sagi, 1991), with no transfer between similar tasks.”

No!

Technically yes, but the focus here shows that “perceptual skills” here are limited to the visual domain (even though the paper starts from talking about sommeliers.

What I propose is that

(1) we need to extend the term “perceptual learning” to all kinds of perception - auditory, gustatory, olfactory, somatosensory etc.

(2) I point out at Baseworks as an interesting example, where focus on perceptual learning (in somatosensory and interceptive domains) emerged organically from the pressure to improve “communicability” of movement among users with different levels of imitation ability, skill level, and cognitive understanding of tasks.

Mehling (2011)

This is one of the most influential papers on body awareness.

My problem with it is not really about “body awareness” as the most general concept. Some practitioners interviewed for this paper even insisted that they mean “self awareness”, and they would only use the word “body” if it means “mind-body” as a single unit. They clearly ignore the somatosensory aspects, focusing on emotional aspects, more abstract construct of self, and also all the stress reduction related mindfulness literature.

In contrast, our research is trying to talk about somatosensory perceptual skills.

What the Safron paper adds to the conversation:

The most important contribution is the Embodied Self-Model (ESM) concept. An ESM is a body map that
functions as a predictive-memory system and cybernetic controller — not a static map of where limbs are,
but an active generative model that produces predictions about body states and uses prediction errors to
update itself. This reframes Proske’s “central memory mechanism” in a way that’s actually satisfying: what Proske empirically identified is an operational consequence of the ESM — the body’s generative model predicting a configuration and reproducing it.

Your observation about Proske is exactly right: he discovered something that exceeds his conceptual
framework and named it with a placeholder. “Central memory mechanism” is what a spindle specialist with no FEP vocabulary would call what a predictive processing person would call “the proprioceptive generative
model.” They’re pointing at the same thing from different directions.

Counterfactual richness specifically:

In Safron’s framework: “the extent and variety of evoked affordance-related predictive abilities”
determines the quality of presence and agency. High counterfactual richness = a rich ESM that can generate many precise, immediately actionable body-state predictions. This is exactly the phenomenology you
describe: the pre-loaded possibility space, the spatial music, the readiness. And it comes with a mechanism: it’s the density and precision of the generative model’s predictions about what configurations are reachable from here.

What I particularly like about this term for Baseworks: it’s the correct flip side of what science.md
already says about GS/FSA expanding the controlled manifold. Expanding the controlled manifold is the training mechanism. Counterfactual richness is the experiential outcome of that training — more of your
body’s degrees of freedom are represented in the generative model with enough precision to generate actionable predictions.

On the representation/enaction tension:

Safron’s position is actually quite close to where you land: he’s explicitly not a radical enactivist (he breaks from Hutto & Myin at the end of section 4.1, citing [174]), but he’s also not a cognitivist in the Cartesian sense. His ESMs are representations, but they’re action-oriented, grounded in sensorimotor
contingencies, and inherently cybernetic. This is probably the most scientifically defensible middle position. For Baseworks purposes: we can say we’re training the ESM — which is real and empirically consequential — without committing to whether it’s “a representation” in the philosophical sense.

On your personal experience with physics:

Safron’s framework makes this very specific. He argues (section 2.2, Figure 3) that all working memory and imagination are grounded in partially-expressed motor predictions — simulated actions that bias attention and organize working memory spatially (Table 2, “spatial biasing”). Mental rotation, map reading,
architectural visualization — these all rely on partially-expressed motor predictions involving head/trunk/eye orienting systems. So your shift from symbolic→spatial strategies in physics is predicted by the framework: as your body-model ESM became more precise and rich, the motor predictions underlying spatial working memory became more reliable, so the spatial strategy became more tractable than the symbolic one. Before: your spatial ESM wasn’t precise enough for physics spatial reasoning, so you defaulted to symbol-shuffling. After: it was.