Lightnews — Scholar-powered news

Mark Haselgrove

@markhaselgrove.bsky.social

3.3K followers 930 following 1.5K posts

Professor of Experimental Psychology at the University of Nottingham. Interested in associative learning, and its application to all manner of stuff. Not really interested in brains. He/him. https://scholar.google.co.uk/citations?user=fObPQPsAAAAJ&hl=en

scholar.google.co.uk

Posts Media Videos Starter Packs

Pinned

Mark Haselgrove @markhaselgrove.bsky.social · Jul 3

New paper 😀

"Mechanisms underlying the accuracy of stimulus representations: Within-event learning and outcome mediation."

By Sandra Lagator, Clara Muniz-Diez, @tombeesley.bsky.social and Me

psycnet.apa.org/record/2026-...

1/7

1 3 13

Reposted by Mark Haselgrove

Mark Haselgrove @markhaselgrove.bsky.social · 18h

Week 1 of using an expensive new system to monitor student attendance.

Out with old static QR code check-in system (which students could simply photo with their phone and send to their non-present mates).

In with with the new system that dynamically updates a new QR code every 30 s.

1/2

1 1 1

Mark Haselgrove @markhaselgrove.bsky.social · 17h

It's worth also remembering that Rescorla also noted (in 1980 I think) that the impact of better contiguity between CS and US is often underestimated in many designs because there is a trade off between good contiguity, and the degree of change in the stimulus conditions between training and testing

Mark Haselgrove @markhaselgrove.bsky.social · 17h

SOP (Wagner, 1981) could probably have a stab at the anti-contiguity effect. Very short CS-US contiguity would mean that the elements of the CS and US are not given enough time to both be in the A1 state (supporting excitation). There is a Goldilocks ISI which is neither too short nor too long.

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 18h

Turns out the same handheld technology that can take a photo can also take a video which can easily be livestreamed.

Who would have thunk it?

a bald man is wearing a red sweater with a star trek logo on the front .

ALT: a bald man is wearing a red sweater with a star trek logo on the front .

media.tenor.com

2 5

Mark Haselgrove @markhaselgrove.bsky.social · 18h

1 1 1

Mark Haselgrove @markhaselgrove.bsky.social · 18h

@gurcelay.bsky.social too. He is my go to person on contiguity.

Mark Haselgrove @markhaselgrove.bsky.social · 3d

Always nice to see papers out in the wild.

Opening the batting in this month's @qjep.bsky.social is "Apparent statistical inference in crows may reflect simple reinforcement learning"

By @davidgeorge.bsky.social, Dominic Dwyer, Me & @mikelepelley.bsky.social

share.google/0IoZwV3sne1N...

1 3 5

Reposted by Mark Haselgrove

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Am slowly making my way through this paper. And it is an impressive body of work by a collection of great researchers.

However, I have a couple of problems with it...

1/n

PsyArXivBot @psyarxivbot.bsky.social · 9d

Benchmarks for Associative Learning Models: https://osf.io/qsgz8

1 6 10

Mark Haselgrove @markhaselgrove.bsky.social · 3d

Mark Haselgrove @markhaselgrove.bsky.social · 4d

To sum up, I reiterate my admiration for the hard work and thinking that has gone into this paper.

But for me it departs from one of the assumptions that we often make about learning: it is continuous. Learning is NOT either there, or not. Similarly an effect is NOT either present or absent.

n/n

Mark Haselgrove @markhaselgrove.bsky.social · 4d

A theory that explains maybe just a couple of phenomena under limited circumstances could be very precise and have great predictive validity within its own small world.

Science is big enough to have a place for these types of theory. They just have a different kind of explanatory power.

11/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

If an effect has (seemingly) prescribed circumstances for it to be observed - then such is nature.

Here our theories have not matured enough to understand these prescribed circumstances, and a seeming lack of generality of an effect is not something that we should benchmark our theories to.

10/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Similarly the conditions for producing summation in rabbit eye-blink conditioning are remarkably different to the conditions for producing retrospective revaluation in humans.

9/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Take as an analogy the production of chemical elements. The conditions under which hydrogen is produced (and which any schoolchild in a lab can reproduce) are remarkably different to the conditions for the production of uranium (not typically done by schoolchildren).

And yet both are reliable.

8/n

1 2

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Problem 2. Explanatory value and generality

Beyond (but not unrelated to) problem 1, I am not fully convinced that a theory which can explain the so-called "AA benchmarks" (but fails on the "C findings") necessarily has more explanatory value than a theory which does the opposite.

7/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

For me, it is these kinds of analyses that move the field forward, and help us better understand the variables that influence behaviour.

6/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Indeed, some authors have argued convincingly for the presence of continua (e.g. Urcelay, 2017; Wagner 2003) that help us understand what a particular manipulation might result in (e.g. overshadowing to potentiation; or summation to generalization decrement).

5/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Even simple non-reinforcement following conditioning doesn't always result in a weakening of the CR (extinction), which you might think you could bet your house on. Sometimes, CR increases for a time (extinction bursts).

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Nonreinforced preexposure doesn't always attenuate subsequent conditioning (latent inhibition) - I have shown that it can FACILIATATE subsequent conditioning (and suggested why, and when this will happen).

4/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Many of the compound conditioning phenomena that are well known and sought after (such as blocking, overshadowing, conditioned inhibition) have a corresponding OPPOSITE effect (augmentation, potentiation, second-order conditioning).

3/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Problem 1. Oversimplification.

Categorising training procedures into effects that are either robust or not and auditing them with this lacks nuance.

Where you place the result of a manipulation within the jargon of the field will influence whether it is thought to be reliable or not.

2/n

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 4d

Am slowly making my way through this paper. And it is an impressive body of work by a collection of great researchers.

However, I have a couple of problems with it...

1/n

PsyArXivBot @psyarxivbot.bsky.social · 9d

Benchmarks for Associative Learning Models: https://osf.io/qsgz8

1 6 10

Mark Haselgrove @markhaselgrove.bsky.social · 5d

It's the picture on the title slide for my first year lecture. Took about 30 minutes to stage it!

Mark Haselgrove @markhaselgrove.bsky.social · 5d

Haselgrove's cat knows a good book when she sees one...

1 1

Mark Haselgrove @markhaselgrove.bsky.social · 7d

What a shame. I think I'd be tempted to withdraw it and go elsewhere.