Dr Miles Tufft — UCL
Behaviours and their associated cognitive mechanisms do not exist in isolation, rather they are embedded in a world that is naturally social, and rich in context. By experimentally recreating minimal social contexts, while maintaining an acceptable degree of control, I investigate how higher order social factors such as social hierarchy, interpersonal synchrony, and group membership come to meaningfully shape implicit cognitive mechanisms. With examples from my recent studies, using the picture-word interference and spatial cuing paradigms, I will describe the online methods we have used to reveal how social contexts afford the offloading or suppression of task-irrelevant, distracting information (social offloading) in ways that are sensitively attuned to the social characteristics of dyadic interactions. In doing so, I aim to share learnings on what is possible when using online experimental platforms for social cognition research, as well as offering insights on the quality and reliability of collecting online chronometric data by comparing lab-based versus online studies. I finish by emphasising the importance of re-worlding participants in meaningful contexts to reveal the embeddedness of behaviour.
Jo Evershed 0:00
Hi, Miles. lovely to have you back here.
Hi Jo. It’s great to be back.
Fantastic screen over to you whenever you’re ready.
Miles Tufft 0:14
Have I shared the correct screen that’s next question. Share the screen. Do you see my slides?
Jo Evershed 0:23
I am seeing your slides. Yeah. One. Yeah. And they’re full screen and I can’t see any notes.
Miles Tufft 0:27
Okay, great. Wonderful. Over to you. Thanks, Jo. Hello, everyone. And it’s, as Jo says, great to be back. I was here two years ago now for beonline 2020, right at the beginning of the pandemic. And back then, I shared a few insights in terms of some of the work that we started to do to try and transition our paradigms into an online environment. And as Jo introduced me as Yes, I research social cognition. And I noticed in the chat earlier, that there are a number of people here that are interested in attention allocation, distractor suppression, social cognition, executive control. So hopefully, those people there will find some things of interest here in terms of how we’ve transitioned our paradigms and social cognition into an environment using Gorilla.
So what I thought I’d start off by, just give you a little bit of a quick background in terms of what my research is about. So rather obscurely described this talk as socially embedded distractor suppression. Most broadly, I’d say that what I’m interested in is the way in which context shapes the way information is prioritised. So, the way information is prioritised in cognition, what we up weight what we down weight, how does context shape that. And one way to think about information, priorities and prioritisation is to think in terms of distractions. So, here’s the most distracting thing that I could find off the internet, the movements. And you can imagine a person in this environment here being distracted.
And you can think of the distractions in terms of things that are task relevant and task irrelevant. So there is a sort of conflict between these two pieces of information. So there is a some element here of requirement for prioritisation, to the stuff that’s important. So down weighting distractors, perhaps up weighting targets in order to adapt to this sort of task relevancy. Now, the way we’ve gone about operationalizing this is actually to think of it in terms of semantic interference paradigms here, in this case, well, interference paradigms more generally. But here I’m going to talk specifically about semantic interference.
So I’m just gonna move this so I can actually see what I’ve got on my slide. Okay, so let me just explain the power of the base paradigm that we’re looking at. So in this paradigm, participants would see a target here, the target being the picture, a dog, and some sort of form of distractor, which would be a word on top. Now, the distractor can come in two flavours. So the task itself is that the participant has to identify the picture here in the sense of a dog. And they have to say, the word dog, and the distractor on top is a stapler, which is semantically unrelated. So there is no semantic connection between stapler and dog, there are different categories.
However, in another condition, there could be a distractor, on top of the word here, mouse, which is actually of the same category, semantic category in which there would be a conflict and these are related. And what you find is that that semantic relatedness actually slows down slightly the identification of the target picture underneath. So relating it back to what I was talking about there in terms of prioritisation of information. Here, we’ve got two pieces of information. And what we want to do is prioritise the target here, which is the dog, but that prioritisation, that target is being interfered with with the distractor information here, the word on top, which is obviously also being processed in some way.
So this, here’s some data, this is some data that we collected, actually pre pandemic in the lab of just a basic paradigm part with participants on their own, just doing the PWI tests. And what you get is this, what we call a semantic interference effect of about 42 milliseconds here. So they are 42 milliseconds slower to identify the dog when the word on top is related in meaning, is the basic is the basic idea.
Now what we’re going to ask here is if we socially embed participants, so we put participants in a social situation where we divide up the task. So what we’re saying here is can we change the way that the distracting information on top is processed depend on whether that’s been looked after by another person. So here we have a second participant come in, and we tell the participant that this other person is going to take care of that mouse information. And then the participant still has to identify the dog. So the partner says mouse and the participant says dog. And the basic hypothesis and idea here was could we change the way that that distracting information is prioritised? Maybe deprioritize because now the participant believes is being taken care of by their partner.
Now what we did pre pandemic is we brought people in the lab we had two conditions and a loan condition where they sat by side by side across a partition, whether you either had the computer disconnected or connected, and they either believed that they were working together or alone. And then the together condition the key manipulation was the belief that your partner was looking after that word, they were doing a task on the word. And in this case, the participant rather than speak the word, they just had to decide where the last letter of the target picture ended in a vowel or a consonant. So remember, dog, it ends in a consonant. So they had to press the nine key on the keyboard if it was consanant, or the one key was a vowel, and their partner looking after the word had to decide how many vowels were in were in the work, but didn’t really matter, just the idea that the partner was in some way looking after that information.
And what we found is that we got this, okay, this was the bass result that I showed you, the 42 milliseconds. And we found this really curious result, that what happened well, when your partner looked after that piece of information, or you just believed it, and let me remind you here, the partner, there is a confederate. So that actual participant partner is actually doing nothing that just sat there, they just believe, that mere belief was actually reducing the degree to which that distracting information was being processed. And you’re negating the semantic intereference, we removed semantic interference, this automatic processing information was being shaped by the context in which you found yourself.
So we were really very excited about this, we were excited about all the different possibilities that we could do moving forward, and then the pandemic hit. So then we had to try and find a way of transitioning it online. And this is some of the work I briefly presented the first time it’d be online. But I didn’t have the punchline yet of how it all worked out. So I’m going to go through some of the basic results that we found. And some of the manipulations that we then followed up on to try and show you some of the things you can do in terms of social cognition research in an online environment.
So there were three areas in which we found in our experience over the last couple of years that have worked rather well, in terms of transitioning online. So first thing was that we gamified. So we turned the basic semantic interference paradigm, the PWI paradigm into something called the picture word game. Okay, and here’s, sorry about that, is that gonna work, yep. So we have this idea, you come into the game, you connect to some service, you login. And this was all possible in the in the Gorilla environment. And then we just tried to be a little bit creative, getting them the sense that there were some sort of player participant that came online into a game to play this picture working, and that they would be playing potentially with someone else. So this was the second thing, we had some social presence when we tried to get them to meet a Confederate online. And we had these social pixels, where they had an avatar in the top right hand corner, that would represent the other party. So here, we pretended that they were finding a word master, they’d find someone so that other person will come online, they would come along, we were able to turn the video camera on the other person, they would meet someone.
So there was also creating this normal social context, obviously, all not real. But we got try to try to deceive our participants into believing this. And then another thing we tried to do, which worked very well was have the shared workspaces where we pretended that they were actually going to share their screen. So we had this pretend shared screen area where they gave permissions. And then they accepted permissions. And then they had this shared display. So they believed that they were working on the same screen, they were sharing their screens, here, they also typed in their name, they logged on, they had their name come in the top left hand corner, and then they saw their partner type on screen to give that sense that they’re also on the shared screen, and then their their name will come in the top right hand corner.
So some of this, I think I presented earlier versions of this the first time, this is the more refined version. And this has worked remarkably well for us, these three pillars have worked well. And we’ve found as you’ll see some very interesting results using this manipulation for the minimal social context online.
Okay, so what does that all culminate in? Well, this is what it looks like the actual paradigm itself, you’ve got the picture word interference paradigm, the stimuli in the middle where you’ve got pictures and words on top half, of them are related half unrelated, as I say, manipulation here is the relatedness slows you down. And then you you’ve got this framing this social frame, you’re on shared screen, you believe someone else is there, you believe you’re working with someone. So what we were doing was comparing whether we would get the same effect in the lab online, if you believe that there was someone there player 2, here Chris, or not someone that no one online playing with you.
Okay, so this is what we found. On the left hand side, I’ve got the lab result, the 42 milliseconds reduce, negated to minus one millisecond. And what we found is we replicated the no partner condition where they weren’t paired with someone, this is the online result, we got a nice 33 milliseconds PWI semantic interference effect, slightly slower on the related word, and once again, we found that that was removed when you believe that you were connected with someone online. So we were very, very encouraged by this. We were a little bit dubious where we would create that minimal context in an online environment there’s lots of distractions and so forth. But actually, we found a lot of people found that very believable, and they very much commented afterwards saying that they really felt say hi to that person would very much you enjoy playing with that person, so forth. And they really got through the gamification, the idea of what we were trying to get out what the task was about. And we were able to replicate the result with great consistency.
So what did we think? So I’m going to go through some of the manipulations that we do. Well, we thought we could play around with this a little bit, I’m just gonna go through this little bit like a whistlestop tour of some of the things that we’ve done a bit of sort of smorgasbord of things that we’ve tried in terms of manipulating the social context.
So this is what it looks like when you meet someone. This is me meeting Phillip, who’s the Confederate that has been the Confederate in about 1000 of the participants that have collected in my studies, he’s been great. He’s famous, but famously underpaid. And what we thought is what we could change who that person is, right? So one of the things that we thought, well, one of the hypotheses was how much we’ve had from previous studies, whether social status affects the degree to which you are able to offload that information. So the idea here is that you’re de prioritising that that interfering information, when it’s looked after by someone else, this gives us idea that you’re offloading that information someone else, would you offload that information to someone else, if you believed that they were of higher or lower social status, perhaps if you trusted them more than more because they’re of high status, they’re more likely to offload than you were someone of low status. How sophisticated is this effect? To what degree can this high level social constructs interact with these low level automatic mechanisms?
So what we did this was actually did with a series of third years, which got third years to record themselves both in low status gear like dress down, messy background, students sort of hovel sort of thing, and then dressed up in high status, with with smart backgrounds and so forth. And we’re trying to look, how does this affect, just the image, effect the degree to which you are going to do this social offloading that we thought this effect we were sort of investigating, and this is the base online results again, so this was just with a partner online, not a partner offline, 33, minus 9.
And what we found is this, with the low status, you get this interference effect of 37 milliseconds and with the high status that’s removed. And this is fascinating, even more so because in both conditions here, both conditions are social. So you’re both meeting someone, the only thing that’s different is the way you’ve construed that person. So in one status you believe they are low status and in the other one you believe they are high status, and the low status is disrupting the offloading of that information to someone else such that you’re bringing back that interference, you’re processing that information again, perhaps you’re not trusting them with that information.
Here’s another manipulation that we ran well, we thought, well, social status comes in different forms prestige, but also competency. Is it the fact that you believe that the other person is competent enough to look after that information. So here using Gorilla, again, we try to manipulate the degree to which you believe they were good at the task your partner. So we either told them that they were the 148th ranked Word master, so that was the name of the person that was your partner in this game, or the 25th ranked so the top 15% of all word masters. Does a high competency person or a low competency person change the degree to which you process that information.
And again, this is the basic online result always on the left. And what we found is yes, in the low competency, you get this standard interference effect here a little bit larger, it 51 milliseconds a strong interference effect, but in the high competence position that disappears. So when you believe just that you’re working with someone of high competence, again it’s both conditions are social, you are able to remove the effect of processing that word information, you are no longer processing that word information in the same way. This is social context shaping information processing.
This is one one of my favourites that we ran, we were thinking about other ways to increase pro sociality. Right, because if this is about cooperation, or trust, and so forth, well, there’s a whole area of literature on terms of synchrony, behavioural synchrony, if you can increase behavioural synchrony between two people, you increase the likelihood of the prosociality subsequently. So here what we did, we had a bit of time, bit of fun trying to work out how to do this, on the top level here is the computer automatically giving a response of a an avatar left and right, okay, and the bottom that is an actual participant responding, so they could press their buttons on the keys, and their avatar would come up spontaneously on the screen.
And what we told them is that your partner is pressing, and you just have to match them. Okay. But in one condition, we had it really easy they were doing ‘bup bup bup bup bup’ but on the other condition, Chris was a bit annoying, and he was all over the place very, very unsynchronized. What we found is that when you perceive Chris to be asynchronous, you had a low self other overlap with one of the ways of measuring affiliation. And when you perceive him to be synchronous, you have a very high level of self other overlap.
But what we’re interested in does this translate into the the offloading effect that I’ve talked about so far, again, the basic online result, and then yes, we could find this as well, right? So in the low synchrony condition, you get the standard interference effect, but in the highest synchrony condition that is removed again. So when you believe that you’ve worked with someone that is in sync with you, you are more likely to not process that information that word information that we’re saying you’ve offloaded to the other person, and therefore, it doesn’t cost you in the same way as it would if you work with a low synchronous person.
Okay, final one I just want to talk about when recently this was actually running in a student lab, we were interested in group membership. So another one of those big other areas in social psychology that we wanted to connect. If we try to group people on the basis of whether they saw a spinning ballerina, this is an illusion. So we have this on Gorilla. Watch this spin around, do you see that moving clockwise or anti clockwise? If you saw it moving clockwise, I think I remember we called you the lefties, you saw it moving anti clockwise, or was the other way around, the anti clockwise the lefties and the clockwise the righties. So we tried to put you into two minimal groups.
This is you know, if you want to look at this good work by Steve Riker talking about minimal group formation, okay. Now does forming minimal groups, in groups and out groups then affect the degree to which you process that other information. And this result actually came out slightly differently than what we expected, we expected that you would offload your in group and not offload to your outgroup. But we found completely the opposite. We found that actually, in your in group, you’ve got this offloading effect, that when you work with an out group member, you remove that interference effect in line with the other. Now I could go on and talk about a little bit some of the ideas why we think that is, but actually just the thing that I want to emphasise in this talk is the remarkable consistency of our online data. And that we were able to take this basic effect, we’re able to then use these very, very global high level complex social constructs and implement them in an online environment and get these very, very low level precise effects on a consistent basis, highly encouraging, we’re consistently able to turn off semantic interference on the basis of the context in which you find yourself in.
So how am I doing for time here, I’ve got just 30 seconds to wrap up. So in terms of theory, well, this is all around this idea for us that cognition is embedded within the social world, there’s some benefits come to face, these implicit attunements, we’re sort of talking about these implicit automaticities being tuned into the social world, they’re not black boxed off, they are sensitive to the constructs we find around us. And this may be a skill based thing, that you’re implicitly implicitly leveraging others to regulate the way that you process information. And we want to talk to social offloading in the skill of being best together.
But I’m just going to rush through that slightly, because I just want to finish by talking about some of the insights that I’ve found in terms of working online. And I keep on getting this thing in the way so I can’t read my slide. So the transition online for has has proved very smooth. And we’ve generated consistent data for us from lots of studies. That’s one thing that hopefully, I’ve shown you. And we want to obviously say that online behavioural data has the sensitivity to pick up on these minimal social effects. Okay, so we were worried that that might not be the case, obviously, these online environments are a little bit more noisy. But actually, what we found is they do translate through and we were finding this consistent effect. We have also received consistent high levels of believability using the framework that I’ve showed you. And we’ve excluded participants who did not believe but we’re entering sort of the mid 90s, in terms of believability. And we’ve not actually one analysis that I wanted to and I haven’t been able to get around to doing is looking at that 5% To see whether we’re not getting the effect in that last 5% that we are in the in the major 95%.
One big thing that I found that has been a big improvement is gamification. That’s been a huge step change in terms of maintaining engagement and the understanding of the task. We do have attention checks throughout our paradigm, we ask people to perform the task to get over a certain level, then they pass and they can go through to the actual main task. But actually, we found that dropout rate significantly dropped once we actually achieved this gamified environment where they believe that where they were playing a game, and they were actually enjoying being a part of this environment in which we were trying to create. So I think I might well stop there because actually that that time went much quicker than I thought thanks to Professor Daniel Richardson, collaborator, Gorilla. Thanks, Jo, Nick and the team and the students for sort of collecting data with me.
Miles, that’s extraordinary.
And I I’ve never had a talk on language followed by talk on social cognition, this semantic Stroop Task essentially interference. And it just, you know, you’ve contextualised so much of what Neil was talking about, because of course, when you’re talking language to another person who’s there by dialect or like you, are you doing this social offloading, because they understand this as well or not, right? Like, we can’t think about language just in the situation where you’re just one person who’s speaking because the whole purpose of language is that it sits within a social context. So you’ve made me have completely new thoughts that I haven’t had before. I’m sure other people have had these thoughts.
Miles Tufft 19:47
No no absolutely, it was a great follow on from Neil. Yes. Yeah.
What a, what a happy marriage. One question that came to me, there are some questions in the q&a that we’ll come to and anybody else who’s online now has a question for them, Miles, dump them into q&a. And miles, if people have other questions, are they welcome to reach out to you?
So people can reach out to you. How do you do all those animations? Because like, obviously, we now actually have multiplayer in Gorilla. So you could actually do this as a multiplayer experiment. But you didn’t, you did a fake multiplayer experiment. And you made it look like a game with all these fancy animations. How did you make those,
Miles Tufft 20:25
so a lot of making of gifts. We created a lot of gifs, a lot of which we then you either created in PowerPoint or some other form of PNG, and then we sort of imported them into the Gorilla platform, and so forth. But obviously, one of the things that we’re very excited about is multiplayer. And so that’s one of the things we want to try and automate that a little bit more, and bring it into that environment and so forth. But this was a lot of this stuff was done about sort of a year and a half ago, when we were first starting out and trying to develop that. So we had to be innovative and creative in terms of the ways in which we could create these environments. Yes.
Jo Evershed 20:59
So one top tip for people, if you want to make your tasks more gamelike it really is just as simple as changing your images to GIFs, putting a bit of movement in them creating a bit of a fiction, and even in the task builder, you can create some game like experiences. Let me go to the q&a. Oh, no, no open questions at the moment. There’s a question in the chat, though. Do you measure social influence or need for uniqueness of your respondents?
Do we measure social need for uniqueness
measure social influence, or the need for uniqueness of your respondents? Maybe these moderate your findings. Dr. Mostanza can you say more about this is this is the idea that your part- that if the participants feel that they’re similar to each other, that they are an in, like one in group? Whereas if they’re more different to each other, they’re more different? Is that the idea of getting at? We’ll see? Yes, yes, that’s the question. So do you look at how different the participant perceives themselves to be to the confederate?
Miles Tufft 22:04
That’s a very nice question. And it’s actually one of the areas that we were looking into degree of social closeness and social similarity. That’s where we sort of went round the membership stuff on a sort of more group level. But yes, definitely, that’s something that would be an interesting way to get at, again, the way that the social world modulates. I think the key point here is that we can’t think of our brains or cognition as separate from the world, we have to think of them as embedded, and therefore built from the bottom up in the world, and therefore shaped by the things that we encounter in this world, one of which could well be social influence. And so social distance and social overlap. And one of those things that we would expect may be shaped the way the information is prioritised or deprioritised. I never talked about mechanism as well. And so mechanism here may be well attention, right. So selective attention the way that we attend to things or downregulate attention. That could be the way that this has been mediated. Brilliant.
Jo Evershed 22:55
Miles. Thank you, you might, Mostanza might want to chat with you further in the chat. We’re now going to go to our next speaker