Popular in Psychology

Books

City Time: On Being Sentenced to Rikers Island

Independent Scholar David A. Campbell

Most Cited

The adapted mind

Dalhousie University Jerome Barkow

Experimental Psychology

Time to investigate human visual shape perception

University of Montreal Martin Arguin

Most Cited

James S. Jackson and the program for research on Black Americans

Harvard University David R. Williams

Psychology

Violence and Mental Illness: Rethinking Risk Factors and Enhancing Public Safety

Duke University Eric Elbogen

Cognitive Psychology

Cognitive remediation – where are we now and what should we do next?

King's College London Til Wykes

Four fundamental dimensions underlie the perception of human actions

View more insights in Social Sciences - Psychology - Cognitive Psychology

View more figures from University of York

View the page of Nick E Barraclough

Read the Study

Nick E Barraclough evaluates the actions of other individuals based upon a variety of movements that can reveal critical information to guide decision making and behavioural responses.

Image courtesy of interviewee. February 6, 2024

Generating Workspace

Export as PDF

This video video discusses the concept of action perception, which involves recognizing and evaluating the actions of others. It provides an example of observing someone crossing the road and highlights how we make judgments based on various traits such as speed, direction, and intention behind the action. The author introduces the idea of "conceptual spaces" — mental frameworks where we categorize and assess actions based on their characteristics.

A detailed study was conducted to explore the conceptual space underlying action perception, utilizing motion capture technology to record 240 different actions performed by actors. The findings revealed four main dimensions through exploratory factor analysis (EFA) that people use to evaluate actions:

1. **Friendliness** – Evaluating actions on a scale from positive/prosocial to negative/untrustworthy.
2. **Formalness** – Assessing perceived power and strength, ranging from dominant to weak.
3. **Plannedness** – Determining intentionality, from intentional and controlled actions to spontaneous and uncontrolled ones.
4. **Abduction** – Focusing on movements of limbs towards or away from the body.

The study found that judgments about friendliness and power are made quickly, likely due to their importance in social interactions, while evaluations of intentionality occur more slowly. The research suggests a possible overlap in the evaluation dimensions of actions and facial traits, implicating shared conceptual spaces in how we assess social information.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.000 --> 00:03.005
The action perception is the detection

00:03.465 --> 00:08.445
and the evaluation of actions executed by other humans.

00:09.065 --> 00:10.205
So for an example,

00:10.415 --> 00:12.805
let's say you are watching someone cross

00:12.885 --> 00:13.965
a road in front of you.

00:14.465 --> 00:17.485
Now we can evaluate that action on

00:18.005 --> 00:19.485
multiple different qualities.

00:19.985 --> 00:23.725
For example, we see the speed that the person is walking at,

00:24.185 --> 00:27.205
we see the direction in which they're moving.

00:27.825 --> 00:31.045
We can see that they're moving quite fluently, for example,

00:31.305 --> 00:32.885
we can recognize what they're doing.

00:32.905 --> 00:36.765
We can see it as a, as a sort of jog may perhaps, or a walk.

00:37.385 --> 00:40.485
Um, we can work out the goal of the action, for example,

00:40.485 --> 00:42.085
that they're crossing the road.

00:42.625 --> 00:46.165
Uh, we can even infer something about the individual's

00:46.165 --> 00:48.005
mental state from their actions.

00:48.145 --> 00:51.565
So you can, you can work out perhaps whether they look happy

00:51.985 --> 00:53.485
or just from the way that they move,

00:53.985 --> 00:56.525
and also also infer things like the

00:56.525 --> 00:58.165
intention of the individual.

00:58.195 --> 01:01.285
Perhaps they're crossing the road excitedly,

01:01.285 --> 01:03.525
because they've seen someone on the other side of the road

01:03.545 --> 01:04.565
and they're trying to meet them.

01:05.425 --> 01:10.165
So when we see an action executed by someone else,

01:10.265 --> 01:12.605
we, we, we make lots of judgments about,

01:12.905 --> 01:14.365
about that actions.

01:14.385 --> 01:17.725
And we can derive lots of characteristics from,

01:17.875 --> 01:19.165
from the way people move.

01:20.665 --> 01:25.405
Now, the principles underlying the way that we see

01:25.585 --> 01:28.085
and make sense of those actions, well,

01:28.085 --> 01:30.485
we think they are represented within,

01:30.485 --> 01:32.565
something called a conceptual space.

01:33.115 --> 01:36.485
Because the way that we recognize and,

01:36.545 --> 01:38.445
and form mental representations

01:38.505 --> 01:41.125
of things in our environments are within these mental

01:41.335 --> 01:45.085
frameworks, or, or conceptual or psychological spaces.

01:45.545 --> 01:48.085
So other things that we see in our environment, for example,

01:48.145 --> 01:52.365
colors or, textures

01:52.945 --> 01:54.645
or movements or faces

01:55.185 --> 01:57.925
and things that we detect with other sensors,

01:58.065 --> 02:00.085
for example, odors or,

02:00.465 --> 02:03.885
or the feeling of, roughness with our hands.

02:04.215 --> 02:05.325
These, these items

02:05.505 --> 02:07.485
or these domains are also represented

02:07.485 --> 02:08.645
within conceptual spaces.

02:09.745 --> 02:14.485
Um, and it's conceptual space describes basically,

02:14.625 --> 02:18.245
the properties of that, of that type of item.

02:19.065 --> 02:23.125
And conceptual spaces capture the similarities

02:23.185 --> 02:25.965
and the differences between different items

02:26.145 --> 02:27.245
within that domain.

02:27.705 --> 02:31.925
So the, this, this, this organization allows the

02:32.445 --> 02:33.605
classification, the naming,

02:34.185 --> 02:36.925
and the responses to that particular item.

02:37.465 --> 02:40.765
Now, within a conceptual space, things that we see

02:40.985 --> 02:42.805
and think of as particularly similar

02:42.865 --> 02:44.885
to each other are represented close to each other.

02:45.345 --> 02:48.005
And things that are, seem very different from each other,

02:48.405 --> 02:51.205
we might respond differently to them, are, are,

02:51.265 --> 02:53.285
are represented far apart within that,

02:53.345 --> 02:54.445
within that space.

02:55.425 --> 02:58.125
So we went into this project,

02:59.315 --> 03:02.375
Trying to think about, well, what is the conceptual space

03:02.375 --> 03:04.335
that underlies action perception?

03:04.395 --> 03:07.495
So we know that we perceive actions on all these different

03:07.495 --> 03:10.095
characteristics in terms of speed, direction,

03:10.285 --> 03:11.655
fluency, so on.

03:12.425 --> 03:14.515
Well, what are the important ones?

03:15.095 --> 03:18.475
Uh, what are those important factors underlying

03:18.655 --> 03:21.035
how we make sense, and,

03:21.095 --> 03:23.155
and represent the actions of other people.

03:23.915 --> 03:25.135
So because we didn't know,

03:25.285 --> 03:28.695
because we didn't know the, the space that,

03:29.095 --> 03:30.975
represents actions, we had some,

03:31.115 --> 03:34.135
we have some other evidence that suggests, that it's,

03:34.155 --> 03:36.895
actions of represented in terms of similarity.

03:37.515 --> 03:41.375
And we knew that there were perhaps certain, levels

03:41.435 --> 03:44.735
of understanding between more concrete, mid-level

03:44.835 --> 03:46.375
and perhaps more abstract levels

03:46.395 --> 03:47.855
of representation of actions.

03:48.355 --> 03:51.255
We wanted to, well just take a data-driven approach

03:51.435 --> 03:55.015
to working out, the conceptual space under action.

03:55.035 --> 03:59.615
So just let, let's, let's look at a big range of actions.

03:59.615 --> 04:01.815
There's bigger range of actions that we could get hold of,

04:02.355 --> 04:06.255
and also to get humans to evaluate those actions on

04:06.255 --> 04:08.895
as number of different characteristics as, as many

04:08.895 --> 04:12.895
as we could sort of reasonably fit into, an experiment.

04:14.325 --> 04:16.865
So what we did,

04:17.765 --> 04:22.145
in our experiment was first of all to, is first of all

04:22.485 --> 04:25.265
to capture this large range of different actions.

04:25.605 --> 04:28.065
So we used, a technique known as motion capture.

04:28.565 --> 04:30.665
So we had actors wearing suits

04:30.695 --> 04:32.545
that were wearing motion capture suits,

04:32.545 --> 04:34.545
which essentially have little,

04:34.905 --> 04:36.545
detectors on bits of the body.

04:36.545 --> 04:40.545
There's 32 separate detectors on, on the head, shoulders,

04:40.675 --> 04:43.745
knees, toes, fingers, elbows, and so on.

04:43.745 --> 04:47.865
So we can capture the movements of those actors within,

04:47.925 --> 04:50.065
the three, three dimensional environment.

04:50.965 --> 04:52.065
And we asked actors.

04:52.245 --> 04:55.785
We had, four different actors, two male, two female,

04:55.885 --> 04:57.825
and they performed a number of different actions,

04:57.925 --> 04:59.625
240 different actions.

05:00.105 --> 05:02.585
Everyone performed every action three times.

05:03.605 --> 05:07.545
Uh, and then we selected the best, most representative

05:08.175 --> 05:11.105
actions across all these different actors.

05:11.205 --> 05:14.585
And these actions range from things like, walking,

05:15.345 --> 05:20.105
standing, cutting, throwing, catching, petting a dog,

05:20.725 --> 05:24.835
just a whole range of different action as,

05:24.855 --> 05:28.195
as many things as we could, realistically

05:28.835 --> 05:30.195
actually capture the certain types

05:30.195 --> 05:31.395
of actions we couldn't capture.

05:32.015 --> 05:34.555
For example, swimming, was not going to work

05:34.555 --> 05:36.275
with our motion capture suits, for example.

05:36.815 --> 05:38.715
Uh, but, but in general, it's just

05:38.715 --> 05:40.355
as many different actions as possible.

05:40.705 --> 05:44.155
Then we took that data that that motion capture data

05:44.415 --> 05:45.835
to animate an avatar.

05:47.015 --> 05:50.275
So, this avatar was a, a three-dimensional,

05:50.275 --> 05:51.875
we call it a volumetric puppet.

05:52.025 --> 05:54.235
It's a three-dimensional a avatar.

05:54.775 --> 05:58.025
It doesn't have a face, doesn't have any clothes on,

05:58.085 --> 06:01.705
but it has the, the shape of the human and it's gray.

06:02.085 --> 06:05.385
And then the reason why we, we had an avatar like this was

06:05.385 --> 06:09.065
to exclude information from other, other types

06:09.085 --> 06:12.345
of body information so we can judge people on their faces,

06:12.375 --> 06:15.065
perhaps on what they're wearing, you know, on the size

06:15.065 --> 06:16.225
of the body and so on.

06:16.605 --> 06:18.945
Uh, we didn't want that information to sort

06:18.945 --> 06:20.425
of confound what we're looking at.

06:20.825 --> 06:23.185
Actually, the, the information we're really interested in

06:23.185 --> 06:24.825
about the actions is the posture.

06:25.045 --> 06:26.745
So that's the position of the body

06:27.165 --> 06:29.865
and also the way that the actor actually moves.

06:30.365 --> 06:34.225
So we use these, this motion capture data

06:34.645 --> 06:35.945
to animate the avatar,

06:36.005 --> 06:39.625
and then we've got 230 people to rate those,

06:39.805 --> 06:42.185
all those different actions on a range

06:42.205 --> 06:44.585
of different characteristics from sort of how,

06:44.885 --> 06:47.945
how fast they're moving, whether they look sociable,

06:47.945 --> 06:50.065
whether they, look trustworthy,

06:50.255 --> 06:52.345
whether they look strong and so on.

06:53.805 --> 06:58.505
And so with that data, we then use the technique known as,

06:58.845 --> 07:01.905
exploratory factor analysis or EFA.

07:02.285 --> 07:05.545
Now, EFA essentially is a data reduction technique.

07:05.545 --> 07:07.305
It takes this data and looks for the

07:07.835 --> 07:09.625
underlying latent factors

07:09.925 --> 07:13.305
or these sort of explana, the, these dimensions

07:13.305 --> 07:15.705
that underlie these, these ratings

07:15.705 --> 07:16.865
of these different actions.

07:17.695 --> 07:22.575
What we found with the, the EFA was that there were

07:23.325 --> 07:27.375
four principle dimensions that underlie the way that people

07:28.775 --> 07:30.615
evaluate the actions of other people.

07:31.355 --> 07:34.455
Now, two of those dimensions were particularly important

07:34.555 --> 07:36.535
and explained a lot of their behavior,

07:36.795 --> 07:40.615
and two, dimensions a little less important, but,

07:40.795 --> 07:42.535
but equally, they were still important.

07:43.275 --> 07:45.455
Uh, so the first two dimensions, which we,

07:45.475 --> 07:48.615
we think are the most important, was the evaluation

07:48.615 --> 07:50.535
of actions on friendliness.

07:50.915 --> 07:54.335
So wherever an action is happy

07:54.795 --> 07:59.295
or seen as trustworthy, or positive or pro-social

07:59.635 --> 08:02.175
and in contrast, perhaps as untrustworthy

08:02.755 --> 08:06.575
or angry, at the other end of this dimension.

08:06.575 --> 08:08.895
So this is a sort of valence dimension.

08:08.915 --> 08:10.095
We called it friendliness,

08:10.955 --> 08:14.655
but it's essentially, it's a, an evaluation of, of valence.

08:15.115 --> 08:19.055
Uh, now, the second dimension that was also important was

08:19.055 --> 08:20.775
that we called it formalness.

08:21.195 --> 08:22.255
Um, but this,

08:22.525 --> 08:25.655
this latent factor really underlies our perceptions

08:25.835 --> 08:29.815
of power, of speed, of strength,

08:30.355 --> 08:32.775
of whether someone is dominant or not.

08:33.235 --> 08:35.895
Uh, and on the other end of that dimension, of course,

08:35.915 --> 08:38.455
is sort of more weaker, and,

08:38.595 --> 08:41.215
and more, more fine types of actions.

08:41.795 --> 08:43.935
So these two, these two dimensions seem

08:43.935 --> 08:45.855
to be the two most important ones,

08:46.195 --> 08:47.615
but there were two other dimensions

08:47.615 --> 08:51.415
that were also pretty important as well, explained, a, a,

08:51.695 --> 08:54.975
a degree of the variance, one we call planned,

08:55.145 --> 08:57.565
But essentially that, that

08:57.635 --> 09:01.405
that explains the intentionality of the action.

09:01.505 --> 09:05.165
So at one end, we'd have actions that were particularly,

09:05.435 --> 09:08.405
intentional or planned or controlled

09:08.825 --> 09:10.645
or precise and deliberate.

09:11.265 --> 09:14.365
And on the other end of the dimension, we had actions

09:14.515 --> 09:19.485
that are spontaneous or accidental,

09:19.705 --> 09:21.805
or uncontrolled,

09:21.945 --> 09:25.405
and of course actions that lay somewhere in between the,

09:25.405 --> 09:27.685
these two, extremes.

09:27.985 --> 09:31.245
And our fourth dimension, there's one we call abducting,

09:31.415 --> 09:34.205
which is essentially, it's quite a, it's quite a sort

09:34.205 --> 09:36.725
of movement focused, dimension.

09:36.785 --> 09:38.925
So abducting is the movement

09:38.925 --> 09:41.085
of limbs away from the body like this.

09:41.465 --> 09:44.685
And abducting is moving, limbs towards,

09:45.185 --> 09:46.485
towards the body.

09:47.185 --> 09:50.045
So actions that involved in pushing things away,

09:50.585 --> 09:51.805
or expelling things

09:52.105 --> 09:55.285
or throwing things would be at one end of the dimension.

09:55.585 --> 09:58.645
And at the other end would be, actions

09:58.865 --> 10:01.965
of people picking things up or bringing things towards them,

10:02.185 --> 10:04.085
or hugging them or drinking

10:04.225 --> 10:06.325
or bringing things in towards the body.

10:06.745 --> 10:10.805
So these are the four main dimensions that we find,

10:11.025 --> 10:13.365
within our, our model

10:13.425 --> 10:15.965
and the representation of, of actions.

10:16.785 --> 10:21.565
We didn't set out to look at how fast people are at, at,

10:21.585 --> 10:23.965
at making these judgements about other actions.

10:24.065 --> 10:27.085
We were actually interested in the ratings of them and the,

10:27.585 --> 10:28.845
and this conceptual space.

10:29.225 --> 10:32.845
But what fell out, fell, fell out of our data was

10:32.845 --> 10:36.085
that people actually make judgments of,

10:36.795 --> 10:40.165
whether someone is pro-social or happy or trustworthy.

10:40.265 --> 10:41.845
We make that judgment the quickest.

10:42.105 --> 10:43.965
It was the, in all our data,

10:43.965 --> 10:45.445
people made this judgment quick.

10:46.225 --> 10:50.685
Um, and we think that well, actually,

10:50.705 --> 10:52.445
if you are trying to work out

10:52.555 --> 10:54.685
what someone in your environment is doing,

10:54.955 --> 10:57.805
this is probably one of the most important things

10:57.805 --> 11:00.125
that we need to be able to make a judgment of.

11:00.425 --> 11:01.805
You know, are they a friend?

11:02.545 --> 11:05.525
Are they being difficult to, they potentially a foe?

11:05.905 --> 11:08.805
Uh, these are, these are important judgements for us

11:08.985 --> 11:12.125
as humans to decide, well, should I interact

11:12.125 --> 11:14.925
with this person or should, perhaps I should avoid them.

11:15.305 --> 11:18.805
So making that decision early on and quick is important.

11:18.965 --> 11:21.485
I mean, okay, maybe may, maybe we don't have to.

11:22.105 --> 11:25.965
Um, it might not save our life at this point in time,

11:26.145 --> 11:27.445
but,

11:27.545 --> 11:30.125
but it is an important judgment even in the modern world

11:30.665 --> 11:34.765
Now, secondly, we also see that judgments of power

11:34.865 --> 11:37.965
and strength and speed are also quite quick, just not

11:37.965 --> 11:40.045
as quick as friendness, but, and nearly as quick.

11:40.665 --> 11:42.045
Um, and,

11:42.105 --> 11:44.925
and that's also an important judgment, determining,

11:45.105 --> 11:47.285
you know, okay, someone is, is a threat.

11:47.555 --> 11:51.495
Well, could they possibly carry out

11:51.495 --> 11:52.815
that threat? Um,

11:53.235 --> 11:54.245
Have they got the strength

11:54.305 --> 11:56.845
or that power in order to be able to carry that strength?

11:56.845 --> 11:58.285
That's also an important thing.

11:58.515 --> 12:00.645
Whereas we found that judgments

12:00.645 --> 12:02.965
of things like intentionality, whether it's a sort

12:02.965 --> 12:05.405
of deliberate or whether it's controlled

12:05.425 --> 12:09.365
or planned action, that, that, that tends to, we, we make

12:09.365 --> 12:10.925
that judgment later on in time

12:11.195 --> 12:12.845
because it's not so important.

12:13.265 --> 12:15.605
Um, you know, you can, you, you can work

12:15.605 --> 12:17.085
that out a little bit later.

12:17.465 --> 12:18.845
You know, first of all, you need

12:18.845 --> 12:21.005
to decide whether someone is,

12:21.225 --> 12:22.965
is performing an action in,

12:22.985 --> 12:24.925
in a friendly prosocial way or not.

12:25.195 --> 12:27.005
What our findings show is these two

12:27.795 --> 12:30.845
principle dimensions in two additional dimensions.

12:31.185 --> 12:34.965
Now, the two primary dimensions, they look very similar to

12:35.785 --> 12:39.405
two of the primary dimensions underlying the evaluation

12:39.405 --> 12:41.085
of f face traits.

12:41.825 --> 12:45.765
So other people have found that we evaluate faces on,

12:45.765 --> 12:48.005
trustworthiness and dominance,

12:48.065 --> 12:49.925
and possibly a third dimension as well.

12:50.505 --> 12:52.845
Now, the, the face trait dimension

12:53.785 --> 12:57.525
of trustworthiness looks quite similar to our,

12:57.525 --> 12:58.845
friendliness dimension.

12:59.025 --> 13:01.405
And the face trait dimension of dominance

13:01.955 --> 13:04.765
also looks pretty similar to our,

13:04.765 --> 13:05.925
dimension of formidable.

13:06.505 --> 13:11.125
Now, this suggests that, well, maybe, maybe there is a cross

13:11.645 --> 13:16.125
category, conceptual spaces that,

13:16.175 --> 13:19.565
conceptual space that involves not only the evaluation

13:19.565 --> 13:21.645
of the actions and the behaviors of other people,

13:21.745 --> 13:24.805
but information you can get from the tr from the face.

13:25.475 --> 13:28.005
Whereas the other two dimensions that we show,

13:28.385 --> 13:30.045
the intentionality dimension

13:30.065 --> 13:33.125
and the abduction dimension, they are very actiony.

13:33.585 --> 13:37.285
The, the, there's no way that another stimulus like a face

13:37.465 --> 13:41.325
or a color or a smell could be intentional, yet

13:42.975 --> 13:47.675
intent of other people is derived from the way

13:47.705 --> 13:50.795
that they move and they act, within their environment.

13:50.815 --> 13:54.435
And of course, abducting this movement of the body away,

13:54.435 --> 13:56.555
parts of the body away from the, from the core

13:56.615 --> 13:57.755
or towards the core.

13:58.105 --> 14:01.955
That of course, also has to involve movement information,

14:02.055 --> 14:04.275
and therefore is also a, a,

14:04.395 --> 14:06.435
a specific action dimension, we think.

14:06.455 --> 14:08.915
And so we think those two are very actiony,

14:09.225 --> 14:13.715
whereas the other two, may,

14:13.895 --> 14:17.955
may represent some part of a, a, of a cross category,

14:18.005 --> 14:20.195
conceptual space involving other types

14:20.215 --> 14:21.235
of social information.

00:00 - The action perception is the detection and the evaluation of actions executed by other humans. So for an example, let's say you are watching someone cross a road in front of you. Now we can evaluate that action on

00:18 - multiple different qualities. For example, we see the speed that the person is walking at, we see the direction in which they're moving. We can see that they're moving quite fluently, for example, we can recognize what they're doing. We can see it as a, as a sort of jog may perhaps, or a walk. we can work out the goal of the action, for example, that they're crossing the road. we can even infer something about the individual's mental state from their actions.

00:48 - So you can, you can work out perhaps whether they look happy or just from the way that they move, and also also infer things like the intention of the individual. Perhaps they're crossing the road excitedly, because they've seen someone on the other side of the road and they're trying to meet them. So when we see an action executed by someone else, we, we, we make lots of judgments about, about that actions.

01:14 - And we can derive lots of characteristics from, from the way people move. Now, the principles underlying the way that we see and make sense of those actions, well, we think they are represented within, something called a conceptual space. Because the way that we recognize and, and form mental representations of things in our environments are within these mental frameworks, or, or conceptual or psychological spaces.

01:45 - So other things that we see in our environment, for example, colors or, textures or movements or faces and things that we detect with other sensors, for example, odors or, or the feeling of, roughness with our hands. These, these items or these domains are also represented within conceptual spaces. and it's conceptual space describes basically,

02:14 - the properties of that, of that type of item. And conceptual spaces capture the similarities and the differences between different items within that domain. So the, this, this, this organization allows the classification, the naming, and the responses to that particular item. Now, within a conceptual space, things that we see and think of as particularly similar to each other are represented close to each other.

02:45 - And things that are, seem very different from each other, we might respond differently to them, are, are, are represented far apart within that, within that space. So we went into this project, Trying to think about, well, what is the conceptual space that underlies action perception? So we know that we perceive actions on all these different characteristics in terms of speed, direction, fluency, so on.

03:12 - Well, what are the important ones? what are those important factors underlying how we make sense, and, and represent the actions of other people. So because we didn't know, because we didn't know the, the space that, represents actions, we had some, we have some other evidence that suggests, that it's, actions of represented in terms of similarity. And we knew that there were perhaps certain, levels

03:41 - of understanding between more concrete, mid-level and perhaps more abstract levels of representation of actions. We wanted to, well just take a data-driven approach to working out, the conceptual space under action. So just let, let's, let's look at a big range of actions. There's bigger range of actions that we could get hold of, and also to get humans to evaluate those actions on as number of different characteristics as, as many as we could sort of reasonably fit into, an experiment.

04:14 - So what we did, in our experiment was first of all to, is first of all to capture this large range of different actions. So we used, a technique known as motion capture. So we had actors wearing suits that were wearing motion capture suits, which essentially have little, detectors on bits of the body. There's 32 separate detectors on, on the head, shoulders, knees, toes, fingers, elbows, and so on.

04:43 - So we can capture the movements of those actors within, the three, three dimensional environment. And we asked actors. We had, four different actors, two male, two female, and they performed a number of different actions, 240 different actions. Everyone performed every action three times. and then we selected the best, most representative actions across all these different actors. And these actions range from things like, walking,

05:15 - standing, cutting, throwing, catching, petting a dog, just a whole range of different action as, as many things as we could, realistically actually capture the certain types of actions we couldn't capture. For example, swimming, was not going to work with our motion capture suits, for example. but, but in general, it's just as many different actions as possible. Then we took that data that that motion capture data

05:44 - to animate an avatar. So, this avatar was a, a three-dimensional, we call it a volumetric puppet. It's a three-dimensional a avatar. It doesn't have a face, doesn't have any clothes on, but it has the, the shape of the human and it's gray. And then the reason why we, we had an avatar like this was to exclude information from other, other types of body information so we can judge people on their faces, perhaps on what they're wearing, you know, on the size

06:15 - of the body and so on. we didn't want that information to sort of confound what we're looking at. Actually, the, the information we're really interested in about the actions is the posture. So that's the position of the body and also the way that the actor actually moves. So we use these, this motion capture data to animate the avatar, and then we've got 230 people to rate those,

06:39 - all those different actions on a range of different characteristics from sort of how, how fast they're moving, whether they look sociable, whether they, look trustworthy, whether they look strong and so on. And so with that data, we then use the technique known as, exploratory factor analysis or EFA. Now, EFA essentially is a data reduction technique. It takes this data and looks for the underlying latent factors

07:09 - or these sort of explana, the, these dimensions that underlie these, these ratings of these different actions. What we found with the, the EFA was that there were four principle dimensions that underlie the way that people evaluate the actions of other people. Now, two of those dimensions were particularly important and explained a lot of their behavior, and two, dimensions a little less important, but, but equally, they were still important.

07:43 - so the first two dimensions, which we, we think are the most important, was the evaluation of actions on friendliness. So wherever an action is happy or seen as trustworthy, or positive or pro-social and in contrast, perhaps as untrustworthy or angry, at the other end of this dimension. So this is a sort of valence dimension. We called it friendliness, but it's essentially, it's a, an evaluation of, of valence.

08:15 - now, the second dimension that was also important was that we called it formalness. but this, this latent factor really underlies our perceptions of power, of speed, of strength, of whether someone is dominant or not. and on the other end of that dimension, of course, is sort of more weaker, and, and more, more fine types of actions. So these two, these two dimensions seem

08:43 - to be the two most important ones, but there were two other dimensions that were also pretty important as well, explained, a, a, a degree of the variance, one we call planned, But essentially that, that that explains the intentionality of the action. So at one end, we'd have actions that were particularly, intentional or planned or controlled or precise and deliberate. And on the other end of the dimension, we had actions

09:14 - that are spontaneous or accidental, or uncontrolled, and of course actions that lay somewhere in between the, these two, extremes. And our fourth dimension, there's one we call abducting, which is essentially, it's quite a, it's quite a sort of movement focused, dimension. So abducting is the movement of limbs away from the body like this. And abducting is moving, limbs towards,

09:45 - towards the body. So actions that involved in pushing things away, or expelling things or throwing things would be at one end of the dimension. And at the other end would be, actions of people picking things up or bringing things towards them, or hugging them or drinking or bringing things in towards the body. So these are the four main dimensions that we find, within our, our model

10:13 - and the representation of, of actions. We didn't set out to look at how fast people are at, at, at making these judgements about other actions. We were actually interested in the ratings of them and the, and this conceptual space. But what fell out, fell, fell out of our data was that people actually make judgments of, whether someone is pro-social or happy or trustworthy. We make that judgment the quickest. It was the, in all our data,

10:43 - people made this judgment quick. and we think that well, actually, if you are trying to work out what someone in your environment is doing, this is probably one of the most important things that we need to be able to make a judgment of. You know, are they a friend? Are they being difficult to, they potentially a foe? these are, these are important judgements for us as humans to decide, well, should I interact

11:12 - with this person or should, perhaps I should avoid them. So making that decision early on and quick is important. I mean, okay, maybe may, maybe we don't have to. it might not save our life at this point in time, but, but it is an important judgment even in the modern world Now, secondly, we also see that judgments of power and strength and speed are also quite quick, just not as quick as friendness, but, and nearly as quick. and,

11:42 - and that's also an important judgment, determining, you know, okay, someone is, is a threat. Well, could they possibly carry out that threat? Have they got the strength or that power in order to be able to carry that strength? That's also an important thing. Whereas we found that judgments of things like intentionality, whether it's a sort of deliberate or whether it's controlled

12:05 - or planned action, that, that, that tends to, we, we make that judgment later on in time because it's not so important. you know, you can, you, you can work that out a little bit later. You know, first of all, you need to decide whether someone is, is performing an action in, in a friendly prosocial way or not. What our findings show is these two

12:27 - principle dimensions in two additional dimensions. Now, the two primary dimensions, they look very similar to two of the primary dimensions underlying the evaluation of f face traits. So other people have found that we evaluate faces on, trustworthiness and dominance, and possibly a third dimension as well. Now, the, the face trait dimension of trustworthiness looks quite similar to our, friendliness dimension.

12:59 - And the face trait dimension of dominance also looks pretty similar to our, dimension of formidable. Now, this suggests that, well, maybe, maybe there is a cross category, conceptual spaces that, conceptual space that involves not only the evaluation of the actions and the behaviors of other people, but information you can get from the tr from the face. Whereas the other two dimensions that we show, the intentionality dimension

13:30 - and the abduction dimension, they are very actiony. The, the, there's no way that another stimulus like a face or a color or a smell could be intentional, yet intent of other people is derived from the way that they move and they act, within their environment. And of course, abducting this movement of the body away, parts of the body away from the, from the core or towards the core. That of course, also has to involve movement information, and therefore is also a, a,

14:04 - a specific action dimension, we think. And so we think those two are very actiony, whereas the other two, may, may represent some part of a, a, of a cross category, conceptual space involving other types of social information.

Faculti AI can make mistakes