BBC Learnings · Section 9

Fake smiles and the computers that can spot them

It's good to see you again, Sam. Really? Yes, of course.

Level: B2 · 125 sentences

Transcript

It's good to see you again, Sam.
Really? Yes, of course.
Can't you tell by the way I'm smiling?
Well, I find it difficult to tell if someone is really smiling
or if it's a fake smile.
Well, that's a coincidence because this programme is all
about how computers may be able to tell real smiles
from fake smiles better than humans can.
Before we get into that though, a question.
The expressions we can make with our face are controlled by muscles.
How many muscles do we have in our face?
Is it a 26, b 43 or c 62?
What do you think, Sam? No idea,
but a lot I'd guess, so I'm going with 62.
OK. Well, we'll see if you'll be smiling
or crying later in the programme.
Hasan Ugail is a professor of visual computing at the University of Bradford.
He's been working on getting computers
to be able to recognise human emotions
from the expressions on our face.
Here he is speaking on the BBC Inside Science radio programme.
How successful does he say they have been?
We've been working quite a lot on the human emotions,
so the idea is how the facial muscle
movement, which is reflected on the face,
through obviously computer, through video frames,
and try and understand how these muscle
movements actually relate to facial expressions.
And then from facial expressions, try
and understand the emotions or to infer the emotions.
And we have been quite successful in doing that.
We have software that can actually
look at somebody's face in real time
and then identify the series of emotions that person
is expressing in real time as well.
So, have they been successful in getting computers to identify emotions?
Yes, he says they've been quite successful.
And what's interesting is that he says
that the computers can do it in real time.
This means that there's no delay.
They don't have to stop and analyse
the data or crunch the numbers.
They can do it as the person is talking.
The system uses video to analyse a person's expressions
and can then infer the emotions.
To infer something means to get an understanding
of something without actually being told directly.
So you look at available information and use your understanding
and knowledge to work out the meaning.
It's a bit like being a detective, isn't it?
You look at the clues and infer what happened,
even if you don't have all the details.
Yes, and in this case the computer looks
at how the movement of muscles in the face,
or facial muscles, show different emotions.
Here's Professor Ugale again.
We've been working quite a lot on the human emotions,
so the idea is how the facial muscle
movement, which is reflected on the face,
through obviously computer, through video frames,
and try and understand how these muscle
movements actually relate to facial expressions.
And then from facial expressions, try
and understand the emotions or to infer the emotions.
And we have been quite successful in doing that.
We have software that can actually
look at somebody's face in real time
and then identify the series of emotions that person
is expressing in real time as well.
So how do the computers know
what is a real or a fake smile?
The computers have to learn that first.
Here's Professor Ugail again talking about how they do that.
We have a dataset of real smiles
and we have a dataset of fake smiles.
These real smiles are indoor smiles in a lab,
so you put somebody on a chair
and then show some funny movies.
and we expect the smiles are genuine smiles.
And similarly, we ask them to pretend they're smiling,
so these are what we call fake smiles.
So what we do is we throw these into the machine
and then the machine figures out
what are the characteristics of a real smile
and what are the characteristics of a fake smile.
So how do they get the data that the computers
use to see if your smile is fake or genuine,
which is another word which means real?
They induce real smiles in the lab by showing people funny films.
This means that they make the smiles come naturally.
They assume that the smiles while watching the funny films are genuine.
And then they ask the people to pretend
to smile and the computer programme now
has a database of real and fake smiles
and is able to figure out which is which.
Figure out means to calculate and come to an answer.
Yes, and apparently the system gets
it right 90% of the time,
which is much higher than we humans can.
Right, well, before we remind ourselves of our vocabulary,
let's get the answer to the question.
How many muscles do we have in our face?
Is it a 26, b 43 or c 62?
Sam, are you going to be smiling?
What did you say? Aha, so I thought 62.
Am I smiling, Neil? Sadly, you are not.
You're using different muscles for that sort of...
Sad look. Actually, the answer is 42.
Congratulations to anyone who got that right.
Now our vocabulary.
Yes, so facial is the adjective relating to face.
Then we had infer.
This verb means to understand something
even when you don't have all the information
and you come to this understanding
based on your experience and knowledge,
or in the case of a computer, the programming.
And these computers work in real time,
which means that there's no delay.
And they can tell a fake smile from a genuine one,
which means a real one, as the person is speaking.
They made people smile, or as the professor said,
they induced smiles by showing funny films.
And the computer is able to figure out,
or calculate, whether the smile is fake or genuine.