Can machines read your emotions? - Kostas Karpouzis
-
0:07 - 0:12With every year, machines surpass humans
in more and more activities -
0:12 - 0:15we once thought only we were capable of.
-
0:15 - 0:18Today's computers can beat us
in complex board games, -
0:18 - 0:21transcribe speech in dozens of languages,
-
0:21 - 0:25and instantly identify almost any object.
-
0:25 - 0:27But the robots of tomorrow may go futher
-
0:27 - 0:30by learning to figure out
what we're feeling. -
0:30 - 0:32And why does that matter?
-
0:32 - 0:35Because if machines
and the people who run them -
0:35 - 0:37can accurately read our emotional states,
-
0:37 - 0:40they may be able to assist us
or manipulate us -
0:40 - 0:43at unprecedented scales.
-
0:43 - 0:45But before we get there,
-
0:45 - 0:50how can something so complex as emotion
be converted into mere numbers, -
0:50 - 0:53the only language machines understand?
-
0:53 - 0:57Essentially the same way our own brains
interpret emotions, -
0:57 - 0:59by learning how to spot them.
-
0:59 - 1:04American psychologist Paul Ekman
identified certain universal emotions -
1:04 - 1:09whose visual cues are understood
the same way across cultures. -
1:09 - 1:14For example, an image of a smile
signals joy to modern urban dwellers -
1:14 - 1:17and aboriginal tribesmen alike.
-
1:17 - 1:18And according to Ekman,
-
1:18 - 1:19anger,
-
1:19 - 1:20disgust,
-
1:20 - 1:20fear,
-
1:20 - 1:21joy,
-
1:21 - 1:22sadness,
-
1:22 - 1:25and surprise are equally recognizable.
-
1:25 - 1:30As it turns out, computers are rapidly
getting better at image recognition -
1:30 - 1:34thanks to machine learning algorithms,
such as neural networks. -
1:34 - 1:38These consist of artificial nodes that
mimic our biological neurons -
1:38 - 1:42by forming connections
and exchanging information. -
1:42 - 1:46To train the network, sample inputs
pre-classified into different categories, -
1:46 - 1:49such as photos marked happy or sad,
-
1:49 - 1:51are fed into the system.
-
1:51 - 1:54The network then learns to classify
those samples -
1:54 - 1:58by adjusting the relative weights
assigned to particular features. -
1:58 - 2:00The more training data it's given,
-
2:00 - 2:05the better the algorithm becomes
at correctly identifying new images. -
2:05 - 2:07This is similar to our own brains,
-
2:07 - 2:12which learn from previous experiences
to shape how new stimuli are processed. -
2:12 - 2:15Recognition algorithms aren't just
limited to facial expressions. -
2:15 - 2:18Our emotions manifest in many ways.
-
2:18 - 2:20There's body language and vocal tone,
-
2:20 - 2:23changes in heart rate, complexion,
and skin temperature, -
2:23 - 2:28or even word frequency and sentence
structure in our writing. -
2:28 - 2:31You might think that training
neural networks to recognize these -
2:31 - 2:34would be a long and complicated task
-
2:34 - 2:37until you realize just how much
data is out there, -
2:37 - 2:40and how quickly modern computers
can process it. -
2:40 - 2:42From social media posts,
-
2:42 - 2:44uploaded photos and videos,
-
2:44 - 2:45and phone recordings,
-
2:45 - 2:47to heat-sensitive security cameras
-
2:47 - 2:50and wearables that monitor
physiological signs, -
2:50 - 2:53the big question is not how to collect
enough data, -
2:53 - 2:55but what we're going to do with it.
-
2:55 - 3:00There are plenty of beneficial uses
for computerized emotion recognition. -
3:00 - 3:03Robots using algorithms to identify
facial expressions -
3:03 - 3:04can help children learn
-
3:04 - 3:08or provide lonely people
with a sense of companionship. -
3:08 - 3:11Social media companies are considering
using algorithms -
3:11 - 3:17to help prevent suicides by flagging posts
that contain specific words or phrases. -
3:17 - 3:21And emotion recognition software can help
treat mental disorders -
3:21 - 3:26or even provide people with low-cost
automated psychotherapy. -
3:26 - 3:27Despite the potential benefits,
-
3:27 - 3:31the prospect of a massive network
automatically scanning our photos, -
3:31 - 3:32communications,
-
3:32 - 3:37and physiological signs
is also quite disturbing. -
3:37 - 3:41What are the implications for our privacy
when such impersonal systems -
3:41 - 3:45are used by corporations to exploit
our emotions through advertising? -
3:45 - 3:47And what becomes of our rights
-
3:47 - 3:51if authorities think they can identify
the people likely to commit crimes -
3:51 - 3:55before they even make
a conscious decision to act? -
3:55 - 3:57Robots currently have a long way to go
-
3:57 - 4:00in distinguishing emotional nuances,
like irony, -
4:00 - 4:05and scales of emotions,
just how happy or sad someone is. -
4:05 - 4:09Nonetheless, they may eventually be able
to accurately read our emotions -
4:09 - 4:11and respond to them.
-
4:11 - 4:15Whether they can empathize with our fear
of unwanted intrusion, however, -
4:15 - 4:17that's another story.
- Title:
- Can machines read your emotions? - Kostas Karpouzis
- Speaker:
- Kostas Karpouzis
- Description:
-
View full lesson: http://ed.ted.com/lessons/can-machines-read-your-emotions-kostas-karpouzis
Computers can beat us in board games, transcribe speech, and instantly identify almost any object. But will future robots go further by learning to figure out what we’re feeling? Kostas Karpouzis imagines a future where machines and the people who run them can accurately read our emotional states — and explains how that could allow them to assist us, or manipulate us, at unprecedented scales.
Lesson by Kostas Karpouzis, animation by Lasse Rützou Bruntse.
- Video Language:
- English
- Team:
closed TED
- Project:
- TED-Ed
- Duration:
- 04:39
![]() |
Jessica Ruby approved English subtitles for Can machines read your emotions? | |
![]() |
Jessica Ruby accepted English subtitles for Can machines read your emotions? | |
![]() |
Jessica Ruby edited English subtitles for Can machines read your emotions? | |
![]() |
Jennifer Cody edited English subtitles for Can machines read your emotions? |