BBC Knowledge AUGUST 2017

74

in the kitchen. Their robots can watch videos

of people preparing and cooking food, and,

by doing so, learn to perform similar actions.

“We use neural networks to acquire knowledge

for our robots by learning the functionality of

objects,” says Prof Yiannis Aloimonos.

“Can this tool be used for scooping; can this object

be used as a container? Our neural networks look

at many examples and they have been taught to

make geometric calculations. The combination of

deep learning with geometry leads to recognition

of the action being performed.”

These AIs learn the underlying ‘grammar’

rules of action so that they can achieve their

intended goal without necessarily needing to

perform identical motions. For example,

the rules of stirring using a spoon to repeatedly

mix a liquid in a pot apply to any liquid and

any pot. A simpler AI might only learn how

to use one specific spoon for one specific pot,

containing one specific kind of soup. This higher-

level thinking using such grammar rules is then

combined with a large number of processes that

track and monitor the hands, the objects, tools

and their movements, all continuously running 

in the background. “All of this implemented in

a robot gives rise to the robots of the future

that ‘understand’ the humans around them,

and learn from them,” explains Aloimonos.

Baomar thinks this form of robotic learning

PHOTOS: VOA NEWS, AFFECTIVA ILLUSTRATIONS: JOE WALDRON

can find countless practical applications.

“I believe that, if we give robots the ability to learn

from humans or even from other systems,

the outcome should be intelligent robots that

are capable of learning a wide spectrum of skills,

ranging from domestic chores to performing

surgery and flying complex machines,” he says.

COME WITH ME IF YOU WANT

TO LOVE

So the robots of the future are likely to be capable

of learning and performing complex, highly-

skilled tasks. But how about emotions? Humans

are complex creatures, unpredictable and often

not entirely rational. Our emotions are just as

important as our intellect in driving our actions.

Affective computing – software that recognises

and interprets our emotions – and human-

computer interaction has started to enable

AIs to detect emotions.

“We know from years of research that emotional

intelligence is a crucial component of human

intelligence,” says Dr Rana el Kaliouby, CEO of

artificial intelligence company Affectiva. “People

who have a higher Emotional Quotient [EQ] lead

more successful professional and personal lives,

are healthier, and even live longer.”

Affectiva is using deep learning, a special kind

of neural network containing many layers of

neurons, to enable computers to detect our

ABOVE:

Robot chefs

could spell

the end of

sweating over

a hot stove

FACING

PAGE:

Dr Rana

el Kaliouby

demonstrates

emotion-

sensing

technology

used by her

company’s

artificial

intelligence

Robotics

science