Artificial Intelligence

0:00
It’s behind the scenes in our favourite apps.
0:02
Making breakthroughs in the lab,
0:04
and constantly in the news.
0:06
Artificial Intelligence is all around us.
0:09
But how does it work?
0:11
Let’s shed some light on the world of AI.
0:15
‘Artificial Intelligence’ is an umbrella term that covers a wide range of systems that
0:19
all work in slightly different ways.
0:21
But many of the most widely known use mathematics to find patterns in data.
0:27
These patterns are then used to make predictions.
0:30
The most common forms of these AI systems use something called machine learning.
0:36
Where algorithms analyse data, arranging the patterns and features into models.
0:42
You can think of models a bit like a map.
0:45
Let’s look at how a machine learning  model might use an image of a koala.
0:50
This image has millions of pixels.
0:54
When we put this digital file through a machine learning model,
0:58
these data points are processed  via many layers of multiplication and addition
1:02
until patterns start to emerge for different features.
1:06
You can think of these like islands.
1:09
The more images we add, the more  comprehensive the map becomes.
1:13
Here on ‘Koala Island’, the west side of the island represents koalas with small ears,
1:19
and the east, big ears.
1:22
What’s really mind blowing, is that ear size is only one feature,
1:27
and to represent all of the different shapes, colours, moods and compositions
1:31
of koala images, we not only have to get 3D
1:35
but imagine thousands of dimensions.
1:39
Maybe for now, let’s keep it at just 2!
1:43
Now, if we take this map that’s been  trained with millions of koala images,
1:47
and ask what an image might look like… here,
1:51
an AI system can generate a completely new image related to that location.
1:56
This is called Generative AI.
1:59
Amazingly, this mapping process works for any data.
2:03
Whether it’s text, images, sound - whatever can be described with numbers!
2:08
When we train models with these different data types together, it’s a bit like combining two maps.
2:14
Here we have a text-to-image model,  trained using images and their text labels.
2:21
These models can take on complex tasks, answer questions,
2:25
write poems and music, and even generate videos from scratch.
2:31
But fundamentally, what we’re doing is giving computer systems a way of mapping information,
2:37
and making connections between patterns.
2:40
It’s maths, not magic.
2:43
So while these outputs are very convincing, 
2:46
it’s important to understand that they’re  only a prediction based on training data.
2:50
If we go back to our text-to-image map,  
2:53
and imagine that our text map was trained using American examples, instead of Australian
2:59
This prompt might give us a much different result.
3:03
Sometimes this can lead to something  called bias. Where unfair or unbalanced  
3:07
outputs are generated that amplify  inaccuracies or gaps in the data.
3:12
Take planning a playlist for example.
3:14
If an AI was trained using only your listening history,
3:18
it won’t have the right data to generate a playlist that appeals to everyone.
3:22
It’s important to remember that human and artificial intelligence  
3:25
are different things.
3:26
For example, humans can instinctively understand context and apply common sense.
3:32
AI systems approach things differently.
3:35
That’s one of the reasons it’s important to understand how AI systems work,
3:40
to ask questions and decide when and how we want to use them.
3:44
In the right hands, AI systems can be incredibly powerful tools.
3:49
Helping manage huge data sets, see patterns that humans can’t see
3:54
and automate complicated processes.
3:57
But it’s up to all of us to ensure it’s being pointed at the brightest future possible.
4:02
To find out more check out csiro.au/ai