“Ask Kirsten about cat croissants, it will make her giggle”, said Dr Rachel Craddock.  So, when I spoke to Kirsten McCormick, I did ask her about cat croissants, and she did giggle.

“They are wonderful!  We noticed them when we were putting together this presentation (next week’s webinar ‘Demystifying Artificial Intelligence’) actually.

We've done the webinar very much jointly, and we spit ball off each other quite a lot.  We were searching common myths about AI, discussing how AI can be tricked by things that look very similar.  We found loads of pictures, some were of croissants and some were of ginger cats or kittens in the shape of a croissant.  And we got very distracted by that.”

This illustrates an issue with AI, how to tell one thing from another when they look alike.  A photograph of a ginger cat resting on a bed might resemble a picture of croissant (think of that browned puff pastry) on a plate.  How does the machine tell the living from the dessert? 

A solution in search of a problem

I put it to Kirsten that AI is a solution in search of a problem, with less practical utility than its advocates would like.  Her response was fascinating.

“I was doing an integrated Masters in Physics, at Nottingham Trent University. Within my master's project I was looking at classifying blood droplets; how they crack, dependent on different stages of physical exhaustion.

We would take blood at different points of exhaustion; at baseline, peak exertion, then after resting for 1 hour, 2 hours, seeing how the crack patterns evolve, using an unsupervised learning approach.  After I began to understand the AI, it grabbed me because it was such a useful tool and can be used in so many different ways to solve things we may not have been able to solve previously.

I actually stayed on and did a summer project, looking at malarial versus non malarial blood, based on the crack patterns.  Because you could use AI for that classification, the crack patterns may have been so similar we would not have picked it up.  However, the machine learning was picking up that every time something happened within it that was a malarial blood.

You think suddenly, that's an incredible use of AI which you may not have achieved if you were doing just a normal algorithmic sort of non-AI approach. “

Repli-catting complexity

When creating Artificial intelligence, or mimicking a brain, we’re up against the range of tasks that a brain, human or animal, can perform at the same time. 

Doctor Rachel Craddock, who’s passion for her subject is infectious, pointed to the cat brain and what it can do.

“Consider a cat whose brain is the size of a walnut.  If you think what that brain actually does, it enables a cat to catch a mouse in the dark without using any tools; to run around the house, climb trees, and annoy people.  The cat is able to have a personality, is able to recognize a certain number of verbal commands from a human, it can recognize the sound of a tin opener from 3 miles away.  Up to a point, the cat can communicate its needs and manipulate its humans.  And that is all in something the size of a walnut.

How on earth does that work?

Most animals’ brains are not only doing the conscious thinking and the actions, but they're also controlling all of the processes in the body. If that bit switched off, I have no idea about how half the internal organs work.

I wouldn't know where to start.”

So what does that mean?

In the course of meeting Ms McCormick and Dr Craddock, it becomes obvious that as a field, Artificial Intelligence requires more than simply solving a practical problem.  It requires a deep level of thinking about building and deploying it.  

Dr Craddock also underlined something else it will force us to do.

“We need to improve our understanding of our own intelligence or biological intelligences to be able to create an artificial equivalent.

We're probably better off now. Currently we've got AI that is very, very good at doing a very limited thing.  It will be better than a human doing that very limited thing.

The minute we try and get it to do everything, are we going to end up recreating humans with all of their fallacies, with all of the mistakes, with all of the learning time that they need?

Is that really the goal? Is it feasible, and is it the right thing to do?” 

Artificial Intelligence is a gigantic, complex subject.  To make it easier to understand and find out about cat croissants, why not sign up for the webinar, ‘Demystifying Artificial Intelligence’; where Dr. Rachel Craddock and Kirsten McCormick will demystify the subject and infect you with their love of the topic.

• Dr Rachel Craddock is a Thales Expert with 30 years’ experience under her belt in artificial intelligence and machine learning.
• Kirsten McCormick is a Systems Engineer and Artificial Intelligence Lead at General Dynamics Mission Systems.