Artificial Intelligence: Sometimes computers are just like children

7 June 2017

One of the latest buzzwords is A.I., or Artificial Intelligence. You increasingly hear about A.I. taking over operations that in the past could only be carried out by people, such as driving cars and lorries, making recommendations for products or films, or detecting diseases in MRI scans. The essence of Artificial Intelligence is that you show a computer how a particular operation is done and then an algorithm learns how to do it by itself. This principle is also known as Machine Learning. Some people, including Elon Musk and Stephen Hawking, even warn us of the power of Artificial Intelligence. Are they right to warn us, or is it actually fine and we should just embrace A.I.?

How does Artificial Intelligence work?

In a previous article about Machine Learning, our colleague, Luc Claassens, explained the Turing test. The aim of this imitation game is for a human not to be able to distinguish between the behaviour of another human being and that of a computer imitating human behaviour. Thanks to Hollywood, Artificial Intelligence makes many people think of robots that can do human activities just as well as humans can, or even better. This is called Artificial General Intelligence, and it is still a long way off. The term Artificial Intelligence is used when a machine is able to perform a single cognitive operation, or a series of cognitive operations. A computer learns to do that by using Machine Learning. You show the computer how it is done and then get it to “copy” it again and again, for as long as it takes through trial and error for it to be able to do it on its own. It’s almost like teaching children: you show them how it is done and then get them to practice doing it until they get the hang of it. The learning process is therefore called Machine Learning and being clever is called Artificial Intelligence. As an illustration, the following clip from the Coursera course by Professor Andrew Ng shows how a car learns to drive itself.

At the top left you can see two bars: the top one is the position of the steering wheel, the bottom one is what the algorithm predicts that position should be. The more the car drives, the more training data the car collects and the better the algorithm learns to drive. You can see this in how the bottom bar becomes progressively more precise and more accurate. Eventually the car is intelligent enough to drive itself, and then you can press a button and the car will continue driving on its own.

Deep learning

One of the current state-of-the-art A.I. algorithms is Deep Learning. Deep Learning generally refers to a group of algorithms called neural networks. The essence of Deep Learning is that the computer, rather than learning to copy an operation, basically learns to think in the same way as the human brain does in all its complexity. Neural networks have actually been around for some time; they were first discussed in academic circles in the 1940s and 1950s. However, they only began to gain major popularity in the last few years because it is only now that computers are powerful enough to perform the complex calculations required. Neural networks are made up of neurons with connections between them and as such they mimic the functioning of the human brain. By activating combinations of neurons, complex patterns can be identified. You could see Deep Learning as being Neural Networks on steroids: they are Neural Networks with a huge number of neurons and extra functionalities.

For example, Deep Learning is used to analyse images, combining all the individual pixels together into a coherent whole. This is demonstrated very well in the following visualisation of various Neural Networks and Deep Learning models. You can see a handwritten number, it’s made up of 50×50 pixels: this is the input. The algorithm then works out what number it actually is by activating particular neurons.

Should we be worried?

Musk warns that computers will one day be more intelligent than people and will take decisions that are not in the interests of society. In 2015, a thousand robotics and A.I. scientists signed a petition against the use of A.I. in wars. Computers have no concept of morality or ethics, and drones can wipe out whole villages or towns on the other side of the world by themselves. Whole populations could be wiped out without a single person lifting a finger.

Musk and Hawking’s reasoning is that the aim of Machine Learning is to teach a machine so that a particular error is minimised, i.e. to tackle a certain problem as effectively as possible. It’s only a matter of time before machines will realise that we, with our human behaviour, are inefficient and will seek to eliminate us.

What I do think is going to be an even greater challenge in the near future is employment. If lorries can drive themselves, you don’t need any lorry drivers. According to the CBS (Centraal Bureau voor de Statistiek [Statistics Netherlands]), we currently have 285,000 lorry drivers and 61,000 taxi drivers, which means we would need to find other jobs for these 346,000 people in the Netherlands. This doesn’t apply only to lorry drivers and taxi drivers, but also to other professions which can be automated once A.I. has reached its next level. This will result in a large surplus of workers, with major economic consequences. Although over the years the economy of the Netherlands has shown itself to be flexible enough to compensate for large portions of unemployment caused by automation, it remains a threat and this should actively be driving decision-making.

Should you be worried too? Oxford and McKinsey both carried out a study into the probability of your job being taken over by robots.

Should we stop A.I.?

Far from it. It is an extremely interesting world with great added value for a wide range of fields. For example, A.I. is used in the medical field to detect diseases and abnormalities in MRI scans of the brain. It is used to convert spoken language into written text (e.g. Siri), or it can make a camera on your car’s dashboard read the road signs so that your car knows how fast to drive.

But also, and that’s what we use it for most, it can help companies to better serve their customers. We utilise Machine Learning to analyse customer behaviour. We look for patterns in behaviour to then be able to predict future behaviour. For example, could we predict who is going to cancel their subscription or policy? Could we predict who is going to convert as a result of a particular campaign? Could we predict which channels, tone-of-voice and propositions best suit each individual customer? Could we predict the Next Best Need of Action so that we know which products are most promising to cross-sell to individual customers?

Would you like to learn more about Artificial Intelligence and how to apply it? On 15 June, our colleague Wouter van Gils and I will be at the Kennisfestival in Deventer where we will take you on a journey through our world of Machine Learning and Artificial Intelligence and show you how we use it to better serve your customers.