Artificial Intelligence

I must talk a bit more about the referendum.  Reading more about the vote I see it was more nuanced than I thought at first. While immigration was the dominant feature there was also a inarticulate cry of anger by those who feel disadvantaged. A sort of “sod you, I’ll be poorer but you’ll be worse off “. This was not just about the EU but an inarticulate shout against social liberalism.

However onto the main topic, Artificial Intelligence

There has been much hoo-ha about this in recent months. Hopefully I can set out what is achieved and make some guesses about the future. Artificial Intelligence ( AI ) has been of research interest for years and until recently had disappointed. However since 2011 or so AI has come along in a spectacular fashion.

This is down to the use of neural networks. These networks, simulated in software, take information and process it through nodes called neurons which have a weighted interaction with other neurons. Crucially these weights ( or probability of transmitting a signal ) can be modified by the input data. Essentially an AI system consists of an input layer, several neuron like layers( can be dozens or more ) and an output layer. The neurons, or brain like cells, are simulated in computer software.

I was involved in using a neural network to try and develop water based corrosion preventives. We bought the network as a package and fed in the results of systematically changing the formulation. Then based on the output we hoped to be able to “tune” the formulation. It was a failure. The output seemed sometimes counter to what our prior knowledge told us. The system was effectively a “black box”; in other words we had no means of knowing why the result was obtained. Modern day thinking would be that we simply provided the network with a completely inadequate stock of information.

From the Economist supplement on AI it seems that the breakthrough idea is to develop a neural network and then train it on gigantic amounts of information. It is the internet which allows ready access to these huge amounts of information and low cost computing which enables the networks. For example the first success was based on image recognition. Vast numbers of portraits were scanned and the AI can sort out say images of men with moustaches. This can be achieved by asking the AI system to simply organise the images by their salient features. Alternatively the fine tuning of the network can be achieved by first picking out a small subset and then telling the AI when it’s right or wrong. A well tuned AI is better than a human.

While these AI can be hugely impressive it is also important not to think they are both powerful and versatile. So far AI seems to be directed at some particular task and a multi- purpose AI is probably decades away. Even so there has been some adverse comment. A few, including eminent scientists, have seen a risk that an AI could plan an even better AI and eventually outpace humanity. This is the stuff of science fiction and far away from present capabilities. A more immediate objection is that AI could displace people doing routine but skilled jobs. For example radiographers examining X-ray pictures for signs of disease could be supplanted by AI which would be better at the job. The other obvious advantage being that the AI would not suffer any distraction and could work 24 hours a day.

There is the usual fear that any type of automation would displace workers. However history tells us that new types of work better suited to human skills arises. Our economic life is an ever evolving dynamic system. Just to take an almost trivial example both my son and son-in-law work in IT in jobs which were simply unknown when I was young. The problem, if there indeed is one, will be the speed at which change may happen. It is a truism of modern life that simply training and then using that knowledge for a working life is no longer adequate.

I can readily see that despite years of training in my own young life it was still necessary to learn new things through my career. Increasingly we must see that education is about learning how to learn not about learning a particular subject. Education is also about providing a framework on which continued learning can hang.

It is easy to write about flexibility in working. It is not so easy in practice. Try telling an unemployed 50 year old who has done one thing all his working life there is a skills shortage. But there is a skills shortage and nationally we ignore this at our peril. I’m very pleased that even in our local primary school I see that learning is more attuned to the future. I’m afraid it is all too easy to wish the world would stop changing but change it will. I can’t see clearly what impact AI will have. It will be profound and may be faster than is comfortable.