Recent progress in artificial intelligence (AI) has been impressive, but the advent of a universal computing machine is still some way off. Interview with Cédric Villani, mathematician and director of the Institut Henri Poincaré in Paris.
Which will win out – robots or human beings? This is a highly topical question. The South Koreans, and many more millions of curious people worldwide, became obsessed with this issue when a match was played between a professional Go player named Lee Se-Dol, and AlphaGo, a computer programme developed by Google subsidiary DeepMind. The match resulted in a 4-1 victory for the machine over the Korean star player. Given that Go was one of the last bastions of the games world to hold out against the learning and analysis techniques employed by cutting-edge computers, this defeat basically symbolises the considerable progress made by deep learning.
Which will win out – robots or human beings? Is this question even relevant? We got some answers from another heavyweight player, Cédric Villani, who is a top performer in his field of Mathematics.
L’Atelier: Some weeks ago, in South Korea, a battle took place that many people got very excited about, between the Go game world champion and AlphaGo, a computer programme developed by Google subsidiary Deep Mind. This excitement tells us something about people’s concern that a machine might go beyond carrying out basic processing tasks and develop truly intellectual ability, in the long term replacing us. Does it make sense to try to pit artificial intelligence against human intelligence in this way?
Cédric Villani: Yes, there is a lot of talk nowadays about artificial intelligence. We’re in an era of progress, on the one hand thanks to the development of some high-performance algorithms, and on the other hand because of the considerable progress made with memory capacity and processing speed. Along with this progress we’ve seen some traditional bastions among major games such as Go fall in the man-machine struggle.
And it’s clear that this once again triggers the speculation – which we’ve seen since the very beginning of computing – about the potential for perfecting artificial intelligence. Perfecting in the sense that AI would be capable of writing novels and coming up with mathematical theorems, or even feeling emotion. Alan Turing, the father of modern computing, was already asking these questions in the 1950s.
“Of course the performance of these machines is impressive. But just look at the extravagant outlay of resources compared with what can be done with a human brain.”
At the present time human and artificial intelligences are very different in nature, even though the techniques, so-called machine learning, have reintroduced a certain amount of automated natural learning by impregnation, as you do naturally for certain tasks. And they have introduced this for algorithms. However, all the AI programmes that exist today are highly specialised. Even the one that knows how to play Go better than any human being is incapable of accomplishing the most ordinary daily tasks. Whether or not an all-purpose AI programme – even one that’s not all that clever but nevertheless capable of doing a bit of everything – can be developed, is still very much open to question. Of course the performance of these machines is impressive. But just look at the extravagant outlay of resources compared with what can be done with a human brain.
What you’re saying comes down to that fact that a truly creative machine is still some way off. Take for example the works of Dostoyevsky translated into French by André Markowicz. The day when our machines can produce as much sensitivity and intuition as an André Markowicz has certainly not yet arrived…
That’s right, you won’t see this kind of machine appearing overnight. In order for a machine to be able to carry out a translation incorporating style and emotion, it needs to have a far broader knowledge base than just to translate. It must be able to understand what a life experience is in the context of a human life. It’s difficult to know how to carry out a creative task without starting off by learning a bit of everything – having general intelligence.
“When you’re finding your way through a problem, you often rely on instinct, on intuition and also on the notion of elegance. Instinctively you let yourself be guided in the direction that seems best. Now, being able to make a machine feel that would be a considerable feat.”
The input of mathematics into the field of computing is obvious, but how do you see the potential for solving a mathematical problem using a computer? In the future will it be possible for a machine first of all to frame a mathematical proof and then verify it?
We need to separate the two. Verifying the validity of a proof will be done very soon. The technology is for the most part available. And a mathematical proof is a sufficiently formalised item to be capable of verification by a computer. On the other hand, framing a proof is a completely different story because it starts in the vast space of the possible. And finding the right strategy among all those possibilities demands a huge amount of talent and input. We mathematicians know very well that when you’re finding your way through a problem, you often rely on instinct, on intuition and also on the notion of elegance. Instinctively you let yourself be guided in the direction that seems best. Now, being able to make a machine feel that would be a considerable feat. The machine has the advantage of being able to explore all the different combinations much faster than human beings. But if it isn’t capable of finding the way, having the right intuition about which direction to go, it’ll find itself stuck in an ocean of possibilities, the ‘curse of dimensionality’ as we call it, because there are so many possible choices when you want to demonstrate a proof. In comparison to that, a game of chess or even Go offer far fewer choices.
“It’s more important to focus on everything that still remains to be done and is within our reach than shedding tears about all the things that are outside our grasp.”
Could it be that, taking the reverse situation, the human brain – faced with this exponential progress in technology – could be capable of grasping everything, of grasping advanced results and great quantities of data?
Well, there are some things that will remain inaccessible to the human brain. It’s actually a fair bet that the brain will stay more or less as it is now. Unless we want to go off into the world imagined by the transhumanists, where we would all have enhanced cerebral faculties and external memories hooked up to our brains. This is just not going to happen tomorrow.
Unless we go off in that direction, there are limitations to the human brain which we can see easily, and which we continue, and will continue to experiment on. What’s important is that even despite these limitations, there are still huge numbers of things to discover, sometimes very simple things. And each year that passes continues to demonstrate this.
There are many discoveries now being made that we can grasp, can understand. Of course we’ll never understand all the details to the nth degree. For example, it is said that there is no single person on earth who is able to understand how a smartphone works right down to the smallest detail. Each person only has a partial view. But that doesn’t mean we can’t understand the general principle. And for each of the sub-parts we can understand how it works, what the main principles are.
It’s more important to focus on everything that still remains to be done and is within our reach than shedding tears about all the things that are outside our grasp.
There’s an increasing demand for data scientists. Are we up to it? Are our schools buzzing with aspiring data scientists?
Well, there are several questions here. One is whether we have good faculties for data scientists. The answer to that is yes. The number has increased in recent years. In France some of our ‘Grandes Écoles’ – top public higher education and research institutes – such as Télécom ParisTech and Mines ParisTech have introduced more technical courses. The universities have also started to build up faculty in this field. Here and there in France we now have faculty that are getting good results.
The second question is whether we’re training enough people. Overall there are still not enough data scientists and we ought to be training far more. There’s a substantial need in industry. Overall we have far fewer that we could have. And if there are young people who read my books and wish to embark on a career in Mathematics, there are far more opportunities now than there used to be. At the moment data science – the science of Big Data plus statistics – is king. So much so that industry needs for professional data scientists cannot be met. There are many fields of application for this discipline.
Not so long ago we had the problem of financial mathematics overshadowing pure maths programmes to an unreasonable degree. Big Data specialist skills are in demand in far more areas than just mathematics for the financial sector. We mustn’t forget the other areas. In particular, we more than ever need specialists in the areas of modelling, scientific calculation, equation simulations and partial derivatives.
“The troops – our young students – are the nerve centre in the war, and they are a country’s most precious resource in terms of the future”
Are we going to face a shortage of simulation practitioners, then?
In all branches of mathematics there are huge needs as regards applications. We ought to be training far more people than we are at present. The question is how enthusiastic are the troops? The keener they are, the more we can train them. The troops – our young students – are the nerve centre in the war, and they are a country’s most precious resource in terms of the future.
On 30 May, Cédric Villani is due to give a lecture at the Maison des Métallos cultural centre in Paris.
If you would like to delve deeper into this subject, in the exalted company of Karol Beffa and Cédric Villani, obtain their book ‘Les coulisses de la création’, (The Corridors of Creation) – not yet available in English translation – published by Flammarion.