Originally Shared Publicly on Google+

I indirectly received, from our Campus PR company, the following questions from a journalist about recent ‘AI developments’. In particular she asked about developments in the context of the recent purchase of Deep Mind by Google. The questions are short, but my answers are longer. It seemed to make sense to share them here.

Q: Do you think 2014 is the year of AI?

A: The focus of major companies like Google and Facebook is turning towards machine learning researchers, in particular deep learning researchers, due to the impressive performance of these models on some tasks that have formerly been seen as difficult. In particular in the area of image understanding. The investments we are seeing are very large because there is a shortage of expertise in this area. In the UK we are lucky to have some leading international groups, however the number of true experts in the UK still numbers in the tens rather than the hundreds. The Deep Mind purchase reflects this, their staff was made up in large part by recent PhD graduates from some of these leading groups. Although even in this context the 400 million dollar price tag still seems extraordinary to many in the field. The year 2014 is not the year in which these developments happened, but it may be the year in which they’ve begun to impinge upon the public conciousness.

Q: How do you think the industry is likely to progress in the near future?

A: Interestingly it is not in industry where the breakthroughs have happened, but in academia. In terms of methodological steps forward, I believe we will see breakthroughs in image understanding, language and speech processing. But this will leave a range of challenging problems still to be resolved. A particular focus of my own group is dealing with ‘massive missing data’: where most of the information we would like to have to base our decisions on is not available. Any one piece of information on its own is not enough to make a decision: assimilating the information to reduce uncertainty is a key challenge. Even then, we are required to make decisions based on only a very small visible portion, a kind of iceberg effect. Beyond my own area of research there are also key challenges in the areas of planning and reasoning. It is not yet clear to me how the recent breakthroughs will effect these areas.

Q: What do you think the long term future of AI is?

A: It is very bright, but progress will be steady, not with large single steps forward, but across a number of applications. Expectations may currently be too high for the immediate future, we are still many years away from achieving many of our goals in artificial intelligence research. The current successes have emerged from an area known as machine learning, a foundational technique that already underpinned much of the data driven decision making of the large internet companies. The methodologies used have mainly emerged from a relatively small annual conference known as NIPS. The recent breakthroughs emerged from a group of NIPS researchers who received very far-sighted funding from the Canadian government (the Canadian Institute for Advanced Research NCAP program). The program spent a relatively small amount of money (tens of millions) on a carefully selected group of people. This group was led by Geoff Hinton (now of Google) and advised by Yann LeCun (now of Facebook). In the UK, for example, large amounts of money are now promised, but it is not at all clear whether it will be well spent. Functional research operates rather like a well tended garden: it needs an understanding of the right sort of plants and the ideal conditions for them. A sudden large increase in funding can have a similar effect to indiscriminate application of manure: something will grow, but it’s not at clear at the outset what it will be. When it comes to harvest time, will we have roses or dock leaves? The Canadian approach was to select the roses first, and then carefully tend them. Other countries would do well to follow a similar approach if they want to reap similar rewards. 

Here’s a link the the article the journalist wrote.