Google and Microsoft Tie in Robotic Caption Contest

AH GOOGLE IO 2014 144


Machine learning has become a big deal in the technology space, and while it’s not quite to the levels where Artificial Intelligent is about to take over the Earth, it is impressive what code can now do on its own. Machine Learning refers to the concept of creating a program that will adapt to certain situations within its original programming. A prime example of this is music recommendation in services like Spotify and Google Play Music. A program has been written to learn what type of music you like, and what type of music you don’t, therefore tailoring recommendations more to your taste than someone else’s. In the recent Microsoft Common Objects in Context (COCO) Captioning Challenge however, Machine Learning was used for something a lot more creative, and Google and Microsoft both tired for first place.

The contest was, on paper, pretty simple. All of the applicants – which included the likes of “China’s Google”, Baidu – were given the same set of pictures and would have computer code generate a caption for it without any human help. These were then judged by the percentage of captions that were as good as, if not better, than human captions and by the percentage of those which passed the Turing Test. Both Google and Microsoft have apparently scored the same, with the results to be announced during the CVPR computer-vision conference in Boston soon. How Google and Microsoft did so well in the test is down to what’s known as ‘deep learning’, a process which tests programs against all sorts of different data to create artificial intelligence neural networks. Deep learning is what makes Google Photos able to display “pictures of cats” despite the fact you’ve never tagged those photos as “cats” and it’s used in a lot of other applications, too.

We have hit a point now where machines and computer code are able to do so much that it’s no wonder people are frightened of what the future holds. Supercomputers like IBM’s Watson are creating recipes on their own, competing against humans in tournaments and there’s seemingly nothing that we can’t teach computer code to do. This has made life much easier in a number of ways, but it does lead us to question just where all of this is headed.