Thursday, June 19, 2014

The future of marketing? Make machines think like humans do.

What's the next big thing in internet tech? It's getting computers to think more intelligently and laterally the way that the human brain does. Artificial Intelligence (AI) is the science and engineering of making intelligent machines (Source : Wikipedia) and it's been around for some time. It has led to some fun and interesting applications like the Turing Test, intelligent supercomputers that beat humans at chess and increasingly smarter robots - the latest one is Softbank Mobile's emoting robot Pepper. But the newest and most exciting trend in AI is 'deep learning'. Put simply, deep learning understands and models how the brain works to improve computing tasks and it is increasingly getting applied in areas like image recognition, voice recognition, language modelling etc. 

Google unsurprisingly has led the way in this field with the "Google Brain" project, an ambitious attempt to recreate the human brain using hardware and software. It's not surprising that Google got interested in this field. They have always been interesting in staying ahead of the curve on what people need to search for and how they search for it - and this would require them to understand and replicate the way the human brain works. Even a driverless car has to be trained to behave in the way a driver would. Baidu, the Chinese search giant, has wooed one of the brains behind Google's experiments, Andrew Ng, to supervise a similar project.

Now, companies are finding this approach to be promising for a more commercial application - understanding how the human brain works and using this to make better product recommendations. Movie rental and streaming service Netflix recently shared in a blog post that they were trying to construct a neural network based on deep learning to improve their video recommendations to users. Pinterest recently purchased a small startup called VisualGraph, which works on the area of image recognition, identifying and co-relating elements to show users pictures that are more relevant to them.Obviously this is fits well with improving Pinterest's user experience.

And more recently, a mobile app called 'Beautiful Me' uses deep learning and neural networks to deliver personalised skin care recommendations. It uses Facebook images of the user to analyse and identify a series of parameters including skin tone and age and prepare a skin profile - an online version of what a beauty expert at the counter might do. This can then become a vehicle for product and brand recommendations. 

What makes deep learning projects more feasible is the increasing availability of big data and the power of the Cloud. For example, Netflix is leveraging Amazon's public Web Service Cloud and their powerful GPU servers which enable quicker processing of complex queries.

And a lot of future research in deep learning will focus on 'unsupervised learning' - where machines will not need to be spoon fed with labels and tags that help them to identify and categorise information. Humans have an ability to teach themselves, and researchers are trying to get machines to emulate this. A huge quantity of data out there is still unlabelled so machines have to be trained to absorb and understand data on their own, without  human intervention to teach them. This will lead to more cost effectiveness and more applications for AI.

This is the future of AI and it while it may not be as interesting as humanoid robots, it definitely holds a more practical commercial interest for marketers!

Sources : Venture BeatWired