I recently completed a wonderful course in Machine Learning taught by Prof Andrew Ng of Stanford on Coursera. Machine Learning, more popularly known as Artificial Intelligence or AI, is not a new topic. It has been in the making for over 60 years. But it has started delivering creditable results in the last few years. The success of the driverless car has captured the popular imagination but progress in the areas of image recognition, natural language processing and anomaly detection is no less impressive and has applications across sectors.
One might argue that AI is part of the digital transformation that is sweeping the world in any case, so what is so special about it. Well there is a fundamental difference. Till this time computers were great for doing repetitive jobs quickly and reliably so they were deployed mostly for automation. They replaced humans for mechanical tasks. But for the first time AI is allowing computers to take up tasks that involve thinking and learning. An AI machine is not limited by the imagination of its programmer. It can actually learn by itself and get better with experience at the task that it has been assigned. This is a paradigm shift that has serious implications for all of us. It has heightened fears of jobless growth on the one hand but holds out the promise of a better quality of life on the other. The lesson of history is quite clear. Progress of technology cannot be denied, but society must find ways to curtail the negative fallout as well.
To get a sense of where AI is today and the long journey still ahead, listen to this TED talk by Fei-Fei Li Director of Stanford’s Artificial Intelligence Lab.
Here are two more videos to provide perspective.
Keynote by Eric Horvitz Managing Director of Microsoft Research's Redmond lab.
TED talk by Nick Bostrom of Oxford University.
A heartening feature of current research in AI is that most of it is open source. A researcher anywhere in the world has access to the latest tools and even very large data sets painstakingly compiled by other research teams. Another interesting feature of AI research is that while it draws draws upon classical mathematics and statistics it is not limited by them. True to its nature, many advances in AI are based on heuristic approach where the researchers themselves are unable to fully explain why their algorithms are working well and what exactly they are doing. This approach actually allows AI to deal well with unstructured data for example from social media.
So what exactly can we expect from AI in the near term. The class of problems that are best suited for AI solutions are those that involve classification and prediction. Once a AI algorithm has been trained on training data it can be used to process actual data and present results. For example recognizing a person in a photo or predicting what movies a person may like from a movie database. AI is being deployed for these types of applications fairly extensively, especially by ecommerce companies.
However AI techniques can provide significant benefits in many other areas also. Here are a few examples.
- Electricity load prediction based on weather conditions.
- Algorithm based credit approval for bank loans.
- Fraud detection and other applications in financial services.
- Migration trends based on geo tagged twitter feeds.
- Diagnostics and patient management in health care.
- Criminal justice.
- Media and knowledge management.
- Transportation and traffic management.
- Weather monitoring.
- Sustainability and wildlife conservation.
There is no doubt that the availability of cheap hardware on pay as you use basis via the cloud, open source advanced AI tools and availability of large quantities of data is driving the uptake of AI technology on an unprecedented scale.