In 2011, ‘supercomputer’ Watson beat human contestants in a quiz game show Jeopardy!, sparking speculation on whether machines would one day replace human beings in the workforce.
In the last six years, Watson has come a long way. It is now a cognitive technology platform that uses natural language processing and machine learning to reveal insights from unstructured data. For example, it can assist doctors by recommending possible diagnosis and lines of treatment.
Views of Sriram Raghavan (Director of India Research Labs, IBM)
How different is cognitive computing from robotics?
Robots automate mundane tasks. But cognitive technology enables automation of tasks that require human beings to think and make decisions. In other words, cognitive computing involves automation of tasks that require intelligence. It is actually ‘intelligence automation. It is popularly called by people as artificial intelligence.
Why is cognitive technology being spoken of as transformative?
Cognitive technology is gaining a lot of attention because traditionally, it was possible to analyse and understand only structured data, which make up just 20% (of all available data). With cognitive technology, we can analyse with precision unstructured data as well.
It is about understanding, reasoning and learning?
Watson Oncology Advisor, for example, has been taught to understand data about cancer. This is revolutionary because the amount of data on clinical trials and research material is so vast that the average doctor can’t keep up. With understanding and reasoning, Watson has become the assistant. The system also keeps learning, with the feedback from doctors.
Its applications are: assisting human beings, scaling expertise and personalisation.
How difficult is it for Watson to understand, reason and learn?
1. The system has to understand if the same information is presented in different ways. For example, you can write in different ways an email to me to say that you will meet me at 6 p.m. tomorrow. But irrespective of the way you write, I will understand what you mean.
2. Another challenge is ambiguity. The same information can have different meanings in different contexts.
A good case study was reported from Japan in 2015. A woman was admitted to a hospital and diagnosed with acute myeloid leukemia, a type of blood cancer. Post-chemotherapy recovery, however, was abnormally slow. That’s when the hospital turned to Watson, which cross-checked the woman’s genetic data with its database and detected gene mutations.
She had mutations in over 1,000 genes, many of them were not related to her disease. Watson found out which of these 1,000 genes were diagnostically important, in just 10 minutes; while scientists would have taken nearly two weeks to do that. Based on Watson's recommendations, the doctors changed their line of treatment, and her condition improved.
In October, you announced the release of Watson Element for teachers to tailor educational material for students. How does it work?
Education is the third area where we are making progress. There is great potential to personalise education. Just as a travel portal personalises the adverts and the offers for you, teachers can personalise the delivery of content to students. Customising helps, as students have their own strengths and weaknesses, and they learn differently.
Traditionally, teachers assess students by looking at their past records or tests or by talking to them. But this is not scalable. Can we have one teacher for every student?
We are working on the notion of cognitive tutor. Its capabilities are understanding students’ strengths and weaknesses, and using that data to personalise the delivery of content.
Technology enables us to track how students use content. Data on content usage – such as starting point, portions of video that the student watched repeatedly, points where the student paused, for how long, how the eye moved while reading, etc. — can provide valuable insights for the software to intervene and suggest study material.
One of the applications of AI is in vetting of resumes during recruitment. But does AI dilute the importance of the human element?
Resumes are huge unstructured documents, and if you want to apply a set of objective criteria (like whether all applicants are graduates), there is no harm in machines doing that, since you don’t have the scale to apply that criteria.
For example, if the existing number of people can cover only 200 resumes, the machine can do 2,000 resumes. But if you are going to extend it and hand over the whole recruitment process, then probably you are exceeding the maturity of technology.