"India is the sweet spot where AI can flourish"

The impact of Artificial Intelligence (AI) will change the distribution of wealth and power across the globe says Dr. Ajay Agrawal, one of the world’s leading experts on the economics of AI. In this interview with Gateway House, Agrawal says that India can be a leader in AI because “much of the work in services that is outsourced to India from the West is right at the sweet spot of what AI is good at and where it will flourish.” Though China is furthest along in AI, now is a window of opportunity for India, as AI can be a new-economy jobs provider.

Gateway House (GH): You said that AI doesn’t do judgement, it only makes predictions and then hands them off to a human to use his or her judgement to determine what to do with those predictions. Technically, is it possible for an AI machine to take decisions at this stage? How can AI be restricted to only predictions and not judgements?

Ajay Agrawal (AA): Artificial Intelligence is a prediction technology, it’s just computational statistics. It does not have any capability in its current incarnation to do judgement. AI can take actions or make decisions. When we say ‘make decisions’ it’s almost anthropomorphising the machine; in other words, it’s implying that the machine has some human capability. AI can decouple the human from the decision. The human can effectively make a decision to enable the AI to act later in time. You can think of it as decoupling the human from the decision in the time dimension. The AI is never making a decision, the AI can be acting on the human’s decision that was made at a prior time.

For example, I just had a meeting with ICICI Lombard. They built an Artificial Intelligence that works on pre-authorisation, it is used when someone is in the hospital to pre-authorise a payment for their hospital care. The AI isn’t deciding; the AI makes a prediction and the human at ICICI Lombard has pre-decided that if the prediction is above a certain threshold, in other words, if the prediction is with 95% confidence that this medical care is covered under their insurance policy, then they will pay for it.

To an outside observer, it looks as if the AI is deciding, but really, someone at ICICI Lombard has decided previously – and given the authority to the AI – that if your prediction is above this threshold then pre-authorise the care. So that is how you can think about the division of labour between humans and machines – in terms of prediction versus judgement.

The other key point here is about when a judgement is made multiple times. Think about autonomous cars. You are driving a car at 60 km per hour and are approaching a Stop light. It’s dark outside, and it’s raining – so the roads are wet – and you’re 100 feet away from the intersection. The question is: should you step on the gas or the brake? That is human judgement.

When the AI observes the human making a judgement many times, it can eventually learn to predict that judgement. So what used to be judgement shifts to the prediction column because now it is predictable.

One way you can think about judgement and prediction is that the judgement is the part of the decision that relies on sparse data. When that data is no longer sparse, when we’ve got lots of examples, then it can move from judgement to prediction.

GH: Many manufacturers in India, particularly the micro, small and medium enterprises (MSMEs), are very interested in implementing digital technologies to solve problems. They have the means to absorb such technologies – unlike the behemoths and their inertia about absorbing AI-like technology. What initiatives can you suggest to the government and the MSMEs so that they can easily assimilate digital manufacturing & AI?

AA: Never over-estimate the innovative capabilities of large companies. I meet with many large companies; some of them have a lot more trouble implementing new technology, which you might attribute to just inertia. The employees of these companies don’t want to use the technology because they’re afraid it’s going to take away their jobs. Others have a certain power or ‘kingdom’ inside their company which they don’t want taken off. There’s a lot of resistance inside big companies and so, they’re often not as fast as you might think.

Secondly, take some types of technological areas, such as AI. Some areas lend themselves very well to big companies, and that’s often because they’re aligned with the incentives of what the company is trying to do.

But other technologies that are not well-aligned – in such areas, the large companies can be quite small.

Therefore, the smarts, the intelligence, the creativity of small companies must be in figuring out those areas of application of the technology where they have an advantage. Their small size gives them an advantage. They have to find those areas where they can establish a foothold and then build their business.

With respect to the government, my view is that its role is in solving market failures. Where the markets aren’t functioning properly – for whatever reason – the government’s role is in regulation. How to handle the security of data because that security is like public infrastructure. So security for data regulation around privacy: how to enable trade across borders so that foreign companies that have different privacy regulations can still participate and deal with Indian companies without being worried about the security of their data. The government can play a very big role in these areas.

GH: The MSMEs will be very happy to hear what you just said!
We in India realise that we are far behind China, the U.S. and Canada in the AI race. There are plenty of options in the standards we can adopt, being a country that believes in strategic autonomy – we can use what Canada, the U.S. or China provides. But we also want to develop our own standards. What model should India adopt – a Canadian, Chinese or American model, or will you ask us to be as diverse as possible? And secondly, how do we fit all these models into India’s industrial scene?

AA: That’s an excellent question. The first part of my answer is that there is an opportunity for India to emerge as the leader in setting standards, not a follower. And the reason for that is that these countries, except for China, are ahead of everyone else. Everybody else is struggling, trying to establish what their new regulatory regime will look like – whether it’s for areas of application, such as driverless cars, autonomous vehicles, in transportation, in medical devices, in financial services, or in the overall general privacy of handling consumer data. In all these areas, there is no free market system that has any well-established regulatory environment yet. It’s just the Wild West and everybody’s trying to figure it out.

There is, therefore, no reason for India to follow anybody. The reason why India can be a leader is two-fold. First, much of the work from the West that’s outsourced to India in services is right at the sweet spot of what AI is good at. And so, India has a domestic leadership in precisely the area where AI is going to develop and flourish.

It’s still very early – just as we were very early in the process: I was just a co-chair along with a professor from Oxford University at a symposium in Washington for AI policy leaders from around the world in May 2019. The question was: what policies should we set to coordinate and cooperate across borders? It was very recent – May – very preliminary, and so, there’s an opportunity for India to lead.

Secondly, this question you asked about standards is going to be far more important than most people realise. The reason for that is that there’s a whole new class of products in not just cars, but in many other areas which we historically have been used to building. Let’s call them inanimate or dumb products that are now becoming so-called ‘intelligent’ or ‘smart’. And in making that transition, those things go from being deterministic to probabilistic. When they’re probabilistic, it means that there’s some uncertainty regarding how they’ll behave in any scenario.

Now, why do we make things that have uncertainty? Because the uncertainty is what makes their intelligence capable, makes them adaptive. But because they’re adaptive, we don’t know exactly how they will react in scenarios that they’ve never encountered. The reason we use them is because, usually, they perform much better than the deterministic version, but occasionally, they behave worse. So how do we handle products that have those types of characteristics?

A good example to draw upon is the pharmaceutical industry, where we have an intermediary who sits between the drug companies and consumers. In the United States, we have the Food and Drug Administration (FDA). And the reason we have them is because their job is to test drugs and see what’s safe for society. And when they release a drug, they authorise it because it meets their criteria: that, on average, it makes society better. But because these products are, in some sense, probabilistic, sometimes they make people feel worse.

This is why when you buy a bottle of drugs in the U. S. it has all the conditions listed on the side: that it may cause this or that, or if it’s an advertisement on TV, it tells you all the potential problems (it may cause). They have to do this, by law, because the FDA has tested it and found that it can cause these problems, but on average, it makes society better.

We are going to need government agencies, like the FDA, or standard-setting bodies, for many new classes of products. Right now, we have them for drugs, but we’re going to need them for many other things, things which are going to become probabilistic and develop characteristics like drugs have.

Standards are therefore going to become much more important. That’s why I’m so interested in your question! I think standards are going to become a defining feature of the countries which get ahead – because those that can set their standards early and standards that are well-designed, will attract companies.

Take an auto company, at the moment. Where can you deploy your autonomous car? You’re not allowed to deploy it in the U.S. and in many Western countries. And the faster some countries get ahead – whether it’s with intelligent cars or intelligent medical devices or intelligent financial service products – all these things require standards, and potentially, regulation.

Professor Ajay Agarwal teaches at the Rotman School of Business, University of Toronto. He is also Research Associate at the National Bureau of Economic Research in Boston, United States.

This interview was exclusively conducted for Gateway House: Indian Council on Global Relations. You can read more exclusive content here.

For interview requests with the author, or for permission to republish, please contact outreach@gatewayhouse.in or 022 22023371.