Ajay Agrawal explains economics of AI and what organisations can do with it

One of the first books that strategy consultants recommend to their clients interested in adopting artificial intelligence (AI) in their businesses is Prediction Machines: The Simple Economics of Artificial Intelligence. Co-authored by Ajay Agrawal, the book says that as the cost of prediction reduces hugely, AI can enable organisations to use predictions for better decision-making, and drive business strategy.

We are meeting over lunch at the late Jiggs Kalra's molecular gastronomy outfit Masala Library in Mumbai, where we order the set lunch. Canadian-born Agrawal is visiting Mumbai on invitation from independent think-tank Gateway House for a talk on the key economic forces underlying AI and explain the trade-offs to decision makers.

He explains that the fall in the cost of prediction means we do more of it better, faster and cheaper as the first course of tempura fried spinach chaat, which combines a traditional recipe with modern technique in line with the restaurant's philosophy, is served. “For example, a bank uses AI in fraud detection, loan sanctions, and know your customer. We also use it is by transforming things that we never thought of as prediction into prediction." So in effect autonomous driving is effectively transforming driving into a prediction issue.

AI increases the value of complements such as data, human judgement or action and decreases that of substitutes like human prediction. “So when people ask if humans are going to be necessary, the answer is that our value goes up as we can apply our judgment.” The value of sensors in driverless cars or that of the machinery used in processing medical images too increases.

“Companies that get an early lead in AI have an opportunity to get a big lead and establish market power,” he says. It is because of how AI, works — it learns through feedback. Google search gets stronger from every link a user clicks, and Uber and Lyft gain market share at the cost of local taxi operators.

The next course is a small portion of pimento soup, followed by goat cheese kebab and roast pepper ketchup being the chef's twist. Agrawal, a professor of strategic management, entrepreneurship and innovation at University of Toronto’s Rotman School of Management, wears many hats. Besides teaching, research and conducting workshops on AI, he also heads the Creative Destruction Lab (CDL), a not-for-profit programme that helps founders transform science-based projects into scalable businesses. He is a research associate at National Bureau of Economic Research (NBER), Boston, and advises the US government on science and innovation leadership for the 21st century. As an entrepreneur, he has co-founded Kindred, a company that builds intelligence in robots, and Sanctuary, which aims to create robots with superhuman empathy.

Agrawal received a bachelor's degree in engineering, and followed it up with a dual masters in engineering and business at the University of British Columbia (UBC) in Vancouver. He became passionate about economics and decided to pursue a PhD after a course in economics of science and technology. “This course and another one in venture finance changed my life.”

A lot of his current work has roots in his PhD at UBC in the economics of technology transfer — taking inventions out of university labs and bringing them into practice. Just as he was starting his research, an MIT professor contacted him saying she too was interested in his area, so he worked with her on his thesis on inventions from MIT. “The early stage commercialisation of university science has been my focus ever since,” he says.  

Canada has been at the epicentre of the recent renaissance in AI, he says. Three decades ago the University of Toronto hired Geoffrey Hinton, who was working on the “lunatic fringe” of computer science (machine learning, neural networks and AI were not mainstream). But after two major breakthroughs by Hinton, one in 2006 and the other in 2012, more and more graduate students and post-doctoral researchers came to Toronto from all over the world. Today, one can trace the lineage of some of the most powerful corporate AI teams to the University of Toronto. Heads of AI at Apple, Facebook, Uber’s driverless car unit, Tesla’s autonomous driving group, and OpenAI have all spent time at the university.

Due credit goes to the Canadian government, which kept funding the research when nobody else was, Agrawal says. But he regrets that Canadian companies and venture firms were not quick to respond or were not technologically-sophisticated.

Agrawal had already founded CDL in 2012 to change Canada’s weak record in commercialising science. “When we started, we had a goal of creating $50 million in equity value from companies that were started from the lab in five years. We just finished our seventh year and we have crossed $4.5 billion in equity value.” There are 373 companies participating in the programme this year and 919 have been through the programme since its founding. Today, top entrepreneurs, business leaders and Silicon Valley investors spend one day every eight weeks at the lab to mentor start-up founders. CDL has also opened the programme at Oxford University in the UK.

He is particularly excited about CDL company North, which makes augmented reality glasses. “It's a platform technology, and if the market responds, it will take off like the mobile phone or the watch. It's another interface that could transform how humans and machines interact,” he says. Atomwise, a CDL alumnus from the first batch, enables AI for drug discovery and can predict which molecule will most effectively bind with which protein. “You don’t need to experiment with test tubes as it can be done digitally.”

He says India must use AI in its outsourcing industry. “The key ingredient to train the AI models, which is the training data, is being produced here. And this is the moment in history where these should be developed here and owned by Indians,” he explains. Some of the work that’s being done here, such as assessing loan applications, claims processing, reading scans and medical images, will ultimately be shifted to machine intelligence. If companies don't move quickly then that will be owned by foreigners. 

India should be doing in services what China is doing in manufacturing. “China is not saying we became the world’s manufacturing hub because we have low-wage labour and so we don’t need AI,” he says. It is investing heavily in robotics and bought more robots than the rest of the world last year despite having all the labour. Even as manufacturing is shifting from humans to robots, it's not leaving China because it is investing ahead of everybody.  

We skip the restaurant’s signature, a palate cleanser, and start the main course of aaloor dom with mustard paste and black dal. No AI conversation can ignore the impact of jobs and incomes, where he says three things will happen. “Jobs that use AI will improve productivity and wages will go up. For example, a surgeon may become more effective, or conduct surgeries in remote locations. Second, for some people it reduces wages as it increases labour market competition — a taxi driver’s knowledge of geography cannot command a wage premium today. Some of these jobs will disappear too.

“Where I think everybody is still holding their breath is what the new jobs will be and where they will show up. It’s hard to imagine because we haven't seen them, but economists are counting on the creation of new jobs. But if it's only replacement of jobs, then we're in a lot of trouble,” he says. So, India should be caring more about AI as it stands to both gain and lose more than almost any other country.

Whether AI good or bad is like asking if fire or electricity are good or bad, according to Agrawal. “It’s both. Is human civilisation capable of harnessing the good and keeping the bad at bay? I don’t have the answer,” he says as our lunch comes to an end.