Throttle AI Innovation at Productivity’s Peril, Says AI Expert

How do you regulate an everything-everywhere innovation like artificial intelligence (AI)?

The White House’s executive order on the safe, secure and trustworthy development of AI is a good a place as any to start, Avi Goldfarb, Rotman chair in artificial intelligence (AI) and healthcare and a professor of marketing at the Rotman School of Management, University of Toronto, told PYMNTS for the “TechREG Talks” series, presented by AI-ID.

“There’s a lot to like in this executive order,” Goldfarb said. “First, it seems to recognize the importance of AI to the economy and to a well-functioning government.”

Drawing from his extensive academic research, Goldfarb added that AI was already revealing itself to be a technology with unparalleled potential to enhance our standard of living.

“It is the most likely general-purpose technology to lead to massive productivity growth…the important thing to remember in all discussions around AI is that when we slow it down, we slow down the benefits of it, too,” he said. 

That’s why there is the need to balance AI’s responsible use in government with fostering a competitive environment for widespread AI diffusion and innovation amongst a competitive marketplace that pushes innovation forward. 

“The thing that I found powerful and useful in the executive order is the emphasis on competition. One of the biggest barriers to the widespread diffusion and increased innovation with artificial intelligence is that if we only have a couple of companies that are developing it, and that are allowed to develop it, then innovation could be limited,” explained Goldfarb.  

“Regulation can make competing onerous for small startup companies… We need a robust, competitive environment to benefit most from AI,” he added, acknowledging that the technology’s risks must be addressed by responsible use. 

Disruption Resulting From the Proliferation of AI is Coming

Lost in the hype around AI is the fact that the advancements in the technology are mainly the result of advancements in computational statistics, which have significantly improved prediction capabilities.

“When you think about AI in 2023, it’s not the Jetsons on the bright side or the Terminator on the worse side, it is just prediction technology. In order to think about the near-term risks and near-term opportunities, it is important to recognize that they’re based on computational statistics. And these prediction tools are only going to get better,” Goldfarb explained. 

Already, the strides that prediction tools and machine learning capabilities have made in the past decade alone have resulted in many robust and diverse applications for machine prediction, extending beyond traditional business problems to areas like medical diagnosis and image recognition.

“What is starting to happen is we are realizing that many things we didn’t use to consider to be prediction can be reframed, engineering wise, as prediction problems,” Goldfarb said. 

“And what is left, if machines are doing the prediction problem aspects of your job well, are the other parts of decision making. Prediction isn’t thinking,” he added. “But it could lead to a situation where an algorithm behind an AI recommendation engine can predict what kind of items a customer might want better than a clerk on the ground could.”

That’s because the growing use of AI software capabilities does not necessarily mean the technology will take over entire industries, but rather that its applications will enhance existing business decision-making processes.

Decision making is where the real disruption from AI will occur, and decision making is everywhere.

System-Level Changes are Needed

Transitioning to AI in healthcare, Goldfarb discussed both the immense potential and current challenges.

“There is no industry where AI has more long-term potential, and maybe no industry with as many short-term challenges,” he explained. “There are lots and lots of people thinking about AI applications in healthcare. And then if you look at the data of actual AI adoption in healthcare, it’s extraordinarily low.” 

Medical diagnosis is a prediction problem, but Goldfarb noted that “if all we’re doing [with AI in healthcare] is what we always did, just a little bit better, it’s not even worth it to bother.” 

That’s why for AI healthcare applications to fundamentally shift the sector forward, there needs to be many and ongoing system-level changes made to unlock the full potential of AI in improving efficiency and patient outcomes, he added.