Check out all the on-demand sessions from the Intelligent Security Summit here.


From top artificial intelligence (AI) scientists warning that deep learning will push radiologists out of employment, to healthcare professionals heralding that AI will redefine the doctor-patient relationship, to tech executives promising that fully self-driving cars are just around the corner, AI has been marked with plenty of failed predictions in recent years.

Despite the remarkable advances in AI, it has yet to play its transformational role in many industries. However, when compared to other technological milestones such as the steam engine, electricity and the internal combustion engine, it is no surprise that AI adoption is slow.

Ajay Agrawal, Joshua Gans and Avi Goldfarb, professors at Toronto University and authors of the new book Power and Prediction, believe that we are at a stage where the power of AI is evident but its widespread adoption has yet to come. And to better deal with the challenges that stand in the way of leveraging the power of AI, we must understand not only the applications where it is used but also the systems in which it operates.

Point solutions and systems

In Power and Prediction, the authors simplify current AI technology as software that can predict outcomes, such as whether a customer buys a recommended product, or a financial transaction turns out to be fraudulent. Today, there is no doubt that machine learning (ML) models have reached the point where, with the right training data, they can make impressive predictions. 

Event

Intelligent Security Summit On-Demand

Learn the critical role of AI & ML in cybersecurity and industry specific case studies. Watch on-demand sessions today.

Watch Here

However, when it comes to integrating the power of prediction machines in applications and products, there are different levels of challenges that organizations must overcome. 

“AI’s technical advances were and continue to be very impressive. So it is natural to expect that their applications may grow at the same pace,” Agrawal, Gans and Goldfarb told VentureBeat. “They haven’t, and our research set out to find out why. We ended up thinking beyond the AI point solution that people were focusing on, thinking about the practicalities of realizing value from AI in current systems. It was clear that there was a problem. To really use AI you have to be open to a wider set of actions but, for many organizations, they are not prepared for that.”

Point solutions are the low-hanging fruit of AI. These are applications where organizations are already doing prediction. One of the examples that the authors  mentioned is Verafin, a Canadian company that uses AI to predict fraud. Based in St. John’s, Newfoundland, Verafin became Canada’s first AI unicorn, acquired by Nasdaq at $2.75 billion in 2020. Neither Verafin nor St. John’s was on the radar of analysts who were making predictions about commercial AI in Canada in the previous years. 

The reason for the success of Verafin is that it implemented an important AI point solution. Predicting fraud has always been an important part of the work of financial institutions, and replacing their old systems with an AI-powered solution that provides better predictions required minimal changes in organizational structure. 

In other domains, AI adoption requires changes not only at the technology level, but also a fundamental redesign at the systems level, including product, organizational structure, company goals, alignment of incentives, and other aspects of businesses. This makes it much harder for companies to adopt AI to its full potential.

“Our focus on the possibilities of prediction machines had blinded us to the probability of actual commercial deployments,” the authors write in Power and Prediction. “While we had been focused on the economic properties of AI itself — lowering the cost of prediction — we underestimated the economics of building the new systems in which AIs must be embedded.”

The “Between Times” of AI adoption

Agrawal, Gans and Goldfarb describe the current situation of AI as the “Between Times” of AI, which means we are between the demonstration of the technology’s capability and the realization of its promise reflected in widespread adoption.

There’s precedence for this. In the 1890s, the main value proposition of electricity for manufacturers was saving fuel costs because people thought of systems from the perspective of the steam engine. But electricity wasn’t just a cheaper steam engine. Its main value was decoupling energy from its source. You no longer needed to have a steam engine installed next to your factory. But this was how most factories were designed and it took until the 1920s for this potential to be fully realized. By that time, new factories were designed with the idea that the power generator could be located miles away, and electricity could be brought to any point of the facility with a cable or a power outlet.

AI scientist Andrew Ng has described AI as the “new electricity.” And Google CEO Sundar Pichai has said that AI is “more profound than electricity.” They are probably right. But in the Between Times, what we’re mostly seeing is the adoption of point solutions, such as ML-powered fraud prediction, video transcription, image classification, etc.

“We are at that stage where, if AI is going to be transformative, we will start to see the seeds of that transformation soon. It will likely first come from startup ventures utilizing AI to launch completely new business models,” Agrawal, Gans and Goldfarb said. 

Currently, incumbents are the winners of point solutions. But history shows that established organizations are slow to adopt the systems changes that new technological revolutions require.

“Startups have an advantage in that they do not have to change the old. They can start from a blank slate,” the authors said. “But, at the same time, history is telling us that current business leaders should be even more vigilant in understanding AI’s transformative potential.”

For example, with several centuries of history, the insurance industry has much to gain from AI. Big insurance companies are already using AI point solutions to tackle some tasks such as calculating premiums and processing claims. However, the real opportunity of AI challenges the business models built around maximizing premiums and reducing claims. New insurtech companies can create entirely new systems and workflows that use AI to predict and mitigate risk instead of transferring it from one party to another. 

“The disadvantage for startups is that it is rarely the case that current incumbent firms offer no value for the new system. Thus, at some point, that will become a challenge for them. In the past, this has led to a round of mergers and acquisitions,” the authors said.

The future of AI adoption

While the tug-of-war between incumbents and startups continues, what’s for sure is that the full potential of AI has yet to manifest itself. And the future of AI will probably be new applications and new systems that are fundamentally different from what we’ve seen today.

“We believe that there are many more opportunities still to be had by adopting AI as point solutions or applications that are not too disruptive for enterprises,” Agrawal, Gans and Goldfarb said. “The real transformation can only come when the technical advances in AI are so pronounced that it is worthwhile to consider building new systems around them. We are hopeful that time will come but there is plenty of value to be had on the ‘smaller’ side of the technology before that point.”

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.