How to Apply Discipline to AI Choices in the Quest for Innovation

Andrew Binns, co-founder of Change Logic consultancy, on how leaders can resist hype while focusing on near-term wins and long-term technology trends.

Artificial intelligence offers creative companies new ways to innovate. Yet many organizations are struggling to realize AI’s full potential as they navigate high operational costs and data integration challenges.

Amid these hurdles, leaders are under increasing pressure to demonstrate a tangible return on their AI investments while championing experimentation and scalability, says Andrew Binns, director of Change Logic, a Massachusetts-based business consultancy, and author of books including Corporate Explorer: How Corporations Beat Startups at the Innovation Game , and the Corporate Explorer Fieldbook.

In this Q&A, Binns discusses the need to focus on the business purpose of AI investments when evaluating risks and rewards, while considering both short-term priorities and the long-run evolution of the technology.

Binns encourages executives to see AI as a general-purpose innovation—like electricity—with the potential to revolutionize tasks, such as coding, provided companies overcome internal hurdles, weigh the risks of experimentation, and learn to move faster.

This interview has been edited for clarity and length.

Cindy Waxer: What is a good way of thinking about AI as an innovation?

Andrew Binns: It’s really important to understand that AI is a general-purpose innovation like electricity. It’s about enabling other outcomes. Some of the earliest applications are not the point of AI; there’s something much bigger ahead. Using AI for self-coding (software  code that autonomously changes to improve performance), for example, will have immense implications over time. But there’s been a long period of experimenting with what AI even is, and how you train it. This period of familiarization, coupled with the hype cycle, has skewed our perceptions of AI a bit.

We’ve seen people, particularly technologists, come too far away from determining what use is AI for, what problem AI will solve, how will it create value for people and why is it better than what I’m already doing today. Many of the AI applications out there are not any better than how we do things now—they just happen to use AI. 

You mentioned the potential for individuals to rely on AI for self-coding. What are some other potentially valuable use cases for AI?

One use case I found really impressive is what the scientific information company Elsevier is doing with Scopus AI. There’s a long-standing problem of discoverability of scientific data. Many papers sound exactly the same and the most important insights often come from the fringe, not from the core corpus of knowledge. Scopus AI enables users to extract summaries of Elsevier’s repository of scientific literature in a way that enables a researcher to identify relevant sources at lightning speed. 

What barriers must organizations overcome to establish more practical use cases for AI?

If you make decisions about your data strategy based on the way things have been in the past, you will fail. You have to think about where are we headed? How do we prepare ourselves for the shocks that will develop as AI’s potential unfolds? These questions are tough because that could take decades.

We have to learn to function at two speeds simultaneously. Businesses have to move fast yet still build systems that integrate. That’s hard work for your average general manager who likes to have things cut and dried, focus-driven, and goal-oriented. You have to live in this world of contradictions. One of my friends and co-authors talks about a both-and strategy. You need a both-and data strategy, ranking short-term priorities over a long-term evolution. 

There has been plenty of talk about AI’s high data center energy costs and consumption. How will these factors impact the scalability of AI initiatives?

AI doesn’t get any cheaper than more of it you have. This is a really substantial problem. At some point, we’ll start to realize the environmental cost of these data centers, and that will impose additional constraints.

There are also these constraints around affordability and mass adoption that may bar some players from entry. There are signs of constraining capacity, but I don’t know when there will be visible limits either on the pace of innovation or the cost of scaling an innovation.

One of the key points I make about innovation is that most of our attention goes to the early stages of ideation and incubation and the bright shiny objects. Actually, innovation is about new ideas applied at scale. Uber displaced taxis in many cities not because they came up with the idea for the app but because it spread to many cities and scaled.

AI success requires experimentation, yet experimentation can result in epic fails. How can organizations increase their changes of realizing a return on their AI investments?

The key questions are, what are the critical assumptions that underpin your investment? What needs to be true for my investment in AI to work? You have to be aware of all of the critical unknowns and the assumptions you have about your business model and test them one by one.

The secret is moving forward in small verifiable increments rather than only in big, bold bets. Sometimes you get to a point where the big, bold bet is appropriate. But you don’t just start there. AI has to be use case driven. What is the high-value customer problem I’m solving? How do I de-risk my investments by making small learning steps rather than big, bold bets?

How might organizations have to change internally to drive greater adoption of AI?

The pace of organizational adaptation needs to increase massively. It’s this old adage—you know you’re in trouble when the pace of change outside the company is faster than the pace of change inside. You have to get your organization to adapt to this completely new workflow, how management functions and how it configures its assets to deliver a product. It probably means that IT as a function no longer exists.

We have found over the centuries that work is constantly being displaced and reallocated to new places, new jobs, new activities. That isn’t always something we can see from our current vantage point. However, I’m confident that what we’ll find is that new jobs and new tasks will get created as a consequence of AI activity. There is likely to be a medium-term shock but long term, my view is that new tasks, new activities will develop, and new jobs will develop. That will be the future.

How is AI changing today’s global competitive landscape?

In any market, disruption happens because a new player can move at speed when an incumbent is still trying to defend its existing business. But if AI allows this rapidity of development, it could collapse the innovation cycle to a short window. It could become a moment where older firms are actually be able to move faster.

If a large company can learn how to adopt AI for software development, can they move at the same speed as a new entrant? Some of these competitive dynamics may actually start to change in ways we’re not used to. The old path of integration and disruption is one we’re really familiar with but AI might flip the script a little bit where the old stodgy companies will move faster. 

What do you see as the value of AI in the years to come?

It’s going to be about ease and everything becoming conversational. All of our interfaces with the digital world will be conversational and faster but also more customizable. Within 10 years, we will be able to speak to our computers to have them write code to render things for us —maybe even in five years. The degree to which we’ll be able to customize and shape our interactions both as businesspeople and as consumers will be extraordinary. It will accelerate the pace of everything. The best analogy is this notion of general-purpose technology. AI will become like electricity—it won’t perform a single function. It’ll be everywhere.