Reimagining Efficiency: The Case for Slow AI in Agriculture.
In a recent talk by Rory Sutherland, Vice Chairman of Ogilvy and Mather, he explores the assumption that optimization always improves experiences—when, in fact, slowing things down can often create better outcomes. This idea has significant implications for agriculture and technology, where the rush to automate and optimize can overlook the human experience and the value found in inefficiency.
I recently came across this talk by Rory Sutherland, Vice Chairman of Ogilvy and Mather, on YouTube. Much of what he said has been bouncing around in my mind ever since.
He talks about the assumption that everyone wants to optimise or speed up processes when, in fact, great experiences can often be achieved by slowing things down. He mentions train journeys—how many people take pleasure in their commute, even subconsciously factoring this in when choosing where to live. These journeys, when optimised by engineers, can turn out to be miserable. But if optimised by someone like Disney, by imagineers, they could become wonderful experiences, making any other option suboptimal or even foolish to consider. Sutherland also critiques how incentives to solve these problems are often designed to remove humanity, uncertainty, and irrationality from solutions—leading organisations to conform to broken funding mechanisms.
I encourage you to watch the talk yourself. If you follow this blog, you’ll know that I think a lot about these kinds of ideas, especially in the context of agriculture and food. In particular, I’m interested in how technology can be introduced into the ecosystem. Rory touches on system design and mentions the work of W. Edwards Deming, who argues that to optimise a system, you must sub-optimise some of the parts.
This brings up some interesting points when we consider agtech. I think very few agtech organisations and startups consider the overall agricultural and food systems. They tend to focus on the specific problem they are trying to solve, often without fully understanding or mapping the broader farming, food, and supply chain ecosystems. There’s also a rush to optimise everything, and the need to sub-optimise frustrates people on both sides of the conversation.
For example, I recently spoke with a farmer who told me that, despite the push to automate tractors, driving the tractor was one of the most enjoyable parts of their day—they didn’t want that taken away. On the other hand, agtech startups tell me that farmers don’t fully understand or appreciate their engineered optimisations. Neither group said this explicitly, but I believe both need an imagineer on their side.
This also applies to farm management software. We’re reaching a point where farmers feel obligated to use it. Conversations have shifted from “Why should I record data about my business?” to “What data should I record, and how do I choose the best software to do it?” Early adopters enjoy the process of recording data and optimizing their activities, but as farm management software becomes more widespread, it risks becoming an obligation. When the long tail of productivity hits, more software tools flood the market, creating an explosion of choice that will leave farmers permanently unsatisfied. There will always be a new tool with one more feature, a different approach, or a lower price point. Data recording is about to become mundane (or you could argue it already has). At what point do we need to call in the imagineers? What could an imagineer do with mundane ag software?
We’ve talked about technology change, system design, and adoption, but what does this mean for AI in agriculture and food? There’s a lot in Rory’s talk to consider. Currently, much of the work in AI aims to shortcut the time and effort people need to invest in reaching an outcome. Rory argues that prompts like “write me an essay” get you an outcome, but rob you of the process of writing. He believes (and I agree) that the value isn’t necessarily in the outcome but in the process of getting there. Perhaps the real value of farming for farmers isn’t the product they produce but the act of producing it. We kind of know this already, describing farming as a “lifestyle” as much as a career.
What we might be forgetting is that there are things we want and need to streamline, but there are also things where the value is precisely in the inefficiency. The automatic assumption is that faster is better. But beyond data, security, and risk, what does responsible AI look like in agriculture? What are the unintended consequences? What does slow AI look like?
I’ll save my thoughts on slow AI in agrifood for another time, but it’s worth thinking about. What does seasonal AI look like? What about generational AI? How do we leverage generational AI? I'm just remembering that Google leaked the Selfish Ledger design fiction touching on some of these ideas in 2016. Something for me to revisit in 2024. Smushing these two videos/ideas together, maybe a better title for this post would have been The Selfish Gene, Are We Now Too Impatient to Be Intelligent? Maybe i'm driftng too far...