AI optimizing the food supply
PRODUCTIVE

AI-Optimizing Our Future Food Supply

Highly technical agriculturalists are hoisting their pitchforks, demanding that we embrace their ideas to reshape our unsustainable food system. Whether our reality is a food desert or a farm-to-table paradise, it’s time we start paying attention: In order to keep pace with a global population that’s expected to surpass 9 billion by 2050, projections show food production would have to increase by 70 percent between 2005/07 and 2050.

AI optimizing the food supply

Researchers at San Francisco-based artificial intelligence company Sentient Technologies are answering the call, working with a major U.S. university to leverage data-powered technologies with the goal of creating ideal food-growing climates in a world disrupted by global warming. Since 2015, the university has been building one such technology: high-tech “food computers.” These sensor-packed and actuator-controlled greenhouses were designed to test out growing conditions for various fruits and vegetables. Researchers teamed up with Sentient, whose evolutionary optimization technology is an artificial intelligence system based on the principles of natural selection.

“If you were able to change anything at will in the growth environment, what would you do?” This was the question asked by Risto Miikkulainen, Sentient’s VP of research and a computer science professor at University of Texas at Austin. Using evolutionary optimization, Sentient and the university are collaborating to answer this question over and over and over again.

When AI Controls for Climate

An evolutionary optimization model starts with potential answers to a given question. For the experimental first phase of the partnership, which began in 2016, the software model simulated the growth cycle of a batch of basil. The question was how to grow the most flavorful basil, and the potential solutions were “climate recipes”—combinations of light, humidity, nutrients and other variables.

Using data from the university’s “food server” (a shipping container filled with food computers and basil), Sentient employed evolutionary optimization to explore which climate recipes would produce the optimal plant. In the case of basil, “optimal” means “tastiest,” but this system can optimize for any trait.

AI optimizing the food supply

In the first simulated growth cycle, most of the climate recipes produced results that the model recognized as bland; but some recipes cultivated relatively flavorful basil. Because of their performances, the latter recipes were selected by the model to reproduce and generate offspring, which were, in turn, one step closer to the desired outcome.

This process—performance, evaluation, culling and reproduction—was repeated for multiple iterations until a handful of well-optimized recipes remained. These improved climate recipes were then returned to the university lab for an actual, non-simulated planting.

“Most of machine learning is focused on modeling and predicting what we already know,” says Miikkulainen. “But this is where evolution is very different. It’s not imitating anything. It’s trying to discover solutions that work well, and the only bias we put in is that we define the search base.”

On the Agricultural Frontier

Sentient approached phase one of the partnership with a pioneering spirit. The company expected to discover more than it knew was possible—and its expectations were right on target.

In addition to reaffirming a widely accepted agricultural principle about the trade-off between plant size and flavor (smaller plants make for tastier basil), Sentient’s evolutionary model discovered something new.

“Common sense suggests that these plants, the basil, should have some downtime: They should be able to sleep,” says Miikkulainen. “[Evolutionary optimization] very quickly discovered that that was not the case. The plants were much happier when the lights were always on: They were more productive, they were bigger, and they were tastier.”

At Scale, a Promising Future

There’s no telling what Sentient’s optimization model might find as the project expands. Phase one used data from only nine climate recipes—but it’s at massive scale that Sentient’s technology really gets humming.

With a global network of farmers using food computers, food servers and food data centers, Sentient’s model will have hundreds of thousands of climate recipes to optimize. Leveraging Sentient’s distributed computing infrastructure, the model will be able to optimize not only for flavor, but also for multiple attributes at once, such as cost, energy efficiency, and shelf life—all crucial considerations in our resource-starved reality.

Phase two of the collaboration is on the horizon in a newly constructed lab that contains four food servers, each with more capacity than that of the single server in phase one. And with thousands of individual food computers being distributed to schools and built-in homes around the world (currently thirty across six continents—and the information to build them is available to the public), Sentient will soon be able to extend its analysis beyond the university’s lab. The goal is to make food production more accessible and more sustainable.

“The optimization so far is not the bottleneck,” says Miikkulainen. “The bottleneck is really to get the data. And the data is essential.”

Sentient’s Data Infrastructure

For its commercial applications—such as optimizing investment strategies and personalizing user experience for e-commerce—Sentient is working with almost unfathomable amounts of data that need massive computing capabilities. But when Sentient got started in 2007, cloud management services didn’t exist. So the company turned to an option called distributed computing. Distributed computing spreads computing power across many computers, servers and data centers. By taking advantage of unused processing power in every imaginable location—in a coffee shop, say, or a gaming center—Sentient can host its data in many places, allowing it to run multiple operations at once in an extremely flexible environment. At this stage in the university collaboration, the data is light enough that Sentient can run its optimization on a local server. The data is produced in a lab, aggregated in an open database, shared with Sentient, analyzed in Sentient’s San Francisco office, and returned to the lab in the form of improved climate recipes. As the project grows, Sentient will have to tap into its distributed resources in order to analyze climate recipe data from around the world.


This content is produced by WIRED Brand Lab in collaboration with Western Digital Corporation.