Sterling Platt

SterlingPlatt@qq.com

AT the recent China Hi-Tech Fair in Shenzhen, Prof. He Liang delivered a harsh reality check to the Shenzhen Association for Artificial Intelligence. The “Green Revolution” — the strategy we’ve used since World War II to grow more food by adding more fertilizer, building more canals, and using bigger machines — has finally hit a wall. For decades, this brute-force approach kept the world fed, but we can no longer squeeze more food out of the ground just by adding more chemicals.

This comes at a dangerous time. By 2050, the global population is expected to rocket toward 9.1 billion people. To feed everyone, the United Nations says we need to grow 70% more food than we do today. This is even harder in China, where there is very little extra land left to farm. In fact, the amount of farmland per person in China is only 40% of the global average.

To fix this, Prof. He is calling for a shift to “New Quality Productive Forces.” While the name sounds complicated, the idea is simple. The last agricultural revolution was about quantity — dumping more water and chemicals on the soil. This new revolution is about quality — using data and computer programs to make smarter decisions. In this future, a big harvest won’t depend on how much fertilizer you buy, but on how smart your algorithms are.

The professor in the field

Prof. He is the perfect person to explain this shift because he lives between two worlds: the high-tech lab and the dusty farm. He is a researcher at Tsinghua University, a professor at Xinjiang University, and leads a cutting-edge lab focused on cognitive computing. In 2020, he shifted his focus from voice recognition at Tsinghua to applying his tech skills directly in the fields of Xinjiang.

He defines “smart farming” as something much bigger than just a self-driving tractor. He sees it as a complete nervous system for the farm. It connects sensors, artificial intelligence (AI), and the Internet of Things into a continuous loop: see, think, decide, and act.

This system watches over crop health, carefully measures out water and fertilizer, and spots pests or droughts before they can destroy a harvest.

The good news is that the first step — giving the farm “eyes” — is largely solved. Prof. He points to the explosion of the drone industry. “Ten years ago, putting high-quality sensors in a field cost tens of thousands of yuan,” he noted. “Today, thanks to mass production and cheap drones, the cost has crashed to just hundreds.”

The ‘one-shot’ economy

We have successfully built the “eyes” of the farm with drones and sensors, but building the “brain” is much harder. Prof. He’s work focuses on teaching AI how to make the right choices for a farm.

In the digital world, this is easy. A computer can play millions of games of Go or Starcraft in a few days. It can lose millions of times, learn from its errors, and eventually become a master strategist.

But a farm isn’t a video game. “Agriculture is a low-frequency science,” He explained. In simple terms, you don’t get to hit “restart.” You only get one harvest a year. You cannot ask a farmer to let an AI “fail fast” on their land just to learn a lesson. If the AI makes a bad call — like not watering the plants during a heatwave or using the wrong amount of fertilizer — the crops die. And if the crops die, the farmer’s family could go bankrupt.

This risk makes farmers deeply risk-averse, a concept tech companies often struggle to grasp. Prof. He illustrated this with the cotton fields of Xinjiang. Even though there are cheaper Chinese machines available, many farmers still pay extra for American-made John Deere cotton harvesters.

They are doing it out of fear. “In the 20-day harvesting window,” Prof. He noted, “mechanical failure may cause hundreds of thousands of yuan in losses.” Farmers are willing to pay more for peace of mind.

Therefore, the transition must respect the unforgiving biological timeline and the farmers’ fragile financial reality. This requires a new level of cooperation among the government, scientists, and the agrarian workforce.

Right now, that cooperation is blocked by “data silos.” In other words, farmers and companies keep their data locked away like secret recipes, not wanting to share their soil records or yield numbers with anyone else.

To fix this, Prof. He proposes a solution called “data assetization.” He argues that in the era of smart farming, data is an input as vital as land, labor, or machinery. It is no longer just a byproduct of farming; it is a raw material necessary to create value. But if we want farmers to share this “digital crop,” they need to be paid for it.

Prof. He suggests that the government needs to set a standard rulebook for how this data is bought and sold. This would allow information to flow freely between different regions and machines, creating a shared brain for agriculture — a goal that both China and the United States are currently racing to achieve.

The digital twin:

Manufacturing experience

For Prof. He, solving the problem of “data silos” is only half the battle. Even if every farmer in Xinjiang shared their data tomorrow, a “biological speed limit” remains: nature simply moves too slowly to train a reliable AI. To learn effectively, an algorithm would need centuries of drought, flood, and pestilence data — time that we do not have.

Since nature cannot be sped up, Prof. He’s team chose to simulate it. They constructed a “digital twin” of the Xinjiang cotton fields. They started with a widely used biophysical simulator (DSSAT) for generic cotton.

To make it useful for Xinjiang, they calibrated this simulator to local conditions by integrating historical climate records, specific soil profiles, and the biological details of a local cotton variety.

The accuracy was striking. When tested against the 2023 harvest, the digital twin matched real-world results with 85.6% precision. The team then used this model to generate 150,000 virtual scenarios based on 43 years of climate data.

This created a massive library of “synthetic data.” It allowed the AI to live through the bone-dry heat of 1982 and the severe drought of 2013 repeatedly, effectively learning the consequences of every irrigation decision without ever risking a real farmer’s livelihood.

The need for speed

This synthetic data foundation allowed the team to solve immediate practical problems. An urgent challenge was deciding, in real time, whether to irrigate when water becomes available.

In the arid regions of Xinjiang, water is delivered via canals according to a strict schedule, but emergency repairs, weather shifts, and the needs of other farms can cause sudden disruptions. With so many changing factors, the team found that they needed to simulate 528 possible scenarios to find the optimal decision.

Traditionally, running these complex simulations would take about 42 hours — far too long for a farmer making daily decisions. However, by training on their massive synthetic dataset, the team distilled the model into a streamlined AI that can do the math in just 24 minutes, well within the decision window. In 2024, applying this fast-thinking AI, the cotton yield increased by 8.5% over the baseline, while water use was reduced by 4%.

Escaping the

‘average’ trap

Beyond immediate logistics, the team tackled a deeper strategic problem: the danger of “average” thinking. Standard AI optimizes for the average outcome, which might hide a small risk of total crop failure.

Since these disasters are too rare to learn from in the real world, the team used the synthetic data as a “time machine,” forcing the AI to survive 43 years of historical climate extremes over and over again. By monitoring 17 distinct health indicators of the living plant, the AI learned to prioritize survival over aggressive pursuit of yield maximization.

The results validated this conservative approach. In 2024, this risk-aware AI outperformed traditional practices, boosting yield by nearly 13.6% while improving water efficiency by 6.7%.

The rise of the digital

agronomist

The urgency of these innovations extends beyond yield charts; it addresses a looming demographic crisis. Prof. He highlighted a stark reality: China’s average farmer is now 56 years old. He observed that younger generations are increasingly disconnected from farming.

China’s goal, therefore, is not to replace the farmer, but to evolve the profession. This is the ultimate promise of “New Quality Productive Forces” — a shift that relies on intelligence rather than muscle. Prof. He envisions a future where the physical burden is finally lifted, and the farmer of tomorrow becomes a “digital agronomist.” In this new era, they will manage algorithms from a screen, securing the nation’s food supply through data rather than sweat.