AI Studies Tomatoes More Than Wheat

The world grows 1.2 billion tonnes of maize a year. And 185 million tonnes of tomatoes.

Guess which one AI studies more.

It's tomatoes. By a wide margin.

What I checked

I counted how many AI plant disease papers exist for each major crop. Then I compared that to how much of each crop the world actually grows.

What I found

Tomato is the most studied crop of all, with 2,848 papers. That is more than rice, maize or wheat.

Once you account for how much is grown, the gap is wide:

• Tomato gets 11.5 times more research per tonne than maize.
• Cassava, which feeds hundreds of millions of people, gets 0.9 papers per million tonnes.
• Sugarcane, the world's largest crop, gets 0.2. The lowest of any crop I checked.

If research followed what we grow, this chart would climb to the right. It does not.

Why it happens

Free image datasets are the likely reason.

To train a model that spots plant disease, you need thousands of labelled photos. A few public datasets exist, and they are heavy on tomatoes. Almost nothing exists for cassava.

So researchers use what is available. Each new paper makes the next one easier, and the literature piles up around the data rather than around the problem.

Some of the imbalance is fair. Coffee and tomatoes are worth far more per tonne than sugarcane, so extra attention makes sense. But that does not explain tomato outranking every staple crop on earth.

Why it matters

The crops getting the least attention are the ones feeding the most people.

Cassava, sugarcane and barley sit at the bottom of the list. These are also the crops where field conditions are hardest and good diagnostic tools would help most.

One more thing: AI really has taken over

A quick check on the wider trend, since raw paper counts can mislead when publishing grows overall.

In 2005, about 1 in 30 plant disease papers used AI. Today it is about 1 in 7. This is measured as a share, so it is not just an effect of more papers being published.

What this does not prove

• I counted papers, not quality. Many papers does not mean a working tool.
• Tonnage is not everything. Crops differ in price and in how many people depend on them.
• Keyword searching misses papers that never name the crop in the abstract.

The dataset explanation is my best reading, not a proven cause. The way to test it would be to catalogue public plant disease image datasets by crop and see whether paper counts track datasets more closely than they track production.

If they do, the fix is not to ask researchers to care more about cassava. It is to build the cassava dataset.