There is only one way to happiness and that is to cease worrying about things which are beyond the power or our will.  – Epictetus

Artificial Intelligence has gripped the attention of many people in the last two years. If you have held stock in NVIDIA, Google, or Microsoft since 2019 I wholeheartedly envy your foresight. If you suffer from bouts of existential angst, then you may be running to refill your prescription for Xanax. Either way I wanted to take a moment to opine and prognosticate on the potential uses of AI in integrated pest management.

The application of AI in agriculture will be in areas that rely on computation and administration. I do not expect any mainstream agricultural institution to roll out a pest management AI. AI that the layperson has in their imagination is called artificial general intelligence. We already have many examples of weak AI platforms around us for the last decade. Predictive and specific task programs such as Siri, Alexa, conversational bots, your google recommendations and predictive typing are examples of weak AI. For pest management, we need a very strong AI.

What would a strong AI platform look like in strawberry pest management? The low hanging fruit for AI applications are in pest monitoring. These systems exist in various forms right now. Pheromone lures enclosed with sticky traps and a camera already report data interceptions of bollworm and diamondback moth. These smart sentinel boxes can report the catches to a user. But if we can correlate trap catches with economic thresholds then it would serve to increase the efficiency of scouts and PCAs.

The ultimate vision would be a fully automated pest management system. In such a system, the sentinel box detects insect numbers over time. Once the economic threshold is reached, the automated platform selects insecticides and mixes them in the appropriate amounts and then proceeds to load a tractor, drone, or an automated chemigation system. The self-propelled sprayer tractor or drone proceeds to the field and either broadcasts over the full area of the field or spot sprays. Or if biocontrol agents are needed, the automated system would order and distribute them on schedule.

It’s a wonderful vision, worthy of an episode of Star Trek or Black Mirror, and unfortunately that is all that it is right now. There are three main obstacles for the daily use of artificial intelligence in strawberry pest management.

1. Insufficient Data

The data for machine learning is not available in the amounts necessary to teach a machine. There are enough makeup tutorials and cooking videos on YouTube that an AI can learn to apply eyeliner and make a fried egg simultaneously if given the physical infrastructure. The resources on pest management are infinitesimal compared with that example. The first thing I am always asked by the engineers is if I have enough good quality data.

Currently, I have a small collaboration with one company to get an automated microscope to count spider mites accurately and it has its challenges. I send them two spotted spider mites, they take the photos, then we sit down and compare what the machine thinks is a spider mite and then I check its accuracy. It’s really good for adults, but weak on the nymphs.

2. The Robotic Infrastructure

The second and the most obvious obstacle, is that we don’t have the optimized physical robotic chassis or bodies for the AI to commandeer and use that can be mass produced for multiple crops. An AI that can drive a car is pretty useless if it doesn’t have the car to drive. There are currently companies such as TRIC robotics and Farm-ng, whom many folks on this blog are familiar with, that are developing the physical infrastructure that can be used to bring this dream to fruition.

Figure 2. An early prototype of the Amiga from Farm-ng designed to deliver predatory mites at a steady rate. Photo captured from video taken by author in June 2024 at a preliminary test run with the Cal Poly Strawberry Center at a site in Watsonville, CA.

3. Regulations, Governance, and Liability

The third obstacle is the governance problem. We would still need humans to be involved as an observer or a guardian of the AI. The thought of an AI Pest Control Advisor is morbidly fascinating and terrifying. Would DPR certify an AI as a PCA? If ChatGPT can pass the bar exam I am sure it can get at least 70% on the PCA exam. Would an AI be held liable in a lawsuit over misapplication because DPR didn’t update its website in time for the AI to update itself? These are the questions that likely will be asked and deliberated on in the next two decades.

Now I am reducing this situation to an absurdly detailed example, but these are problems that need to be addressed first before strong AI is rolled out in agriculture. Teaching an AI to write a novel is surprisingly simpler than finding a Lygus bug and zapping it or picking a strawberry from within a canopy. Unless the capital and knowledge investments are made, these systems will simply not exist. And without the basic machine automation in place, there’s no room for an AI to take over.

In the meantime, kick up your feet and toss a couple of berries down the hatch. You have job security for now, but your neighbor’s kid that decided to go to film school to be a screenwriter; not so much.