Mind Matters Natural and Artificial Intelligence News and Analysis
desert-locust-schistocerca-gregaria-is-a-species-of-locust-a-periodically-swarming-short-horned-grasshopper-in-the-family-acrididae-stockpack-adobe-stock.jpg
Desert locust Schistocerca gregaria is a species of locust, a periodically swarming, short-horned grasshopper in the family Acrididae

AI Tool Now Predicts Attacks of Locust Swarms for African Farmers

Under the right circumstances, data from the past can be used to predict data in the future

A new free AI tool now forewarns African farmers about impending locust attacks: “Farmers and pastoralists receive free SMS alerts 2-3 months in advance of when locusts are highly likely to attack farms and livestock forage in their areas, allowing for early intervention.”

The Kuzi early warning tool is one of a number of new tools that can predict reasonably expected futures. This sort of forecasting is possible if there is large body of oracle ergodic data to train machine intelligence.

“Oracle ergodic” simply means that data from the past can be used to predict data in the future. That’s not self-evident. Flipping a coin, for example, is not oracle ergodic in the sense that a history of past flips tells you nothing about the outcome of future flips. Tick data from the Dow Jones Industrial Average is not by itself oracle ergodic in day-to-day trading. Unforeseen events could transform the stock market overnight. There are, however, a wealth of problems that can be characterized by oracle ergodic data. Forecasting behavior of locust swarms in the future is one of them because locusts, like other insects, will tend to do in the future what they did in the past.

The earliest example of this type of forecasting using machine intelligence of which I’m aware is that of Stanford’s neural network pioneer Bernard Widrow. Widrow (pictured) trained his ADALINE (Adaptive Linear Neuron conceived in 1959) to forecast the weather. Using seven years of historical pressure patterns provided by the San Francisco airport, ADALINE would forecast rain or no rain. Given current weather pressure patterns, ADALINE announced whether or not it would rain tomorrow. Widrow’s neural network outperformed the official weather forecaster at the time.

Apart from this accomplishment, Widrow’s amazing neural network accomplishments in the 1960’s include transformation from spoken word to typed text and near optimal blackjack play. Can you balance a broom on your fingertip? Widrow’s neural network can. All of Widrow’s accomplishments were created by training a neural network with oracle ergodic data.

Widrow reported his accomplishments sixty years ago. Thirty years ago in the second wave of AI, I was coauthor of the first paper using neural networks to forecast load demands for power companies. If power companies do not generate enough power, they must purchase additional power on the market. Likewise, if too much power is generated, a power company must sell the power at the market price. Power companies, wanting to avoid the uncertainty of the market, are motivated to forecast power consumption as accurately as possible.

In one experiment, our power load forecasting neural network was trained on today’s average, low, and high temperatures. The neural network’s output was power consumption for the next day. Historical data provided by Puget Power was used to train the neural network. There are numerous variations of power forecasting neural networks, witnessed by the fact our paper has been referenced over 1500 times. Six years later after the publication of our paper, neural networks were being used by 32 major North American utilities to forecast load. And the interest has not waned. In the last three years (1918–2020), our 1991 paper has been referenced more each year than during any preceding year.

There are many other examples of training a forecasting machine as an intelligent oracle. A neural network can probably be used to forecast whatever you want if (a) there is enough training data and (b) the data is oracle ergodic. Part of the problem is finding the right data to allow machine learning. For rain forecasting, Widrow found pressure patterns were important in the forecasting of rain. Temperature allowed forecasting power load demand.

So what is the secret sauce data for today’s Kuzi tool for forecasting locust swarms?

Using satellite data, soil sensor data, [and] ground meteorological observation, …Kuzi can predict the breeding, occurrence and migration routes of desert locusts…

Muhammed Akinyemi, “AI Tool Will Help African Farmers Fight Locusts Using A Warning And Prediction System” at Space in Africa (January 6, 2021)


Learning from experience (i.e. from historical data) remains one of the most powerful applications of machine intelligence thus far reduced to everyday practice.

About those locusts:

At present, preventive methods rely on aerial fumigation with pesticides but biological controls are under development.


Robert J. Marks II

Director, Senior Fellow, Walter Bradley Center for Natural & Artificial Intelligence
Besides serving as Director, Robert J. Marks Ph.D. hosts the Mind Matters podcast for the Bradley Center. He is Distinguished Professor of Electrical and Computer Engineering at Baylor University. Marks is a Fellow of both the Institute of Electrical and Electronic Engineers (IEEE) and the Optical Society of America. He was Charter President of the IEEE Neural Networks Council and served as Editor-in-Chief of the IEEE Transactions on Neural Networks. He is coauthor of the books Neural Smithing: Supervised Learning in Feedforward Artificial Neural Networks (MIT Press) and Introduction to Evolutionary Informatics (World Scientific). For more information, see Dr. Marks’s expanded bio.

AI Tool Now Predicts Attacks of Locust Swarms for African Farmers