A beekeeper holds a shelf of the bees in their hand
An arrow pointing leftHome

Saving the bees just got a boost from AI

  • Justin Caffier

A machine learning model that predicts pesticide toxicity promises to offer salvation to our long-suffering buzzy buddies.

The plight of the bumblebee has been relatively low on our collective priority list as of late, what with the daily harbingers of impending societal collapse and climate apocalypse. For a minute there, the world seemed ready to listen to the increasingly concerned scientists raising alarm bells about what we now know as Colony Collapse Disorder: the mysterious phenomenon of worker bees suddenly abandoning their colony, leaving just the queen and few caretakers behind.

Our enthusiasm for this cause peaked in 2014 with the hashtag activism campaign #BringBacktheBees, in which Honey Nut Cheerios put a chalk outline of their mascot on boxes for a spell. And then, embarrassingly but predictably, our attention drifted elsewhere. The magnificent creatures responsible for pollinating a third of everything we consume were facing an existential crisis and the best we could muster was giving them their own #KONY2012.

Thankfully, the scientific community is less capricious than the general public. With enough government, economic, and ecological institutions grasping the legitimacy of the crisis at hand and supporting the cause, researchers and bee advocates have tirelessly worked to crack this enigma for almost two decades. Today, many unknowns remain about the granular details of Colony Collapse Disorder, but it’s possible that bees may still have a chance at a comeback. There’s little consensus among researchers beyond no single contributing factor being to blame. Of the many possible culprits studied, some of the most damning evidence for causality has been produced for man-made ones like climate change and pesticides. Thanks to these revelations, bees just landed a powerful AI ally.

Researchers at Oregon State University have trained a machine learning model to analyze the molecular structure of proposed insecticides, herbicides, and fungicides to determine its potential danger to bees. The project team, led by Cory Simon and Xiaoli Fern and with support from the National Science Foundation, recently published their findings in The Journal of Chemical Physics.

Using data from experiments on honeybees that measured the toxicity effects of over 400 different pesticides, the team taught their algorithm to extrapolate possible new pesticides that would be created with additional molecules added and then determine if those novel compounds would be harmful to bees based on their similarities to the existing pesticides.

The molecular structure of pesticides manifests in the form of a mathematical concept referred to as a “random walk,” where each step of the molecule chain is determined by chance. Big Ag loves these as all that variance allows for the creation proprietary and, thus, patentable compounds. But a big drawback of playing mad scientist with the building blocks of the universe is not knowing how the additional molecules will affect living organisms.

By mapping these random walk structures onto graphs, the OSU team’s AI is able to recognize the existence or absence of patterns in hypothetical or candidate pesticides. In an ideal scenario, where it were somehow mandated that companies must submit proposed insecticides, herbicides, and fungicides to be screened by this model, chemical researchers would be alerted to the likely toxic effects of their compound before it had a chance to go to market and decimate colonies in the wild.

Though this is not the first instance of AI being employed in the fight to save the bees, it’s arguably the most proactive one developed thus far. With the clock ticking, and quarterly profits still valued at a premium over long-term solutions, bees are going to need all the help they can get, so we can only hope the OSU research team’s accomplishment inspires other creative AI assists for the fight. But how about we leave the #AIforApianIntervention hashtags in the brainstorm meeting this time?