Preventing Bias in Agricultural AI

This is Tim Hammerich of the Ag Information Network with your Farm of the Future Report.

As more growers turn to digital tools for decisions on the farm, one of the emerging challenges is making sure those technologies stay unbiased. John Kempf, founder of Advancing Eco Agriculture and a leader on the development of FieldLark AI, says building fair and reliable AI for agriculture may require taking a step back and developing a broader perspective.

Kempf… “ There's all types of bias that can occur there. There can be just the bias of having 60 peer reviewed papers saying one thing, and then three or four other papers saying something that is in direct conflict, in direct contrast to those. You also have a bias of recency versus historically. So synthesizing all that is, you have to think through fairly carefully how to, how to evaluate those. But it was given the direction of removing all biases. FieldLark doesn't make recommendations for anhydrous ammonia, for example, or potassium chloride. Because it's looking at the broad array of evidence and looking at the additional externalities and saying there's much smarter ways and much better ways of applying nitrogen than using anhydrous ammonia. So we haven't given it any instructions, for example, to take a biological agricultural approach or to give it a region of agriculture approach, because it needs to be bigger than that. The training needs to have a truly macro and global perspective, and not a biased perspective. But in the process of removing bias, some users perceived the presence of bias, which is fascinating–a reflection of our own biases that we have when we're using it.”

FieldLark AI is an artificial intelligence system that has been specifically trained to help with on-farm questions and decision-making.