How to Address the Challenges of Data Science in Agriculture
Recently, Joe Byrum wrote an excellent article on the state of machine learning in agriculture. In it, Joe highlighted one of the biggest challenges facing FMIS and other tech venture backed groups:
“… most agricultural technology startups today are pushed to complete development as quickly as possible and then encouraged to flood the market as quickly as possible with their products. This usually results in a failure of a product, which leads to skepticism from the market and delivers a blow to the integrity of Machine Learning technology.”
A machine learning algorithm can do amazing things that a human brain cannot scale to. It can create decisions combining immense amount of data from weather, soil, plant science, imagery and chemistry in a matter of seconds. The potential is immense with this technology but an algorithm is only as good as the insights that a human can provide it. By rushing it out the door and not doing due diligence on understanding the particular field, the model fails and grower sentiment wanes.
What’s the potential when machine learning works?
Industries such as healthcare have already seen some of the benefits from machine learning but what about an industry with exposure to weather? Uptake Technologies, with one of its strategic partners in the construction space, ingests hundreds of different datasets from sensors on trains and construction equipment engines. The OEM supplies the data, Uptake analyzes and applies its algorithm and the end result is a product in the OEM’s platform that accurately predicts when an engine is going to fail days before it does. How? The first step was consulting with the OEM and all of their dealers, technicians of 15+ years, maintenance managers, and rail yard managers to understand what the data could tell them.
Where machine learning/data science big tech needs help in ag is with two things: 1) getting access to large amounts of data, and 2) working with growers and ag tech that not only know their fields, they can explain the problems they deal with on a regular basis. Most growers are drowning in data because at this point a lot of products that exist today are data visualizers meant to bombard the grower with data for him to process and make the decisions. The growers and agronomists that have thrived with using these tools are extremely valuable because they hold the key to developing an enriching machine learning algorithm.
One road block to advancement of machine learning is that ag tech feels threatened by big tech. Big tech shouldn’t threaten ag tech, rather it should be welcomed as a symbiotic relationship. Big tech has the potential to stick to their strengths but it needs to look at the data in order to come up with the machine algorithms and fine tune them with end users (growers and agronomists). Ag tech has spent years establishing themselves as a trusted adviser and has deployed the man power needed to do so. Plus, they’ve had the most experience designing the products for growers. By sticking to their strengths, FMIS and Big Tech can share the risks and reduce the strain on FMIS development teams by allowing them to focus on products while big tech takes on a larger portion of the data science. That way, they combine each others strengths and produce insights that are valuable to the grower.
Joe’s article is one of hope for the potential of machine learning in agriculture but also one of honesty on the state of the science. The key takeaway is the need for honesty and collaboration with the growers, ag tech and big tech.