2023 will go down in history as the year Artificial Intelligence burst from the realms of science fiction and out into reality for everyday people. Conversations about AI echoed across pubs, board rooms, universities, and dinner tables around the world. In these rapidly changing times, there is both optimism about the future AI could help create and concern for the impact that disruptions like AI inevitably have on our jobs and society.
Nowhere are these views more hotly contested than in agriculture.
Food Agility is investing in research across a range of projects, from yield prediction of major horticultural crops to carbon measurement, grazing planning and disease detection of livestock, each working to unlock the potential of AI for the sector – and understand it’s limitations.
It’s an important question with a counterintuitive answer: To understand the potential (and limits) of AI for ag, we need to first unlearn the way we comprehend agriculture. That is, we need to pause our big-picture thinking of agriculture as a system, and instead focus in on the food production chain as the sum of its parts.
Let me explain.
There is a long list of complex and dynamic variables that shape a healthy, profitable, and sustainable agricultural business. Think of these variables like pieces of a puzzle.
These pieces are always evolving. Sometimes pieces will need to be added or removed or may change in their size and significance depending on the growing season and the business lifecycle.
Why can’t we just look at the puzzle as a whole? Because AI is data-driven. It is only able to interact with the pieces of the farm-puzzle that have high-quality, reliable digital records. When data is good, AI can be great. It can find complex interactions between elements in a growing system and piece the puzzle together continuously.
However, with current technological limitations in our sector and related fields like meteorology, we are never going to have every high-quality piece of the farm-data-puzzle.
Given this reality, producers and supply chain actors are faced with an optimisation problem: which puzzle pieces should they prioritise? Will installing new field sensors, upgrading management technology, adding attachments to existing machinery, etc enable AI to improve their business and bottom line?
The cost-benefit analysis is also nuanced. It requires an appreciation of the types of value that AI could unlock for an individual business and an understanding of the costs ($ and time/effort) involved in securing the relevant data streams.
It’s also important to remember that collecting pieces of farm data isn’t a binary 'yes’ or ‘no’ equation. There is a sliding scale of data quality which has critical implications for the capacity of AI to deliver actionable insights.
Any piece of the farm-data-puzzle needs to be evaluated for key quality characteristics, such as completeness, accuracy, consistency and validity. The puzzle pieces ideally need to be in the same scale, frequency and duration as the puzzle pieces they are connecting to. Crucially, they also need to be able to connect with other pieces, and not be treated as though they are the entire puzzle, as has been seen with some new tech in the past.
Recently we have been seeing great examples of leading Australian agtech players breaking down some of these boundaries between systems and datasets, such as AgriWebb launching their Marketplace Integrations service which enables their customers to connect in live data relating to water management, pasture availability, and animal growth records all under the one farm management application. Additionally, The Yield’s partnership with global robotics manufacturer Yamaha also offers a glimpse into the kind of exciting potentials that will arise when we harness connections between complementary technologies.
Ultimately, we are finding that the decisions that need to be made regarding data capture and AI application rely on the wealth of knowledge that has been accumulated in our farming community over generations. AI won’t overcome the need for this expertise any time soon.
Plus, increasingly volatile factors such as changing climate, shifting global markets, and evolving disease pressures, mean that AI is unlikely to ever reach the point of omnipotent crystal ball-style predictions. AI is, after all, only as good as the historical data that it is fed and the quality of any contingent predictions it relies upon (such as climate/weather forecasts).
Understanding the limitations of AI is essential to harnessing its value. While AI isn’t a cure-all, we are seeing tremendous outcomes when AI is applied intuitively and responsibly across our research projects. To find out more about AI and its applications in ag, take a look:
For more information reach out to AI & Robotics Lead Ash Rootsey via firstname.lastname@example.org or online via @ashrootsey.