Trillions of dollars are being spent on AI and generative AI, but the real value, Dr Chad Jennings says, lies in narrowing down data for on-farm decisions

Artificial intelligence (AI) and generative AI (gen AI) would bring “mind-blowing” changes, Almanac’s Dr Chad Jennings told the GRDC Grains Research Update in Perth.

Dr Jennings, who has spent more than two decades working on geospatial analytics, recently joined Almanac after working with Google Cloud. The Canadian-based Almanac began in late 2024 with the merger of Semios and the Australian farm information management system Agworld.

Dr Jennings said trillions of dollars were being invested in these technologies and many would be repurposed into agricultural solutions.

Data and decisions

“Agronomists and growers face a new challenge – an overwhelming amount of data, yet a gap in actionable decision-making. It’s wisdom that makes the call.”

AI data analytics and geospatial analytics could play a “simplifying” role, giving the “right data to the right person in the right quantity and at the right time”.

AI and gen AI were the biggest technological disruptions seen in the past half-decade
with an estimated US$5.7 trillion being spent annually on them, an amount that increases
by US$500 billion each year.

The value, he said, was “sitting between this crazy disruptive investment and what we all do on any given Tuesday”. That is, problems need to be solved and questions need to be answered.

Canola example

Using Agworld and spraying canola as an example, Dr Jennings walked the audience through how AI and gen Al worked.

With the technology available now, growers and agronomists can use Agworld to look into spraying options. After inputting information, label data can be downloaded into the management system.

The next step, using Alma – the generative AI assistant within Agworld – already sees that label data, as well as all of the other products supported in the system, displayed.

“It essentially brings up a sidebar with recommendations and key pieces of information, instead of having to go through the label one row at a time or one detail at a time,” he said.

“One thing that we haven’t done and that we don’t intend to do is have gen AI create the recommendation automatically.”

The reason why is accuracy. “The accuracy rate is getting better, but it’s not perfect. Generally, it’s about 70 to 80 per cent. Three years ago, it was 60 per cent, so it’s getting better quickly.

“However, we definitely need humans in the loop to verify what the results are.”

Outside information

To help with recommendations, the system includes references. “So, if you want to see where the model is getting a particular bit of data, then you can go to the exact right place in the PDF and find it and verify it for yourself.

“Labels don’t have to be the only repository that we take advantage of for the benefit of
our users.”

He said the future could see, for example, GRDC guides in the Agworld library. Another possibility would be geospatial indexing. “This is one place where current large language models are not particularly good. Adding geospatial awareness to that generated AI intelligence can also add a new level of recommendation.”

More information: GRDC Past Events.