Kiwifruit crops are benefiting from the use of AI.
Somewhere in New Zealand a computer is learning from an expert horticulture pruner the best place to cut a branch. The computer will go on to help a beginner pruner make the right decision.
On a kiwifruit orchard in the Bay of Plenty, researchers are working out how counting and calculating the density of buds and flowers will maximise the harvest.
In that small aircraft above them is a tool to analyse nutrient content and water stress in the foliage, while over the Kaimais in the Waikato, a dairy farmer knows a cow is unwell even though he can't see her.
Artificial intelligence at work in rural New Zealand. Some of it hasn't been commercialised yet, and there's concern New Zealand isn't investing enough and we risk getting left behind by our agribusiness competitors, but AI is alive and kicking in our orchards and paddocks.
If for you this conjures up images of Terminator-types lurking behind the milk tank or in the wheat crop, a definition of this increasingly used, but poorly understood, buzz phrase, may be comforting.
As AI Forum NZ says, it's important that more New Zealanders have a high level understanding of AI so there is a clearer link from the technology to its many potential applications, but it can be difficult to define what AI is.
No wonder when the term is used to refer to everything from neural networks to autonomous robots in sci-fi movies to the search engine you use to look up pictures of cats.
The forum offers the OECD's definition of an AI system. "A machine based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. AI systems are designed to operate with varying degrees of autonomy".
Then there's Dr Mark Begbie's more easily digestible: "A good working definition of AI is the science of enabling computers to make decisions in fuzzy and uncertain circumstances."
And the chief executive of Tauranga-based independent research institute PlantTech is quick to make clear AI is not robotics, but the way a robotic machine goes about understanding the world by using AI.
While robots are examples of AI, so are self-driving cars, smart assistants, disease mapping, predictive texting, automated financial investing, social media monitoring and virtual travel booking agents.
AI Forum NZ again: "The theory and application of AI as we know it today has been around for decades….however many of the most interesting applications of AI today result from approaches that use machine learning."
Time for another definition: Machine learning is a subset of artificial intelligence which often uses statistical techniques to give computers the ability to "learn" with data — creating models to process, interpret and respond, without being explicitly programmed with a predefined set of rules".
Enter AI in New Zealand agribusiness and the concern of some experts. Machine learning is about big data — we need more agricultural and horticultural data and the investment to get it.
There's some exciting AI work going on in the $6.65 billion export horticulture sector but it is to some extent still catching up with agriculture progress, because the horticulture environment is more complex, says Begbie.
He compares that environment to that in which one of AI's early big success stories, the self-drive vehicles, operates. It's extremely variable — even a traffic light is challenging.
"But when you go out into agriculture and horticulture the world is even more variable. At least a traffic light has three lights and you know what colour is at the top and the bottom and it's always a vertical row.
"But if you're looking at a tree or a plant it's a much more variable structure and varies massively over the year — from sticks in the winter to being covered in leaves in summer."
AI is finding success in progressively more complicated problems and challenges, Begbie says, but to create a reliable AI tool, like training a child, you have to be able to show it a variety of life, which is where data capture comes in.
"Horticulture is typified by smaller players so there's less opportunity for large scale data capture…"
However he's far from glum about progress and offers plenty of examples of groundbreaking work under way, such as the earlier mentioned projects.
While we don't have huge industries like the US citrus sector to supply data, we have very substantial and good "living labs" in our kiwifruit, apple, avocado and wine sectors to develop solutions, he says.
Dr Amanda Williamson, a Waikato University management school lecturer in strategy and innovation with a strong interest in AI, is about to launch a national survey to measure how far agribusiness is down the track with AI and what barriers there are to progress. The results will form a white paper.
She says AI is important because the business landscape is increasingly uncertain and requires leaders to analyse complex information at a greater pace than ever before.
"Emerging AI technologies help people make sense of information, provide data-driven insights and made rapid evidence-based decisions — think supply chains.
"With data we can improve processes and performance, by augmenting decision-making and building digital products and services through AI.
"Thanks to digital transformation, AI adoption is relatively and cheap to implement for augmenting decision-making around hiring people and selling products.
"But it is not often easy in the early stages of the value chain in primary industries and that is a big concern I have for New Zealand."
Williamson says unless primary sector businesses can invest in collecting good data that pertains to the early stages of production — or partner with businesses that can do it for them — they are likely to be left behind. "AI is only as good as the data it learns from."
In most industries, she says, businesses do not need to worry about collecting data — they collect it every day because most of their work is digitised.
"Data collection is a lot more complicated when it comes to the early stages of the primary industries' value chain. How do you significantly improve plant production if you are not collecting good data on the variables impacting your plants?"
Often collecting data in the early stages of production requires proactive investment in observational tools, Williamson says.
"The organisations that have invested in collecting data are poised to be the leaders in the market. The intermediaries that own the data may have all the power and leverage in future. My message to businesses: data is a strategic resource, invest in collecting and owning it."
AI will have a transformative effect on primary industries, she says. (AI Forum NZ in a 2019 report predicted AI could bring more than $50 billion in benefits to the New Zealand economy by 2035 — the primary industry is a cornerstone of that economy.)
"Executives cannot yet predict the exact monetary benefits of AI and therefore may provide an insufficient business case for investing the necessary time and resources in its adoption.
"We lack business cases now to illustrate the dollar value of adopting AI and data collection initiatives. AI adoption is not cheap."
Williamson's concerns about lack of investment are echoed by Waikato University's new Artificial Intelligence Institute. Associate director Jannat Maqbool says we are not investing in research and building the talent pipeline needed to develop the models New Zealand requires.
"AI can enable predictions across the supply chain as well as downstream demand for products, meaning we can better plan and prioritise resource usage and more effectively manage and minimise the environmental impact of food production.
"There is a lack of data to train models given the slow pace of digitisation in the sector. This presents a challenge for the short to medium term and is an important consideration in terms of New Zealand being able to move forward in leveraging AI.
"Disparate infrastructure and connectivity in rural New Zealand means we really need to be developing AI to operate on site — at the edge, something that many other parts of the world do not have to consider."
Maqbool says AI can help in enhancing productivity and increasing yields, as well as food quality — important features as the world looks to feed growing populations more sustainably.
Williamson says an obstacle to business decisions to invest in AI is that the benefits are not often immediate. "This is problematic for primary industries, because to continue to compete in the ever-advancing international market we need to make investments in AI in the primary industries now."
Begbie might disagree. He doesn't believe proving the business case for AI is more challenging here than in many other countries.
On the delivery speed of benefits after investment, he cites a PlantTech result on a project for global kiwifruit marketer Zespri. The on-the-vine fruit sizing project, showcased at the recent Fieldays, delivered a credible solution in 18 months.
"We've gone from a standing start to having a credible solution… showing it's scaleable and meets the accuracy requirements of the industry in under two years."
The current kiwifruit sizing system involves manually measuring and weighing fruit on the vine.
Begbie says it barely covers a few thousand fruit and there's high uncertainty in the result.
Working with Zespri this year, PlantTech took the data from four operating AI systems for a week.
It processed the data in two days and gave Zespri calculated weights for around 23 million individual kiwifruit, with four times less error than the manual system.
Why does sizing matter? "It means Zespri can much better target high value markets knowing more accurately what fruit they have. It will bring millions of dollars extra into the economy because they are better able to target markets," says Begbie.
"There are a lot of companies globally trying to count and calculate the weight of fruit while its in the orchard and I can say hand on heart the solution we have developed is the most accurate in the world for kiwifruit."
PlantTech is now looking to apply the system to other fruit crops.
"We are confident will be up there with the best in the world."
Another target for the system is the post-harvest sector.
"If they have a better understanding of what is coming in from the orchard they can better prepare around how much, and which, cool storage facilities to have.
But also there's real potential to release working capital.
"Without an accurate knowledge of what is coming in, there is contingency buying of cardboard packaging, for example, because it varies by the size of the fruit."
Packhouses bring in millions of dollars worth of packaging that may be carried over unused from one season to another.
That working capital could be doing something much more valuable than gathering dust.
Begbie doesn't believe growers need deep pockets to be keen on adopting AI.
There were plenty at Fieldays "really driven" by the desire to operate more efficiently and sustainably, he says. "We are building relationships and looking to work directly with them."
But he agrees more data gathering wouldn't be a bad thing.
AI, or at least robotics, have had a lot of airtime as the primary sector grapples with a critical labour shortage. The issue is not new but has been emphasised by the pandemic, with thousands of seasonal workers and backpacker labour shut out of New Zealand.
As Begbie says, robotics and AI are not the same thing but robotics has an assistant role in AI.
There's been a flurry of work in the past two years to find robotic harvesting solutions that don't damage fruit, which has to be picked by the piece.
A broad range of demonstrations has shown the dexterity challenge has been solved without damaging fruit from berries to apples to kiwifruit, he says.
But pace is a problem.
"If you look at a lot of these demonstrations of picking, they're extremely slow. In a kiwifruit orchard with an efficient (human) picker, it's fruit per second, not seconds per fruit. I think we are still five, maybe 10, years away from that picking rate."
Meanwhile the structures and operations of orchards are changing to improve productivity and the labour shortage problem while improving health and safety. And the AI work goes on.
In Hawke's Bay a prototype is hoovering apples off the tree, and a human on a platform directs them to a conveyor belt which takes the fruit.
Work is under way to enable orchardists to be less selective about skills.
Begbie says Auckland, Waikato and Canterbury universities are working with partners trialling augmented reality to teach pruning skills. Augmented reality is an interactive experience of a real-world environment where real world objects are enhanced by computer-generated perceptual information.
Using AI, a computer learns from an expert pruner to understand where to make a cut.
"Then you can bring in lower skilled pruners because the computer will help them make that decision. These are examples of robotic or AI technology not solving the whole problem but making it easier for humans to solve."
Commercialisation is probably at least two years away, Begbie says.
Meanwhile, PlantTech has been busy this year using AI to analyse nutrients and water stress in foliage using an aircraft with hyperspectral imaging which analyses a wide spectrum of light instead of just assigning primary colours, red, green, blue.
The aim is to provide feedback on where a grower doesn't need to use nitrogen and how to optimise water use.
"We can also look at biodiversity which is one of the things that excites me. If we can do this across sectors, if we can get data moving and sharing between sectors we can start to understand the environment at the catchment level."
New generation satellite data offers the early promise of opportunities to manage the whole environment for sustainable yield, Begbie says.
Smaller satellite orbits much closer to earth would enable daily data gathering instead of weekly, and while an aircraft could cover 2000 ha a day, a satellite could do the whole country.
All that data will lead to much more informed decisions about how the whole ecosystem is working.
So, plenty of promising projects, but when will working people get to use the results?
Begbie says kiwifruit sizing should be delivering value in Zespri's core systems next harvest and PlantTech is talking to a player in post-harvest kiwifruit and some other crops about deploying the technology.
"I don't think it's unrealistic to expect that being used at a small scale commercially next year."
The US is a potential export market.
On the aerial assessment research, PlantTech's talking to several parties how to get the work into a commercial tool. But Begbie doesn't see why it shouldn't be at early scale commercialisation within 18 months.
As for the biodiversity and sustainability catchment assessment work, it's likely to get the attention of authorities for its biosecurity and biodiversity management potential.
"We are talking to government agencies about how we can support these sorts of issues.
That very rapidly is likely to roll down to the likes of regional councils who will be charged with gathering this data."
But it will be commercial companies which take PlantTech's work to market.
"We create core AI engines but our model as an independent research organisation is to partner with commercial companies so we can embed new capabilities into their products.
"(This way) we can scale much more rapidly. We don't have the slow start-up phase, we can drive revenue back into technology in the form of licence fees and whatever.
"Also by building trust and confidence we can encourage those companies to reinvest in medium term research."
