The following is a long-form version of an educational article we published in Greenhouse Grower Magazine.
Here is a link to the original article and below you will find fleshed-out explanations of data and the training of artificial intelligence technology.
All the talk of artificial intelligence (AI) propelling us towards a digital revolution in agriculture is hard to avoid--and with good reason: this discussion is more than just hype. With the fall in computing costs coupled with greater access to agriculture data, a growing slew of high-tech AI tools continue to enter the market. Greenhouse growing is but one sector where growers today are beginning to see real-world impacts from this technology. The time is ripe for us to become more familiar with this novel tool for growers. Let’s begin here with an overview of how the technology works and follow with real-world applications of AI we can use to make greenhouse growing more efficient and profitable.
Smart software that learns from repetition
Putting to rest the robot misconception is a good place to start: though some equipment and robots have AI features that enable basic decision-making, AI technology is distinct from robotics. Most AI services today involve sophisticated software that works without any machines other than a computer processor.
Sophisticated AI software mimics the most basic way us humans learn: by repeating a task with the aim of getting better and better at a defined goal. Becoming an ace at poker provides an apt example. We begin by learning rules that set limits on what we can or cannot do. The rules make us focus on what is most relevant to achieving the goal of winning by playing the strongest hand. At first, you play a series of poor hands but begin to understand why you lost. After playing more rounds, your follies become less common and you start to learn strategies that improve your chances of winning. With practice, you learn to recognize with a high degree of confidence when your opponents are bluffing and when you have a winning hand.
With AI software we start by setting a basic framework, often termed an ‘algorithm’, that sets the ‘rules of the game’ and a defined goal. Rules can be, for instance, defining the most relevant growing conditions in your greenhouse, and a goal can be predicting the future flowering time for a variety of ornamentals. With the framework set, we now need to train the software to understand how to follow the rules and then make assessments that will likely achieve the goal with a high degree of confidence. Training involves having the algorithm ‘play many hands of poker’ by analyzing data you have on hand. Data can be any information available in a digital format, ranging from spreadsheets of your past heating bills, to records of how fast ventilation fans spin at given times of the day, to measurements of dissolved solids in your irrigation system. Soon enough the algorithm will understand how diverse growing conditions determine the future flowing time for an ornamental plant. Now trained, the smart software can analyze new data from your greenhouse and predict with precision future outcomes for your greenhouse.
AI is remarkable in its abilities to analyze large amounts of information within seconds, which enables the tool to automate common, complex tasks and predict outcomes with far greater precision than us humans are capable of. That said, current capacities in AI are limited. Just like being an expert in poker does not make you skilled at winning chess, an algorithm that predicts flowering time has no understanding of when best to harvest tomatoes. If you only play a few hands of poker, you’ll never get good at playing the game; if you have little greenhouse data of quality, the algorithm will also fail to learn. Thus, if you started cultivating a new variety of crop, you must complete a few harvest cycles in order to have sufficient data needed to develop AI tools for that new addition to your greenhouse. Moreover, applications of AI in greenhouses must be specific and are useless towards completing any kind of creative work that requires the occasional bending of the rules. Now let’s take a closer look at examples of these specific applications of AI.
AI as your digital assistant
Greenhouse tasks that require detailed comparisons or trial-and-error assessments are prime targets for an upgrade with an AI digital-assistant. Diagnosing diseases and pests in your greenhouse is one example. If you don’t recognize a crop illness offhand, you’ll need to do your research in order to identify the problem and know which pest management strategy will stop it in its tracks.
No longer. We now have thousands of digital photos of all major crops plagued with common ailments. Using these images to train an algorithm, smart software can now recognize many crop diseases and recommend treatment strategies. You now can download a free app for your smartphone and then snap a picture of an ailing greenhouse plant to obtain a diagnosis and recommended treatment within seconds.
Knowing the growing conditions that cause skin cracking and blemishes in vegetables is another problem of particular interest. Six to twelve per cent of a tomato harvest is unfit for the fresh retail market because of cracks in their skin. Growers often adjust humidity, temperature and irrigation factors through a process of trial and error before they find the right mix to control the problem. Forget trial and error. AI can learn patterns in growing conditions that cause skin cracking and recommend what to adjust first in order to maximize your yield of picture-perfect produce.
AI makes growing precise
Imagine calculating all the financial information for your greenhouse by hand. Not only would this manual task be a frustrating waste of time, you might also feel unsure about the accuracy of your results since there’s a good chance you forgot to carry a one or mistook a scribbled digit of “5” as a “6”. This is why you use spreadsheets and a calculator to do complex math. AI is like a calculator because it too can do complex calculations and avoid errors that pop up from manual work.
Forecasting harvest yields is a case in point. Growers need to know if they will harvest enough produce to meet contractual agreements with their buyers. Predicting the weekly harvest yields of, say, tomatoes and peppers is a manual process that requires counting underripe fruits and making informed guesses as to how current growing conditions will affect the speed at which they ripen. Not only is this a time-consuming process, it’s also often imprecise; errors in forecasts can be off by 30 per cent from the actual yield that week. AI can automate harvest forecasts with greater precision because an algorithm can conduct a continuous analysis of your growing conditions. The algorithm can learn how growing conditions will affect the speed by which vegetables ripen, enabling growers to know weeks in advance their future harvest yields within a single digit percentage point.
AI provides a means to store skills and knowledge
Growers are feeling the pinch from growing labour shortages of both skilled and unskilled workers. Concerns in the industry are mounting that should a member of your team--such as your head grower--retire or get poached by a competitor, it may take quite some time to find a replacement. And when you do, it can take years of training for this new team member to get enough of a feel of your greenhouse operations before they can be a fully autonomous employee.
AI can help ease the pain of losing a skilled employee. Once you train an algorithm to learn a task, that skill will remain embedded in smart software. Recall the previous example of harvest forecasting. Armed with the AI-automated forecaster, your greenhouse can continue this essential task even if you lose your head grower. After you find a replacement, they no longer need to learn how to conduct harvest forecasts, freeing their time to focus on more creative tasks that extend beyond the capacities of AI. Overall, your new employee can learn the ins-and-outs of your greenhouse operations in less time.
Fear not and ask questions
The potential for AI to automate and streamline operations in industries ranging from finance to healthcare demonstrate its broad impact on the future of business. The greenhouse industry is no exception. The time is now for growers to understand how AI will disrupt common practices in growing. Given the novelty of the technology, we are in the early stages of outlining how best to use AI and defining what rules should govern the provision of data from your business. While its novelty does raise questions, it is counterproductive for growers to shy away from this impressive technology. Should you have any questions about AI, let us members of the agriculture technology sector know about your concerns. Rest assured, we are ready to engage in further discussions about how this high-tech tool can assure bountiful harvests in your greenhouse.
Would you like to learn more about artificial intelligence and how this technology can strengthen your greenhouse business? We are ready to answer any inquiries; please forward your request to our Sales and Customer Service teams by filling out this short form and we will contact you soon after.
By using data from your greenhouse, our team of data scientists will develop custom-made automation services made possible by the latest artificial intelligence technology. We respect that your greenhouse data contains a wealth of information about your business that you never want to be shared with anyone outside of Motorleaf. Fret not: we’re leading experts when it comes to data management and you can count on us to protect your information at each step when developing your high-tech automation tools.
Your greenhouse produces a wealth of data and growers need to understand the value and power of all this digital information. Technology companies can use your data to develop cutting-edge automation services that make your business more efficient and competitive. Sharing your greenhouse data, however, raises issues about cybersecurity, privacy and data ownership; moreover, categories of data sharing are distinct and raise particular challenges.