AI is making cannabis cultivation smarter

(This story is part of the cover package in the May-June issue of MJBizMagazine.)

Machine learning is becoming increasingly common in indoor cannabis grows, as cultivators use sophisticated sensors and cameras to maintain optimal growing conditions, sound the alarm about threats such as pests or disease and reduce labor costs associated with both menial and high-level cultivation tasks.

“Cannabis has always been the enabler of some of these bleeding-edge technologies,” Nick Genty, CEO of North Carolina-based AgEye Technologies, said in an interview with MJBizMagazine.

“They’ve had the margins, and they’ve had the budgets to support investing in new technology versus some of the vegetable guys who don’t.”

There are two main reasons why cannabis and other indoor agriculture companies are implementing artificial intelligence or machine-learning technology in their facilities:

• The labor, energy and other resources required to grow cannabis indoors are expensive, and technology powered by artificial intelligence (AI) can reduce those costs.
• Consistency is key in any indoor agriculture facility, and AI can help deliver that consistency – even across facilities located in different states.

For now, AI is powering sophisticated data collection and monitoring solutions, and technology can help identify pests, diseases and leaf discoloration, Genty said.

AI for indoor agriculture is getting closer to being able to monitor and highlight what isn’t visible in a grow, such as growth-rate changes in plants.

In the future, AI will be able to enhance growth rates once consistency has been reached.

“You’ve established that you can do consistency across that baseline,” Genty said. “Now let’s improve the crops.”

More precision, less labor

For Stephen Hess, the head of cultivation at Arizona-based cannabis company 22Red, AI’s gift is eliminating much of the manual and logistical work involved in cultivating cannabis.

That means Hess can save on labor costs and free up time to focus his energies elsewhere.

“I don’t want to look at 47 different set points, quantify those numbers, create nine overlays and say, ‘Here’s the solution,’” Hess told MJBizMagazine.

At 22Red, Hess implemented the Spyder from San Francisco-based tech company Neatleaf.

The technology includes multiple sensors that traverse the canopy of indoor grows to track data such as leaf color, temperature and humidity multiple times per day for individual plants and microclimates within the facility.

The Spyder’s AI technology then assesses that data to detect patterns or signs of stress; its findings are presented on a dashboard with charts, heat maps and light maps for Hess and his team to pore over.

Before the Spyder, Hess deployed his team with handheld sensors to gather the same kind of data throughout 22Red’s cultivation rooms; workers spent five to 10 minutes per plant or microclimate, he said.

The labor cost savings using the Spyder are significant.

“I only need six growers on-site versus 10,” Hess said.

Rather than paying $50,000 each in annual salaries to four employees, he’s paying $600 for the Neatleaf subscription, plus another $600 per month for evaluations – another benefit of the service.

Plus, Hess said, the Spyder’s recommendations are more precise because the software is making calculations based on things the system has learned gauging crop deficiencies at other cultivation sites it has monitored, Hess said.

“It gives me more consistency, which then gives us more quality throughput with what we’re trying to do from an evaluation standpoint to utilize that data to make the best decision for the canopy,” Hess said.

Specialty data

Elmar Mair, a co-founder of Neatleaf, said the company has so far implemented the Spyder at 30 facilities.

The company has a backlog of clients and a two-month waiting list that includes non-cannabis cultivators, such as berry growers.

One of the most powerful aspects of the Spyder is its ability to look back in time and even compare the health of one plant to examples of the same strain grown in previous years, he said.

“Plus, you can do it all remotely,” he said. “You don’t have to be on-site.

“You can look at the garden without being in the space. (Multistate operators) can leverage expertise without having to fly around.”

Not all grows can accommodate the Spyder, however.

The Spyder can’t collect data from areas that are too low or blocked, such as in vertical cultivation facilities or grows where pillars would obstruct the sensors.

Consistency at lower costs

Rather than implementing one AI service, Florida-based marijuana MSO Jushi Holdings has worked with mechanical engineers and contractors to ensure that equipment with machine learning drives its building-management system.

That approach enables Jushi to efficiently collect and analyze data as well as reduce energy costs and produce cannabis more sustainably, said Ryan Cook, Jushi’s executive vice president of operations.

Each mechanical unit, such as an air conditioner, connects to the building-management system, which gives the cultivation team a bird’s-eye view of the facility.

“It makes me sleep better at night knowing that Josh (Malman, Jushi’s director of cultivation) has access to all of our systems and that he can log in there and he can make changes,” Cook said.

But any manipulation made in the system should be done strategically, Cook cautioned.

Everything from humidity to temperature to air flow is impacted by everything else, so making manual tweaks can be risky.

“The reality is that our systems are so interconnected – and everything is learning on a regular basis. If you make those manual adjustments, it takes the system longer to relearn the knowledge that it originally had,” Cook said.

Malman identifies what set points are required for the grow, and when compressors click on, they adjust the climate to meet preset requirements.

Machine learning allows the system to understand whether it should work at 70% or 100%, and for how long.

“Just those minor modifications of the controls make a big difference in how the room can actually manage itself,” Cook said.

While the system must “learn” each facility separately, it learns more quickly by assessing data from multiple facilities, he said.

Jushi’s technology is not an autonomous system – although Malman said he’s curious about vision systems in LED lights and robotic trimmers.

With the addition of AI, training cultivation staff is as much about technical skills and data analysis as it is horticulture.

For now, building management is about fine-tuning every aspect of the growing process to ensure it’s optimized and working synergistically.

From irrigation to fertigation, Malman and Cook said analysis also allows them to predict what will be required next.

“We’re changing methodologies and our cultivation systems simply based on the fact that we’re capable of seeing the future now,” Cook said.

“You can predict where we’re headed simply based on being able to analyze that data faster.”

Kate Robertson can be reached at kate.robertson@mjbizdaily.com.