How Might AI Impact The Coffee Industry?

All images in this article were generated by Midjourney, an image-from-text generative model that generates images based on natural language descriptions. Prompts used to generate the images are included with each image.

It is largely assumed that machine learning and data mining are precursors to AI and that the current large language models (LLMs) such as ChatGPT and image-from-text generative models such as Midjourney will help usher in artificial general intelligence — machine sentience. Most of what is currently called AI is effectively machine learning through pattern recognition as a result of data mining.

The terms we use such as AI versus learning model versus algorithm are still being worked out. I wanted to be strict and not call anything AI yet because I don’t believe we yet have AI. But the term is being widely used and generally understood to represent a broad collection of algorithms and machine learning models. Therefore I use the term AI in that manner; a broad category of disparate machine learning models and algorithms.

Artificial Intelligence (AI) could impact the coffee industry in any number of ways. One of the ways I envision it impacting the industry is through novel ways of improving smallholders’ livelihoods, ways that have not historically entered mainstream thought. I think this will happen through novel pattern recognition and unique insights.

To explain, I need to take a bit of a side-journey into some recent developments in the progression of AI.

AI improves cooling of Google data centers

In 2016, researchers at DeepMind, an AI company owned by Google, developed a machine learning model to improve the cooling of data centers. Despite some of the world’s best engineers already working on the program, the model improved efficiency by 40%.1 Google called it the “Machine Learning Control of Datacenter Cooling” (MLC).

Our machine learning system was able to consistently achieve a 40 percent reduction in the amount of energy used for cooling, which equates to a 15 percent reduction in overall PUE [Power Usage Effectiveness] overhead after accounting for electrical losses and other non-cooling inefficiencies. It also produced the lowest PUE the site had ever seen.

Google’s MLC approach uses machine learning algorithms to predict the cooling needs of different parts of the data center in real time. The algorithms analyze data from sensors throughout the facility, such as temperature, humidity, and air pressure, as well as data on how the servers are being used.

Based on this analysis, the algorithms can adjust the cooling system in real time to provide the right amount of cooling to each part of the data center. This approach is more efficient than traditional cooling because it allows for dynamic adjustment of the cooling system based on actual needs, rather than relying on a fixed temperature setting.

The MLC found a new way of thinking about data center operations.

movie scene style like movie Fifth Element, artificial intelligence designs cooling system for massive data center --v 5 --ar 15:10 --q 0.5
movie scene style like movie Fifth Element, artificial intelligence designs cooling system for massive data center --v 5 --ar 15:10 --q 0.5

AI masters chess

In 2017, DeepMind announced that AlphaZero, a machine-learning algorithm designed to learn to play games, had mastered chess. It mastered chess simply through self-play. The programmers of AlphaZero did not program into it the principles of chess — no pre-programmed moves, combinations, or strategies. It learned them on its own. After only 24 hours of training, AlphaZero taught itself how to defeat world-champion chess programs such as Stockfish and Elmo.

Part of what was significant about AlphaZero’s accomplishment (aside from the fact that it learned the pricinciples on its own) was that it seemed to play with insight. According to a New York Times article; “It played like no computer ever has, intuitively and beautifully, with a romantic, attacking style.” The article goes on to say: “While conducting its attack in Game 10, AlphaZero retreated its queen back into the corner of the board on its own side, far from Stockfish’s king, not normally where an attacking queen should be placed.”2

Grandmasters had never seen anything like it. AlphaZero had the finesse of a virtuoso and the power of a machine. It was humankind’s first glimpse of an awesome new kind of intelligence.

As Henry A Kissinger, Eric Schmidt, and Daniel Huttenlocher said in their book titled The Age of AI: And Our Human Future: “The AI did not just process data more quickly than humanly possible; it also detected aspects of reality humans have not detected, or perhaps cannot detect.”3

AlphaZero learned a logic of its own.

faceless AI boss playing chess against Stockfish --v 5 --ar 15:10 --q 0.5
faceless AI boss playing chess against Stockfish --v 5 --ar 15:10 --q 0.5

AI identifies new pharmaceuticals

In 2020, researchers at the Massachusets Institute of Technology (MIT) built a machine-learning model that can idenfitfy new antibiotics. To develop the model, researchers fed it a 2,000-molecule database consisting of molecular structure as input and antibiotic effectiveness as output. The researchers then provided the trained model a larger database of about 6,000 molecules. The model proposed an antibiotic researchers now call halicin, named after Hal, the computer in the movie 2001: A Space Odyssey. In the movie, Hal decided to kill the crew of astronauts to protect the mission.

Halicin is effective in killing many strains of bacteria known to be resistant to conventional antibiotics. According to an MIT press release; “The drug worked against every species that they tested, with the exception of Pseudomonas aeruginosa, a difficult-to-treat lung pathogen.”4

Halicin was an existing molecule, already known, but never considered as an antibiotic.

In many cases in which the AI-chosen antibiotics were effective, no one knows why they are effecive. The model was able to identify them as effective, based on patterns of qualities and characteristics humans have not yet learned to associate with an antibiotic’s effectiveness.

The MIT deep-learning model discovered its own form of pattern recognition.

comic book panels showing artificial intelligence developing antibiotics for humans --v 5 --ar 15:10 --q 0.5
comic book panels showing artificial intelligence developing antibiotics for humans --v 5 --ar 15:10 --q 0.5

Applying this to the coffee industry

Novel ideas and new ways of thinking about problems are the golden eggs of AI so far. Having AI to perform tedious, mind-numbing work is good. Yes, it will make a lot of paid-jobs obsolete and that will impact a lot of marginalized and impoverished workers. We’ll need to make sure we do our best to bring them along rather than leave them behind.

AI, or the algorithms and learning models we’re currently calling AI are already disruptive to some creative portions of the coffee industry; namely marketing departments. You can, today, use a language learning model (such as ChatGPT) and a text-to-image model (such as Midjourney) to create entire marketing campaigns. You can use a language model to generate a Twitter thread. You can use Midjourney + ChatGPT to generate an Instagram post. You can design package labels, logos, icons, etc with the tools currently available.

Entire paragraphs in this article were written by AI. Can you guess which ones?

AI is already disrupting the industry. We just need to make sure we collectively benefit from the disruption.

The following image was generated by Midjourney, using a prompt generated by ChatGPT. Both outputs are impressive.

My prompt to ChatGPT was:

In a short paragraph or less provide a detailed description of an oil painting that depicts the entire coffee value chain, from seed to cup, in one single image. Use long-tail descriptions, lots of adjectives, choose a medium type, and name an artist who might paint such a picture. The description will be used to generate a compelling image using an AI text-to-image synthesis system.

ChatGPT’s response:

The oil painting that depicts the entire coffee supply chain is a vibrant and dynamic work of art. The painting showcases the journey of coffee beans from their origin to the moment they are served in a cup. In the center of the painting, a tall and majestic coffee tree is depicted, surrounded by colorful workers picking and sorting the ripe coffee cherries. Moving along, the beans are processed and dried, with workers carefully monitoring the quality of the beans. Next, the painting shows the beans being roasted in a traditional roaster, filling the air with a rich aroma. Finally, the painting shows a barista expertly brewing a cup of coffee and presenting it to a happy customer. The use of light and shadow in the painting highlights the beauty of the different stages of the coffee supply chain. An artist who might create such a painting is David Hockney, a British painter known for his vibrant use of color and dynamic compositions.

Midjourney’s response using ChatGPT’s prompt:

The Coffee Supply Chain as envissioned by Midjourney

Michael C. Wright

Michael is a licensed Q Grader, licensed Q Processor Pro, an Authorized SCA Trainer (AST), and most recently, a graduate with a degree in horticulture and a concentration in horticultural business management. He has over ten years experience in the coffee industry operating on both the supply and demand sides of the value chain.