As I celebrate my ten-year homebrewing milestone this upcoming Summer, I look back on my journey with fondness. From the early days of extract brewing my first IPA with humble pots and pans, to my current all-grain setup with a custom-built Mash/Lauter Tun, and the all-electric and half-automated Brew-In-A-Bag (BIAB) system that has transformed much of my garage — I’ve come a long way. But it’s not just about automation; it’s about the joy of enhancing my hobby with productivity tools and geeky gadgets that make each brew day more satisfying.
My trusty Alexa, who I adopted the moment it hit the market, is still one of my faithful companions — even if it’s just playing music. I even penned an article about it in 2016. And then there’s the Tilt device, one of the more recent additions to my homebrewing arsenal. This wireless hydrometer was a game-changer. A small, waterproof gadget that nestles in my fermenter, monitoring everything from specific gravity to temperature and fermentation progress, all in real-time. I track this vital data on one of my ancient Kindles, an essential companion that allows me to perfect my brews. As I reflect on my journey, I know that these tools have brought me closer to my craft, and that the best is yet to come.
Enter Generative Artificial Intelligence (AI).
Generative Artificial Intelligence (AI)
Generative AI is an exciting branch of artificial intelligence that creates original and unique content like images, videos, and text. The image you see above is a perfect example of this — it was generated using prompts I provided to DALL·E, a generative AI model developed by OpenAI. Unlike traditional AI that is programmed for specific tasks, generative AI uses machine learning algorithms to learn patterns and create content autonomously.
Prompt: robot making beer in a garage lab, graffiti style.
The applications of generative AI are far-reaching and have immense potential. In fields like art, design, and marketing, generative AI can produce one-of-a-kind and captivating content that catches the eye. Beyond DALL·E, other generative AI apps include: Artbreeder, RunwayML, GPT-3, Artisto, DeepArt.io, and Amper Music to name only a few. Additionally, in scientific research, generative AI can simulate complex systems and create new hypotheses. But it’s not just limited to these fields. This technology can be applied in homebrewing, opening up new possibilities to improve the brewing experience.
10 ways to use Generative AI in Homebrewing
With its endless possibilities, generative AI can have an immediate impact on the homebrewing community. In this article, we’ll take a look at 10 ways it can assist with the brewing process before diving deep into recipe development.
Recipe Generation: Generative AI can be used for recipe development of homebrewed beers by analyzing a vast dataset of existing beer recipes (public or private) and generating new and unique recipes based on specific parameters or criteria. The AI model could use machine learning algorithms to learn patterns and relationships between different ingredients and brewing methods and generate new recipe ideas based on that learning. One of the most significant advantages of using generative AI for recipe development is the ability to explore new and innovative combinations of ingredients and brewing techniques that may not have been considered before. By providing specific prompts or constraints, such as beer style, desired ABV, IBU, or available ingredients, the AI model can generate unique and customized beer recipes. We’ll explore this deeper later in the article.
Ingredient Selection: Generative AI can be used to assist with ingredient selection for homebrewing by generating recommendations based on the desired beer style, flavor profile, and other criteria. For example, a user could input the desired style of beer they wish to brew, and the AI could suggest a list of grains, hops, and yeast strains that would be appropriate for that style. The AI could also take into account the user’s preferred flavor profile, such as hop bitterness or malt sweetness, and suggest ingredients that would achieve that profile. Additionally, the AI could analyze previous brew data to suggest ingredient tweaks for future batches to improve the beer’s taste or aroma.
Label Design: Generative AI can be used to design unique and creative labels for homebrewed beer. By analyzing existing label designs and using machine learning algorithms, the AI can generate new label designs that match the style and personality of the beer.
Beer Name Generator: Generative AI can be used to generate unique and fun beer names for homebrewing by analyzing a vast dataset of existing beer names and generating new ones based on specific criteria. By providing specific prompts or constraints, such as beer style, flavor profile, or brewing technique, the AI model can generate customized and unique beer names that catch the eye and stand out from the crowd.
Quality Control: Generative AI can be used to analyze the quality of beer by analyzing data such as alcohol content, pH level, and bitterness. The AI can use this data to identify potential issues in the brewing process and suggest adjustments to ensure the beer meets the desired specifications.
Brewing Calculations: Generative AI could be used in hydrometer calculations by analyzing data from previous hydrometer readings and generating predictions for future readings based on that data. This process is called predictive modeling. The AI model would be trained on a dataset of historical hydrometer readings, which could include variables such as time of day, temperature, and specific gravity. The model would then use machine learning algorithms to identify patterns in the data and generate predictions for future hydrometer readings based on the input variables. For an example around water analysis, see Mastering Homebrewing: A Guide to Analyzing Water Profiles with ChatGPT and PDF Integration.
Food Pairing: Generative AI could be used to help with food pairing with beer by analyzing data on the flavor profiles of different beer styles and the taste of different foods. The AI model could learn the relationship between different beer styles and the flavors that complement them, and generate suggestions for food pairings based on that learning. For example, a homebrewer could provide the generative AI model with prompts such as the type of beer they are brewing and the food they plan to serve with it. The model could then generate suggestions for complementary foods based on the flavor profile of the beer.
Hop Research: There are hundreds of hop varieties available to homebrewers, each with their unique characteristics, such as bitterness, flavor, aroma, and alpha acid content. The number of hop varieties available for homebrewers can be overwhelming, but it also means that there are endless possibilities for creating different flavors and aroma profiles in beer. Generative AI can help with hop selection for homebrewing by analyzing data on the flavor and aroma profiles of different hop varieties and suggesting combinations that would result in the desired taste. The AI can use this data to generate new and unique hop blends based on specific parameters or criteria provided by the homebrewer, such as beer style, desired flavor profile, and available hop varieties. Additionally, generative AI can analyze historical brewing data to identify the hop varieties used in successful brews and suggest similar varieties for future brews. This technology can also assist in predicting the impact of different hop combinations on the final beer flavor profile, allowing homebrewers to experiment with new and innovative hop blends with confidence. I've explored this in Flavors of the Sunflower State: AI-Driven Brewing Adventures with Kansas Hops and Wheat.
Certification Prep: There are several certifications available for homebrewers, including the Beer Judge Certification Program (BJCP) and the Cicerone Certification Program. The BJCP offers a range of certifications, from Recognized to Grand Master, based on knowledge and experience in beer evaluation and styles. The Cicerone Certification Program offers several levels of certification, including Certified Beer Server, Certified Cicerone, and Master Cicerone, which focus on beer service, knowledge, and pairing. Generative AI could be used to simulate exam questions for homebrewing certifications by analyzing a vast dataset of existing exam questions and generating new questions based on specific parameters or criteria. By providing specific prompts or constraints, such as the level of certification, desired focus area, or type of question, the AI model can generate a unique and customized set of exam questions for practice. This would allow homebrewers to prepare more effectively and have a wider variety of questions to practice with, potentially leading to higher success rates on certification exams. What it can’t do, is take the tasting exams for you :) I've explored this one further in this in Become a Certified BJCP Beer Judge with the Help of Generative AI.
BeerSmith Analysis and Automation: BeerSmith is a popular brewing software used by homebrewers and professionals alike. Generative AI could analyze a vast dataset of Beer.xml files, such as the ten years’ worth of brewing data I have, to derive insights on patterns and correlations among various brewing techniques and ingredients. Based on this knowledge, the AI model can then create new and distinctive recipes. With specific prompts or restrictions, like beer style, ABV, IBU, or available ingredients, the generative AI can generate customized and one-of-a-kind beer recipes.
Although there are numerous other scenarios that can be explored, I believe that the above-mentioned ten examples are both practical, educational, entertaining, and can be implemented in your home brewery immediately. However, let’s dive deeper into the first scenario, as it is a crucial aspect of the entire homebrewing process.
Recipe Development with ChatGPT
Previously, I have talked about generative AI in a broad sense, but now let’s turn our attention to ChatGPT, a large language model created by OpenAI that offers a simple interface. We’ll be using ChatGPT and prompt engineering techniques to fine-tune the AI to generate a recipe for a Kölsch style beer, one of my preferred beer styles. We’ll also look at one of my existing recipes and feed it into the “frozen” pretrained language model for feedback. We’ll then continue to build out a food pairing for the Kölsch and wrap it up by giving it a name, fun beer description, and a marketing pitch. While all of this is geared towards homebrewing the same concepts can apply to commercial brewers.
Prompt Engineering
Before we get going, it’s important to introduce the concept of Prompt engineering. Prompt engineering is a technique used in natural language processing (NLP) and generative AI that involves creating specific prompts or input data to achieve desired outputs. This requires selecting the appropriate words and phrasing to communicate the intended meaning to the AI model, which then generates a response based on the provided prompts. It’s akin to searching the early days of the internet, where choosing the right keywords was crucial to getting relevant results. Prompt engineering is an emerging skill that requires a thorough understanding of the data and desired outcomes. In homebrewing, this means having a deep knowledge of the craft to achieve the desired result.
Prompt engineering with generative AI can be used to produce a good beer recipe by providing specific parameters or constraints to the AI model that generate a recipe that matches those criteria. For example, a homebrewer could provide the generative AI model with prompts such as the desired beer style, ABV, IBU, and types of ingredients they have available. The model could then generate a recipe that fits within those constraints. You could also feed the model and existing recipe you have and pass it new parameters to enhance or change the overall flavor profile.
Prompt engineering can also be used to experiment with unconventional ingredients or flavor combinations. You could provide prompts like “brew a beer that tastes like mango and habanero,” and the generative AI model could create a recipe that incorporates those flavors.
💡For a closer look at prompting in homebrewing see Prompting 101 for Homebrewers.
Designing a Kölsch
Let’s walk through the process of creating a beer from scratch. The BJCP style guide describes a Kölsch as a crisp, clear, light-bodied beer that originated in Cologne, Germany. It is a pale, straw-colored ale with moderate carbonation and a slightly dry finish. It has a delicate malt flavor with a subtle sweetness, and a low hop bitterness that is balanced by a floral or fruity hop aroma. Brewing a Kölsch beer can be challenging, especially for beginners, as it requires careful attention to detail and adherence to specific brewing techniques to achieve the style’s signature crisp, clean, and refreshing flavor profile.
If you’re an experienced brewer, chances are you have a go-to recipe for this style of beer. However, if you’re looking to experiment with a new recipe or trying your hand at brewing it for the first time, a simple prompt can be a great starting point:
Prompt: Generate a recipe for a Kölsch
And ChatGPT will respond with something like this:
Now, on the surface this looks like a decent recipe but there are a few things wrong with this response in my opinion and this is where prompt engineering, critical thinking, and domain expertise come into play:
The AI assumed I had a 5-gallon setup, but maybe I don’t.
The grain bill could be improved by adding Munich or Vienna and specifying that you want traditional German malts.
It’s not clear what the ABV or IBUs will end up being, it looks to be in the ~4.5% ABV range with maybe 25 IBUs, but I’d rather the AI provide more details.
Who uses Magnum hops in a Kölsch? Magnum hops will produce a slightly different flavor profile than using traditional Kölsch hops. I’d stick with a more traditional flavor profile, such as Hallertau or Saaz.
The Kölsch yeast was not specific enough and I’d have to go do further research on the various styles that I could use. I prefer Wyeast 2565.
Personally, I’d pitch the yeast at cooler temperatures in the 55-degree range and then lager for 4–6 weeks. The instructions are minimal at best here.
I’d use fining agents in a Kölsch for clarity and also add yeast nutrient.
This assumes you’re an all-grain brewer, but what if you wanted an extract version instead?
So, let’s build on that and try to provide more constraints through prompt engineering. And this is where you can also incorporate role prompting. Role prompting is a technique used in prompt engineering to guide the language model to generate text in a specific tone, style, or context. In role prompting, the language model is provided with a prompt that includes information about the role or perspective the text should be written from.
Prompt: Act as an expert all-grain homebrewer. Generate a new version of this recipe that has an ABV of 4.5% and IBUs of 20. Incorporate Munich into the grain bill, do not use Magnum Hops, be specific on yeast variants I can use, and use whirlfloc as a fining agent. Stay with a 5-gallon batch size. Ensure the instructions have the fermentation temperature at 55 degrees.
ChatGPT’s response was much more comprehensive, even including a suggestion to add yeast nutrient to the boil. The response also provided detailed information on the Original Gravity (OG), Color Profile (SRM), and International Bitterness Units (IBU), but unfortunately didn’t include the ABV. To clarify, I would simply follow up and ask for the estimated ABV of the beer based on the given recipe. You could even ask it to refactor that recipe and stay true to the BJCP style guidelines for Style 5B.
Overall, as an experienced homebrewer, I’m pretty happy with the results after tinkering with it. But let’s make one final adjustment to the recipe.
Prompt: Create that recipe again but make it so that the aroma is spicier.
Here you can see that Hallertau Mittelfrueh, known for its classic floral and spicy notes with a subtle hint of citrus has been added in place of Hallertau Blanc which is known for its fruity and earthy flavors with a touch of white wine grapes, passionfruit, and grapefruit.
Now that I have my recipe, I can load it into my brewing software and fine-tune it for my setup. But why bother with manual data entry in BeerSmith when we can simplify things even further?
Prompt: Act as an expert homebrewer that knows the Beer.XML schema. Convert this recipe to Beer.xml so I can import in into BeerSmith.
And there you go, you can take the XML as a starting point if you so desire.
As an AI language model, ChatGPT is trained on a large corpus of text data and uses statistical patterns to understand and generate human-like text. It does not have any specific knowledge of the Beer XML schema or any other technical schema. However, it can still provide output based on the text data it has been trained on and the input it receives from users.
This is one approach of many for how you might design a Kölsch from scratch, highlighting key components such as the beer’s color, carbonation, malt flavor, hop bitterness, and aroma through prompt engineering and role prompting to improve recipe accuracy and ensure adherence to BJCP style guidelines.
Improving an Existing Kölsch Recipe
Now let’s look at modifying an existing recipe. If you want to get feedback on an existing beer recipe, generative AI can be a helpful tool. All you need to do is give the AI your recipe and tell it what kind of feedback you’re looking for. With that information, the AI can generate responses that offer suggestions for improvement, point out any potential issues, and even provide insights on how your beer might taste, smell, and look. Let’s put this into action and see what happens when I test with a simple prompt on one of my Kölsch recipes. This specific recipe I developed has won some local and regional awards and even went to Nationals in 2016. So, I’m curious what ChatGPT might say.
In this example, I’ve exported plain text output from BeerSmith and provided this prompt to ChatGPT:
Prompt: I’d like to increase the earthiness of this recipe. Any thoughts?
In this example, ChatGPT provided some vague feedback around how I could make my Kölsch earthier by suggesting modifications to both the grain bill and hop profiles. But to be fair, I had a wide-open question for the AI. You could take this a step further and have it re-write that recipe based on that feedback and keep tweaking it if you want.
I’m pretty happy with my base recipe I have right now along with how it performs in my brewery, but I do think I might try incorporating Perle hops in an upcoming brewing session. Conducting such experiments sparks new ideas and motivates me to constantly innovate and challenge recipe development norms.
Food Pairing
Now that we have our beer, let’s match it with a food pairing for our AI generated Kölsch! Just like with wine, certain types of beer can complement or contrast the flavors of different foods. There is a really good book on this titled Beer Pairing: The Essential Guide from the Pairing Pros. For example, a crisp pilsner can pair well with spicy foods, while a rich stout can be great with chocolate desserts. Similarly, the bitterness of an IPA can be a nice counterbalance to the richness of a fatty meat, while a sour beer can cut through the richness of a creamy cheese. Beer pairing can be just as nuanced and enjoyable as wine pairing!
Prompt: Can you suggest some food pairings that would complement this recipe?
Prompt: Create a light pasta dish recipe that would pair well with this beer.
And there you have it. We’ve created a nice simple pasta salad recipe that should pair well with your new AI generated Kölsch.
Naming the Beer
Let’s give that beer a name and fun beer description! When Craft Beer & Brewing Magazine incorporated a beer name generator in their website, I had a blast playing around with it and generating names for my various homebrew recipes. A couple of my favorites were Luscious Hallucination and Double Ripe Rain Trip. I gave both of these monikers to a couple of my Hazy IPAs. The limitation with their tool when they rolled it out was that there was no input capability. I couldn’t do any sort of prompt engineering. But they have now incorporated ChatGPT in the backend with some variables you can provide.
But let’s continue to use the ChatGPT client to generate our name and description.
Prompt: Generate 10 beer names for a Kölsch that that has a fruity, floral and slightly spicy aroma. Base the names off of various unique planets inspired by science fiction. Ensure the names include at least three words.
Oh, I like #3 so I think I’ll go with that one and write a quick description about the beer (at least as a starting point).
Prompt: Generate a fun science fiction beer description and marketing pitch for “Zalanthan Fruity Nectar Kölsch”
I’d definitely spend more time refining this content, but you get the idea.
Prototyping Artwork
Finally, you could spend some time using generative AI to start prototyping your artwork for your homebrewing labels for the Kölsch. While generative AI can be a valuable tool in the prototyping process, it is important to remember that it should not replace the expertise of a human designer. A human designer can evaluate the generative AI output, refine and customize it as needed, and ensure that the final product is of high quality and effectively communicates the desired message to the audience.
Presenting the Zalanthan Fruity Nectar Kölsch, complete with a unique beer label generated by AI technology.
In Summary
Through the use of generative AI, we’ve dove into the process of creating a unique and delicious Kölsch beer. With the help of prompt engineering techniques and ChatGPT, we’ve generated a recipe that perfectly fits certain desired parameters, fine-tuning it to include specific ingredients and flavor profiles. We’ve also modified an existing recipe, leveraging generative AI to receive suggestions and feedback for improvement.
We’ve also seen how you can use generative AI to pair your beer with complementary foods, hopefully elevating your overall drinking experience. And to add some fun and personality to your brew, we’ve even turned to generative AI to generate unique and creative names for your beer, inspired by science fiction.
Final Thoughts
While generative AI has numerous exciting applications as we’ve seen here, it also carries several potential drawbacks. These include biases in the output, limited human control over the generated content, potential issues with quality control, and legal and ethical concerns. This has the potential to benefit nearly every aspect of our lives — so it must be developed and deployed responsibly. Companies like OpenAI are taking a proactive approach to addressing the potential downsides of generative AI and working to ensure that this technology is developed and used in a responsible and ethical manner, but it is early days.
The potential benefits of generative AI in homebrewing (and commercial brewing) are too great to ignore. With the critical thinking, the right expertise and the right datasets, homebrewers can use this technology to push the boundaries of their craft and create truly unique and personalized beer experiences. So, if you’re a homebrewer looking to take your craft to the next level, it may be time to start exploring the exciting world of generative AI.
To learn more about Prompt Engineering, I recommend checking out the course on Learn Prompting, a website created by OpenAI that offers guidance and resources for prompt engineering in natural language processing and generative AI. Additionally, The Art of Prompt Engineering is a book I just picked up to go deeper in this emerging skill and it looks like a decent read.
🍻Cheers to the endless possibilities that generative AI offers to the homebrewing community!
תגובות