In Conversation with Garden AI

Sure I like plants, gardening, landscape design and just being outdoors but I also like technology and some of the weird and wondrous ways these two topics can intersect. Case in point - the new ChatGPT playground from OpenAI is an advanced chat bot that excels at natural language, learning and occasionally drawing conclusions - even when the topic happens to be... Garden Design. I don't claim to be an expert (or even a novice) when it comes to topics like Natural Language Processing, NLP,  I've just always had an interest since trying to write my own rudimentary language processor almost 20 years ago. I'm just happy at this point to be a user who gets to marvel at this tool we've been handed.

So where am I going with this? Believe me - I have an interesting concrete example of chat intersecting with gardening to share. First though I should say that ChatGPT has access to a lot of training data - meaning it's general understanding of the world is broad - kind of like someone whose read the entire internet. However it doesn't have access to a lot information locked away in mediums such as books as far as I understand. Falling asleep last night I thought about what got me into NLP all those years ago. My mother had died recently and going through a giant pile of boxes I found among old pictures and letters a set of journals she'd written as a young adult. Yes, I know - generally private stuff and I possibly shouldn't have read them - but occasionally her writing was hilarious and often thoughtful - as a result her memory felt less distant. At one point I thought how incredible it would be if I could somehow bring this writing to life! Fast forward to today and it turns out others have had similar thoughts about bringing back relatives, historic figures or even their own, younger selves using OpenAI. Maybe I will too some day but for the moment I went a different route.

Like a lot of you I have dozens of books about gardening, design, plants etc with too much info to ever remember ( for me at any rate). It might be interesting to help ChatGPT ingest the information from one of these books and then ask it for gardening advice. Sounds like science fiction I know but as it turns out not so difficult after you get used to crafting the perfect question. 

So here's what I did. First I chose a book that addressed various garden types for example: water gardens, dry gardens, walled gardens, kitchen gardens etc. Next I used an OCR (Optical Character Recognition) app on my phone to take pictures of the books text and turn it into actual, editable text.

A pciture of text in a book.
An example page I scanned from a garden design book


The fun part is using this text as training data for the chatbot. There are no rules when interacting with the chatbot, no correct way to interact so that it can understand you - although it can get tripped up by ambiguity just like us humans. Interestingly the chatbot will give feedback about not being human if you start making anthropomorphic assumptions such as asking what it likes or hopes or dreams etc

telling chatgpt what I'm about I'm going to say
Priming the Chatbot with an overview of what's to come.

After entering information about several garden types I'm curious what is the book's writing style? Perhaps the AI can help clarify.

chatbot provides feedback on writing style

Can we hear about the garden types that have been entered so far in a different voice than that of the book? The book did seem a tad formal so let's get creative with our next prompt.

In the voice of a virtual Monty Don

What's interesting about this response is it definitely sounds like a passable version of what Monty might say! Phrases like "I have had the pleasure of visiting..." and "One of the most striking examples..." are spot on target. Another point of note is the mention of "Highgrove" - this is not information I entered but must rather have come from the chatbots general understanding of the gardening world.

Ok, so far, so good but let's turn up the science fiction a notch and ask Monty for some advice on a specific situation.

asking a virtual Monty Don for advice
Too many Garden styles?

Hmm - interesting and not bad advice given the input but I think my prompt should have been crafted with more thought since it seems too many garden styles are stuffed into the response. I'll fix that later but first I'm wondering if Monty isn't a bit busy to be giving us advice. I'm certain he must be so let's ask someone else and perhaps expand to some different genre's (?) Hmm...

Asking a virtual Walt Whitman for garden advice
Poetic garden advice

Or maybe we should seek advice from a darker source...

garden advice from H.P. Lovecraft
"Futile attempts to impose order"

The possibilities are endless and believe me I stayed up late exploring how notable figures from the past might respond. Eventually though I came to the conclusion that the responses felt shallow or rather they seemed like a thin-ish veneer. My guess is that it was a combination of how I phrased the prompt and the chatbot's current context so my next approach was to have a conversation about a given designer so that both I and the chatbot could craft a better discussion.

For whatever reason the Landscape Architect Thomas Church came to mind - possibly because he was a local hero of my professors in school. My memory of Thomas Church and his famous book were very faint so I asked the AI for a refresher.

A summary of a book by Thomas Church

Yes, it's coming back. He practiced in California and even advocated a "California Style". Playing with the chatbot I'd found that entering quotes or common phrases from a given person added the appearance of depth to it's responses. 

A somewhat amazing analysis of a quote by AI

I was surprised at the AI's understanding of metaphor and what the author might be suggesting. I tried another quote and played the part of an English teacher asking a tough question.

This response does not seem as if it has been regurgitated from a Wikipedia article - it seems like actual analysis. That's impressive but does it understand one of those most human of traits... humor?

can AI understand humor?
"Leave room for unexpected beauty"... words to live by.

The conversation about Thomas Church went on for some time. It was actually a fun exercise and great way to learn about his approach to design. At this point I felt there was enough information in the AI context and enough background information for myself that I could ask a good question.

Garden advice from a virtual Thomas Church
Thomas Church comes back to life and shares his advice


I think there is more to explore here but for now I'm heading outside where sun is shining.


What will the future be like? I'd answer "The same, but different". In other words much of what we will know of life in the future will be the same elements we appreciate today but aspects will be changed and probably cause unexpected secondary effects. For example think of smart phones... their still phones but more importantly powerful, elegantly designed computing devices which inadvertently super charged social media and changed society at a fundamental level. We still live in houses and apartments. walk our dogs, enjoy sunny days and dig in the garden -  the same could be said of people living 200 yrs ago. 200 yrs from now we may be doing the same but with layers of small changes that somehow add up to fundamental change.

AI response to my predictions for the future

Thank you AI for your time and thoughtful comments.

Until next time happy gardening and chatting!


  1. Well, you really put the Chatbot through its paces! I'm surprised you got so much time to interact with it. An LA Times columnist recently claimed that it put her off for some time before deigning to have a conversation with her. Many of the exchanges your relate made me think of undergrad-level papers - there was still a lot of regurgitating information even when there was breadth to the responses. I can't say that ChatGPT is ready for poetry yet either ;) I fear for the difficulties high school and college teachers will face with this technology on the loose but I hope educational institutions will work out controls with OpenAI sooner rather than later.

    1. These next years will be interesting for sure. btw - my sessions with chatgpt were spread out - and I often ran into times where the session just stopped working... maybe it timed out. It still felt a little like magic though :)

  2. Interesting. Back in the dark ages (1990's) when I got my CS degree, what there was of NLP was pretty primitive. One of my least favorite classes. Being outside with plants turned out to be more my style.

    1. Eventually we all end up outside with the plants!


Post a Comment

Popular Posts