Interpretive Writing Cage Match: You vs Chat GPT 4o.

Hideous AI drawing of an orchid typing on a laptop.
Sorry, if I have to have nightmares about all of this then so do you.

Recently I wrote that I have been using AI for a lot of my work functions: summarizing data, cleaning up text, and other dreary tasks. And the more I work with it, the more I have come to depend on it.

But I feel that there are some of my core services that are soon to go the way of the dodo. Sadly, I think interpretive writing might be one of them. Let me show you what I mean.

In my interpretive signs course, I do an activity with my group. I want to challenge you with the same activity, and see how you do. Then I want to show you what my little AI chat buddy did in exactly four seconds. Are you ready?

Interpretive sign writing assignment

I would like you to write me an interpretive sign. It’s a wayside sign for family audiences about Calypso orchids. I want it to be readable, interesting, and clever. I would like it to have a title that catches the attention, no more than five words. I want it to have a subtitle that is a complete sentence and states the main message of the sign. I want it to have a body section of about 80-100 words. I want you to take this scientific abstract and turn it into that sign.

Calybso bulbosa var. occidentalis is a western North American terrestrial orchid. In the northern California populations studied, these plants are pollinated by Bombus queens and Psithyrus females. Bee visitation, however, is infrequent. Calypso bulbosa var. occidentalis is a generalized food-flower mimic and must rely on exploratory visits of naive bees for pollination. Abrupt rises in the number of flowers pollinated corresponds with sudden increases in available pollinators. Fruit is set low (11-34 percent), and is comparable to that mentioned in reports of other temperate food-deceptive orchids.

A large number of orchids provide no reward to their pollinators. Their visual, olfactory, and sometimes tactile cues are deceitful. Perhaps the most striking examples are the pseudocopulatory orchids,Ophrys, Trichoceros, and a variety of Australian species (Kullenberg and Bergstrom, 1976; Pijl and Dodson, 1966; Stoutamire, 1975). The majority of non-rewarding orchids, however, are pollinated by visitors apparently searching for food. Very few such systems have been studied in detail. Recently, much attention has been focused on Calypso bulbosa (L.) Oakes, a strikingly beautiful circumboreal orchid. Mosquin (1970) and Stoutamire (1971) found that C. bulbosa var. americana is nectarless and hypothesized that pollinators are attracted to the flower because of its conspicuous coloration. Furthermore, the flowers appear to rely on newly emerged naive bumblebees to effect pollination.

Source: https://www.jstor.org/stable/41424311

Are you up for the challenge? Start your timer and get to work.

Post your development time and your sign text in the comments below if you like.

All done? Excellent.

Here’s what Chat GPT did in exactly four seconds. Here’s my prompt – the same prompt as above.

Image of Chat GPT prompt. Same instructions as written above for the sign writing assignment.

Here’s the AI output, below.

Title:
The orchid with a clever trick

Subtitle:
Calypso orchids deceive bees with beauty, not nectar.

Body:
The stunning Calypso orchid may look like a tasty treat to bees, but here’s the secret—there’s no nectar! Instead, this clever flower tricks young bees into visiting with its bright colours and sweet scent, only to leave them empty-handed. But the bees still help! On their first visit, they unknowingly pick up pollen and carry it to the next orchid, ensuring pollination. Though bees don’t stop by often, when they do, Calypso orchids make the most of it, continuing the cycle of nature’s trickiest beauty.


Four seconds. It took four seconds.

Is it as good as something you could write? Because I have done this exact activity with about five different classes of professional interpreters, and it’s better than about 60% of their first drafts.

Am I bothered?

Yes, a little. The writing is on the AI wall, so to speak; I think there’s a risk that our managers and clients will be turning to AI for their writing needs. On the other hand, I do think I could write a better sign than the above—and I can also work with the client to help them think about whether this is the right message on the right sign in the right place for the right audience. I think that’s still worth a lot.


Hey does this kind of writing tickle your brain? Sign up for monthly goodness from donenright.com.

Wanna see what past newsletters look like? Clickity click.

Get monthly (ish) updates via email from Don Enright. I write about interpretation and visitor experience. I never sell or share my lists.

This field is required.

Comments

7 responses to “Interpretive Writing Cage Match: You vs Chat GPT 4o.”

  1. Don Enright – wondering if we are related – my grandmother was Martha Enright Ward – from Ellicottville , NY. We have quite an extensive Enright family tree and active researchers. We believe The Enrights were from County Kerry or Limerick. Dan Ward.

  2. I appreciate where you are coming from, and sure the output was pretty good, but for the AI to do its work, someone had to decide that Calypso orchids were something worth interpreting, that a wayside sign was the way to do it, that it was being done for the benefit of family audiences, and then direct the AI to a reliable source of information on which to draw.
    What about all the questions about layout and presentation on the panel and creative stylistic choices for the panel itself that weren’t addressed?
    It might also be expected that this sign would be part of a larger interpretive provision on a site, and presumably share a thematic style, which adds a level of complexity to the challenge.
    Also, five minutes on Google Scholar found a couple of snippets of information that might be worthy of inclusion that either an AI wouldn’t be directed to find, or if it was give freer rein, would then require a real human to fact-check before publication.
    I think this is as much evidence of how badly thought out, designed and written so many panels are, and how little many clients understand about what ‘good’ might look like, than how clever AI is.
    On the other hand, i could certainly imagine asking it to generate a bunch of different options and then playing around with them to come up with the final product.

    1. Don Enright Avatar
      Don Enright

      Hi Ian- yes, absolutely to all of your points. But that’s not what my article is about. I’m talking about the specific skill of interpretive writing, which is distinct from (or perhaps a sub-set of) interpretive planning and experience design. Interpretive planning and experience design are hard. They’re terrible tasks for AI. Interpretive writing—crafting text as inspired by an interpretive plan and based on existing scientific research—is what Chat GPT nailed here. And that type of interpretive writing—the pure creative wordsmithing—forms a substantial part of my contract work. That’s where I’m coming from with this.

  3. Raymond Nakamura Avatar
    Raymond Nakamura

    Pretty impressive. It kind of reminds me of how word processors made many people think they could be designers. Except, as some have pointed out, AI can start to decide to do stuff on its own. I just think it’s fun to put words together on my own.

  4. ron chamney Avatar
    ron chamney

    Hi Don, I am still looking for my jaw after it hit the floor. I have done (more than one) interp blurb for the Calypso. Rest assured they took me far longer than 4 seconds. Is the AI blurb finished? No, but it shows real promise and it was able to capture everything you outlined in your prompt. If only my interp staff was able to follow all my concept outlines so well. The thing that impressed me most is that it had some good hooks that made me want to know more and it actually seemed to have a style to it. With deeper parameters in the prompt I am certain that this AI would produce something solid that I could sculpt into what I could use. Now, Don, help me find my jaw! Well done… Ron

  5. Clark Hancock Avatar
    Clark Hancock

    I am reminded of the observation that in this evolving “Information Age”, what has increased in value is not answers but the questions asked. With AI, that manifests with the prompts used such as the extensive ones you created for this exercise. AI is a tool for the interpretive artist. As with all tools, it is up the artist to determine appropriateness to a specific situation. As with any skill, it requires a level proficiency developed through conscious and deliberate practice.

  6. Clark Hancock Avatar
    Clark Hancock

    Pardon the typos, should have used AI

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.