I just got back from 2 weeks across 3 events, and I have a lot of thoughts about how AI* is showing up in conference presentations.
* [for these purposes, let’s define “AI” as the umbrella term for LLMs, generative tools, text-to-image generators, automatic prompts and suggestions from tools, and pretty much anything but “intelligence”. But that’s the language we’re using, so.]
We’re talking about AI. We’re all talking about it, even more than we were about blockchain. How to host it, how to monitor it, how to feed it, how to use it effectively as a manager, as a team, as an organization, as a marketer, as a devrel, as a technical writer.
Talks about AI and LLMs and ML are all over the conference circuit, and rightfully so. It’s a technology that’s going to change a lot of how we do work. At SREcon EMEA, there were talks about building the datacenters needed for model learning, and how to make sure your data is what you need, and using machine learning for troubleshooting. At LeadDev West Coast, there was a talk on scaling ChatGPT, both machines and people.
My third event was an AI Summit by Heavybit. Since it was an unconference, we didn’t have a set agenda, but there were topics like how to secure AI models, how to use AI to create specific models, how to opensource AI or run a small instance on your laptop, what opensource and AI could do together, and how AIs and APIs will interact.
If you have a talk about AI in action, not just in theory, then the world of conference organizers and attendees probably wants to hear it.
Creating slides is not easy. The way I do it, I put my outline into slides, put in some speaker notes, and then spend a few hours searching for images that are:
- right for my message
- licensed for re-use/modification
- appropriate to the venue and audience
DuckDuckGo is my new favorite tool for this, because unlike Google Image Search, it remembers that I’m searching for open-licensed images, even if the search term changes.
However, for the first time, I saw a bunch of conference slide images that were generated with DALL-E or other text-to-image prompt tools. It’s so satisfying, because I frequently have an idea of what I want, like “a bunch of analog clocks that are set to different times, photorealistic, black and white”. And I can’t always find that, and I end up hopping through a bunch of dead-ends and almost-rights. What if I could just… type it?
Here’s the thing: I don’t feel right doing it. I know too many artists who have had their art scooped up and fed to the great homogenizing sausage-maker that is AI. The images are indexed by humans training and fine-tuning models. You don’t think a computer knows what “high-fantasy art” looks like unless a human tells it, do you? It doesn’t. Computers don’t know anything, they just remember what we tell them. And that goes triple for generative AI.
So while I’m not going to judge anyone for using text-to-image to create their slides, certainly not the way I’m side-eying the person from a company who can absolutely afford to pay licensing fees using a stock image that still had the watermarks on it… while I’m not judging, I don’t feel good about doing it myself. You should think now about how you feel, because it will come up.
Prompts and guessing
My stance above about art is in contrast with this part. I think it’s fine and useful to use LLMs to summarize text, create outlines, and generally help you get started on writing. What is the difference between scraping the world’s art and the world’s text? I think the key lies with my understanding of fair use, and of the distinction between creative and business writing. You can’t just use, say, a photograph of Prince to do whatever you want, even if you are Andy Warhol. It has to be transformative, or comment on the original piece, or meet some other tests. (In the US. Copyright law varies.)
Visual art, especially art that is not work-for-hire, is a creative expression. Creative writing is clearly creative expression. But business writing? Eh.** The document as a whole is something you can copyright, and it’s shitty to rip off recognizable chunks, but I have written a lot of Quick Start documents, and from the time when they said “Insert the CD and follow the wizard” to the present day, there’s just not a lot of latitude in how you can or should describe starting quickly. It’s not plagiarism to have Step 1 followed by Step 2. It’s just… the shape of the document. And so I think it’s fine for the LLMs to offer us Notion documents and slide templates that have prompts and likely things that we’ll want to use to get started. I think there’s nothing wrong with tools that summarize or outline our free-form writing. And to do that effectively, they need to have trained on copyrighted material. But they are not trying to make us think that this is human creative expression. Business writing has never been that.
**[Ok, there was a time when I whistle-blew on a manager who wanted me to plagiarize the frickin’ AMA, but that was an entire document, in situ, not the same. But yes, copyright does technically apply to technical documents, and would you want to try to defend yourself in that situation?]
What’s a speaker to do?
No one is going to come drag you off the stage for bad image attribution or using a slide template or getting a boost from a writing assistant. The vast majority of people don’t care about the fair use implications of DALL-E or Midjourney even a little bit.
This is your talk, your representation in the world, to people you don’t know. How do you want to present yourself? Maybe it is with amazing, adorable, candy-colored owl slides. Maybe it isn’t. Maybe your talk, or your personality are all Noun Project, or all NASA images (always free), or all pictures of fish you have caught. Maybe they’re hand-drawn stick figures. Whatever it is your talk is, be sure you’re comfortable with how it represents you and your values.