Art & Craft

Robot waiters, chatbots, and AI image generation

I love technology. I work in tech and I am a big consumer of tech in its various forms. But I’m not blind about the shortcomings of a lot of technology, and part of what I find fun about using new technology is comparing how it actually performs with the marketing bluster about it.

Yesterday I went to a hotpot restaurant which had a robot waiter. Not so much the sleek, cool and futuristic robot that you might have imagined if you were asked a decade ago what the robots of today would look like. This robot had a cat face, played muzak-style elevator tunes, and needed rescuing when it got stuck on the edge of a wall. It was cute and novel, but only capable of bringing out dishes and returning to the kitchen. I’m sure if the human cooks and wait staff there were asked if it made any significant impact on their workload, the answer would be a no.

Here kitty kitty.

My background in customer service means that I got to witness all the hand-wringing that occurred when chatbots started to become adopted by businesses. Then I started creating my own chatbots and training people how to build them. I saw that even the very best chatbots could never be any true substitute for human help. Even today, decision-making, humor, and empathy are things that AI can emulate, but never originate.

Robot waiters and chatbots are barely any threat at all to the human capabilities that we rely on to get good customer service or be served well at a restaurant. But one area of technological advancement that I think is right now challenging human capabilities is the advent of AI-generated art.

If you’ve never played with text-to-image generation via AI, I invite you now to give it a go. Most online image generators allow you to create an account with them, claim a handful of free credits, and play. Nightcafe is probably the best web-based generator available at the moment, which uses the Stable Diffusion model to create art from the text prompts that you provide it.

“A person reading a blog”. This is what Stable Diffusion thinks you look like right now. Seems SD doesn’t really understand that blogs are not paper-based.

There are a number of different text-to-image generation models out now which work in different ways. If you’re interested in how Stable Diffusion works, check out this excellent primer from Jay Alammar. Nightcafe also lets you access Open AI’s Dall-E 2, which was for a long time, in my opinion, one of the best AI models out there.

Here’s Dall-E’s version of “a person reading a blog”. Looks like it’s drawing from a lot of stock image sites. Note that text-to-image generation still has no idea how to render text…

Just lately I have been bowled over by Midjourney, accessible via Discord.

Last August, I was part of Midjourney’s beta before it opened formally to the public. I found that creating the best images required you to craft longer and more intricate text prompts that called out specific artists, art styles, lighting and color that the model could draw from (around this time, there was even talk of AI prompt creation becoming a fully-fledged job.) I was fairly impressed. But only fairly.

The prompt I used here was “Cheeky cat”

“Black and white illustration creative design human brain biopunk body horror by Junji Ito”

Midjourney has just released it’s V4 model. And it produces gorgeous images, right off the bat, without needing to craft elaborate prompts.

But don’t just take my word for it. Check these out.

Vanilla Midjourney of “a person reading a blog”
Midjourney V4 “a person reading a blog”.

What’s interesting about Midjourney V4 is that it’s almost too good. It’s not perfect – it struggles to render text, and some hands it produces have a few too many fingers. But some of the images it produces are not only beautifully composed, but also indistinguishable from images created by people, by hand.

This is simultaneously impressive, and problematic. There’s a lot of debate happening right now about the ethics of AI image generation, given that artists can claim to own an image that they put no physical effort into producing. Deviantart saw a major backlash from its users when it implemented its own image generation tool, with many artists leaving the site over what they saw as its proliferation and promotion of low-effort “art”.

I certainly would never generate an image, put my name on it, call myself an artist and call it a day. Where Midjourney V4 has been genuinely changing the world of art for me is in its ability to help with my creative process.

I’ve always been creative, but never truly considered myself artistic. I have an eye for things that look nice, but I don’t have the artistic chops to draw things from scratch. This is problematic because while I enjoy designing things from a starting point that already exists, I never want to copy from existing artists or do anything that could be considered to be impinging on anyone else’s original design.

But AI-generated art has solved this problem for me. I can design from the starting point that AI provides, without worrying that I’m stealing someone’s work, or in danger of touching on anyone else’s copyright.

So I’ve been using AI to help me in the crafty things that I enjoy. Some of the crafts I like doing are:

  • I’ve been very into fiber arts for a long time and have a long history of sewing, knitting, crocheting and the like. For me, good craft is tactile as well as visually pleasing. I love to hold soft beautiful things in my hands and to give those soft and beautiful things to other people, knowing that they can be snuggled into and loved in a very childlike sort of way, like a favorite teddy bear.
  • I also really love dying things pretty colors.
  • I am also a big fan of block printing.

Here’s an example of an image that I created with AI…

And then moved it over to Photoshop and messed around with it there…

And then I carved it into a rubber block, hand-printed it onto fabric I dyed and made patches with it.

This patch project was the first time that I used AI in my process and I fell in love with it. I really love taking an AI-generated image which is almost perfect, and then making it tactile and human again through hand-carving and printing with a rubber block.

While robot waiters and chatbots won’t be impinging on our livelihoods in a significant way any time soon, if they were, it would certainly result in all the ethical dilemmas that we’re seeing in the world of AI image generation right now. I think where we can be free of ethical dilemma in the world of AI image generation is by taking what AI does as a starting point and then using that to create the type of art that only humans can create.

I’d love to see others using AI for the same purposes. Please let me know if you’re doing, or planning to do similar!

Leave a Reply

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