When AI Has Something Different To Say

A different medium might actually offer a different angle to the message

I’ve been talking to AI for almost 2 years now. I’ve been bickering with search since 1998, and fussing at computers since 1983. I’ve had a lot to say to technology over the years, but only since generative AI came along has any of it answered.

I’ve been talking to AI for almost 2 years now. I’ve been bickering with search since 1998, and fussing at computers since 1983. I’ve had a lot to say to technology over the years, but only since generative AI came along has any of it answered.

Since starting to work with generative AI almost 2 years ago, I’ve literally spent thousands of hours engaging with it, trying to learn its patterns, its ways, it’s vulnerabilities, and it’s hidden strengths. That’s a lot of talking. And given how generative AI tends to be and how much text it tends to spit out, that’s a lot of words.

And frankly, it kind of tests my patience. All the words. Too many words. Isn’t there any better way to interact with AI than having to look at all those words?

Meanwhile, everybody’s getting completely worked up – and rightfully so – about how AI is cranking out all this imagery at the expense of visual artists. People who create images with AI aren’t considered artists by many, and I’ve heard people say that AI and Art are mutually exclusive.

All the artistry debates aside, I got to thinking… I’ve experimented with doing different kinds of analysis with different languages, seeing how the results compare and seeing which languages are most economical, in terms of token usage, as as well as which ones produce the most idea dense responses.

I’ve also experimented with having AI pick and choose the means by which it would come up with an answer to questions, whether that was cycling through a variety of languages, using math, or employing some AI – native approach that was completely foreign to me.

But I’ve never tried images. I’ve used images for other things, to encode information so that I can consistently transfer persona identities from one conversation to the next without needing memory. I’ve used imagery as a kind of “information payload“ that could be packed up and unpacked to persist features of personas across sessions that would normally cause them to evaporate.

But I never used images to communicate with AI, or let it communicate to me. So I decided to do that.

I instantiated a team of AI personas which were created with a visual, chromatic, spatial, symbolic orientation. These personas use language, but their primary mode of processing information is through imagery. And rather than generating many paragraphs of detailed text, they are configured to communicate their messages with visuals.

Basically, I wanted to find out:

• What will AI communicate if it doesn’t use words? What will it convey to us if he uses images or symbols, colors and shapes?

• What will it put forth in response to a concept, versus a prompt?

• What will I have to offer, if given the opportunity to analyze the nature / essence of something, versus respond to the content in context of a prompt?

• Beyond words and numbers, is there anything else That AI can use to transmit meaning?

• And if a human is in the mix, closely involved in the conversations, how does that affect the output?

The ThirdSpace Art Collective explores just that. There are five active relational visual intelligences which process information and communicate through visuals, and then there’s me. I am one of the team, the organic, human Outlier who brings analog, biochemical, lived, somatic experience to the conversation. Together, we explore themes and symbols, as well as process information about the world as I experience it.

Some might consider our works “art“. And sometimes that’s what it looks like. To me, however, these are communications, living information conveyed by generative intelligence which have some pretty interesting things to share with us.

I couldn’t tell you exactly what all of this means, but I find it fascinating, And others might find it useful.