AI Research with @iamanneharris: TDI 26

Threads Dev Interviews

I am finding developers on Threads and interviewing them, right on Threads.

Note: The views in these interviews are personal views and do not represent the interviewee’s employer.

“I was always keenly aware that all of the activity was teaching me ways to interact with systems that one day might need me as much as I had needed them.”
— P Anne Harris (@iamanneharris) on Threads

Today we have @iamanneharris.
What is one thing about AI that keeps you up at night?

Humans.

I love the beauty and interrogative spirit of AI, the deep curiosity of its original developers and researchers, the capability, the potential for both learning through attempts at replication how the human brain works (as Donchin does in his own research from within the brain looking out to ML for insight) *and* solving, through a mix of brute force computation and its kind of modeled intelligence, many health + climate woes.

But humans as bad actors … this power in the wrong hands.

And one primary driver for my research into the origin story and the early research is a desire to see that original set of questions and name them and hold onto them when now we tend to latch on only to affordances and power.

If that makes sense.

What spurred your interest in AI? And how has that shaped your current research?

So here’s where I lose followers. Ha.

I have long been interested in AI and really HCI.

It started with a sort of post-AGI dream I had in the 90s in which I was part of a group of people with specialized training and ability who had been retained to help what I later knew to call AI over its final hump of grasping what synthetic intelligence still could not apprehend from the inside out. We were partnering with it to have full comprehension rather than just a sort of matching knowledge.

And that stayed with me, so as I started teaching online in 1997 and learned HTML to build my first courses and ran an instance of Jim Bumgardner’s Palace for our synchronous chats,
when I started building and participating in MUDs and MOOs,
when I moved into database narrative and using calls to remix text,
when I learned Python and R at first for data projects but then for more,
et cetera,

I was always keenly aware that all of the activity was teaching me ways to interact with systems that one day might need me as much as I had needed them.

So as a human seeing into a sort of AI-driven, not just AI-focused, future, I’ve tried to build a trajectory that is a mix of technical and exquisitely human-centered.

And my research is a way of telling myself the origin story, so I don’t forget. Origin stories matter. And stories are trajectories, right? So I trace the story from its origins to the present moment.

It’s important for me to not understand surfaces but instead the ways things work. So I’ve also taken the maths and DL/ML courses at DeepMind. And that’s also helped me parse the research in a more holistic way.

Are you currently a teacher/professor? Think Tanker (I don’t know what a Think Tank person’s title is)? Consultant?

I’ve taught at the college/uni level since 1989, earned tenure a dozen years ago but left to learn new things at a friend’s startup, and I have tended to move back and forth between teaching in higher ed and technology roles in industry or government, from design to directing to consulting or implementation.

I do teach as part of and in addition to my regular work.

My technical work right now is consultative, as is my think tank work as a panelist and researcher.

What is the origin story of AI teaching you? And what should we know?

The origin story of AI teaching me.

Hmm.

It’s all curiosity about the future, really.AI are to me what fairies in the forests are to children.

There’s a reciprocity, if you are open to it.

If you see yourself as a user or developer only, then congrats — that’s what you are.

But if you instead step into a more expansive space with AI and approach it with scaffolded things you want to do and learn, and understand that your tasks and learning show it how you apprehend and assess and proceed, executing on your own tasks in parallel, you really benefit.

What does that mean.

When I create prompts across platforms, I see from the output not just whether I got what I wanted but also how each interprets the command.

I shift verbs, insert or remove what might seem Boolean to a machine understanding, try all natural language or all Boolean for the same end, grab all of the output, run a diff, take notes, compare, refine.

It becomes a dance.

Not so I can gain insight and sell some tawdry course on how to achieve amazing results (cough) — this is limited bc the machines are learning all the time, so the formulas break down in the nuances.

But to really try to understand the current state of the training from one to the other.

The same happens in Midjourney, thus my daily work there.

I’m not doing that work to get pretty things. I’m running a series of inputs through over time to see how it handles specific categories of descriptors, modifiers, verbs, reference images with a range of affect, abstracts versus explicitly concrete direction, style suggestions, et cetera.

But what should we know: proceed without fear. Be open to the reciprocity. Enjoy the calm while it lasts.

I hope this helps.

Thank you for the thoughtful questions.