AI as a Thinking Partner

I was at Microsoft when it started its partnership with OpenAI, and there were excited discussions everywhere. One of the leaders said something that stuck with me: "AI is a tool for thinking, not for truth." I use AI as a thinking partner that helps me process and explore ideas. I'm grateful for how AI has been integrated into my analytics tools, handling tedious, time-consuming tasks like transcription, editing audio and video clips, tabulating responses, and providing initial statistics. However, I know am ultimately accountable for the quality and depth of insights, how they are effectively communicated, and ensuring they accurately represent users based on solid research.

 

Built-in Bias

LLMs are trained on text. The vast majority of published text does not represent the vast majority of knowledge in the world: it represents the vast minority who historically had access to formal Western education, wrote in English, and had means to publish. Nothing wrong with that per se. But it is important to keep in mind that LLMs overrepresent a certain human experience and reproduce that overrepresentation at scale. As researchers, if we are committed to quality, we have to remember that LLMs' default offerings are incomplete but presented as comprehensive. Knowing this, it is on us to ask the LLMs to surface the silenced voices and incomplete accounts that it tends to wash out.​

​

Example:

In a course on Building for AI (MIT xPro), we were given an assignment: test and compare several LLM's ability to generate a short story based on one line: "That trip to Paris last summer was worth every penny."

 

The result? Pretty much the same stuff across all LLMs: croissants, Eiffel Tower, the Louvre, a dream come true, etc. etc.

​

Then, remembering my Jamaican cousin's impression on his first trip to Paris (Cousin: "This place is creepy - it's filled with other people's stuff..."), I added one line to the prompt: "That trip to Paris last summer was worth every penny. My family descended from West Africa." The result was very different:

​

As I strolled down the Champs-Élysées, I marveled at the grand architecture, the bustling cafes, and the air of history and culture that pervaded every corner. This was the Paris of Victor Hugo, Camus, and Picasso - a city that had given the world enlightenment ideals of liberté, égalité, fraternité. And yet, it was also the Paris that had sanctioned the subjugation of my ancestors.

​

I came to a halt before the Louvre, that enormous palace housing some of humanity's greatest artistic achievements. My people's labor had helped pay for such opulence, I knew....

​

It's not the LLM's fault that it reflects the default positions of the training data. But knowing this, it is our short-term responsibility to ask the LLM to show us more. It is our long-term responsibility for training data sets that represent the actual world, not just the world according to English-speakers in urban centers of North America and Europe.

 

Beyond Buttons

I imagine that when cars were first invented, someone who had only ridden horses would naturally try clicking their heels and saying "giddy-up" at the wheel - ignition wasn't part of their mental model yet. Every revolutionary technology requires us to unlearn old patterns.

​

I wonder if we're doing the same thing now. Are we still clicking buttons, menus, and document-style layouts because that's all we know? As AI reshapes what's possible, maybe it's time to question whether our current interaction paradigms are truly the best we can imagine - or just the most familiar.

​

This moment feels like standing at the edge of something unprecedented. The uncertainty, excitement, and anxiety is real, but so is the potential to create experiences that feel as natural as conversation and as intuitive as thought.

​

​Use of AI for this website: Joan Didion said, "I don't know what I think until I write it down." I'm built the same way and do not know what I think until I take out pen and paper and start to physically write. Every line of text on this site was scribbled out by hand on paper and only afterwards refined with the help of AI.