Should We Write Differently for AI?
I watched the video of Reid Hoffman interviewing his AI self and, if you haven’t seen it, I recommend that you take a few minutes to watch it (see below): it’s an interesting demonstration of the art of the possible when it comes to personal and personalized AIs.
After thinking about the video and, more importantly, the idea that it represented, I found myself thinking about what I’d expect of and hope for in a Dennis AI. I’d love to be able to query, reflect on, and refine my thinking over time, but I’d also want other modules that would support and contribute to my efforts at bettering myself.
Reid Interviews Reid
First, Reid’s video if you haven’t seen it:
The AI avatar gives a pretty good sense of the data it uses: “I’ve read all his books, watched all his speeches, and listened to his podcasts.” In an effort to get a sense of what that means, I ran a couple searches:
ReidHoffman.org, the online repository for his posts and essays, has 282 pieces dating from 2012 to 2023. If we assume an average of 7 pages (based on a very lazy sampling of the first two pieces), we’re looking at over 1,900 pages of thinking. I’d assume more as that no new pieces have been added since November 2023.
Looking at his five books on Amazon, there’s something in the neighborhood of 1,400 pages of material.
Not in this calculus are articles in which he’s been interviewed, op-eds he’s written, talks he’s given, panels he’s sat on, podcasts that he’s done, etc. For the sake of argument, let’s assume there’s another 1,900 pages of content at play. I suspect that number is low.
So, at a minimum, Reid AI is working with 5,200 pages of content that reflects human Reid’s thinking on a variety of topics.
Yes, and…
When I had the opportunity to work with the good folks at IDEO, they broke me of the habit of saying “Yes, but…” and instead saying “Yes, and…”. The former is often a idea killer whereas the latter opens the door to more creative thinking.
After watching Reid’s video, my “Yes and…”s revolved around expanding the scope of the data that Dennis AI—a hypothetical personalized AI that represents me—would draw on:
The articles that I’ve written and talks that I’ve given;
Articles that I’ve read and talks that I’ve attended;
Tabs that I’ve opened;
Books that I’ve read;
Movies and television shows that I’ve seen;
Content that I’ve shared with friends; and
Content that my friends have shared with me.
The diversity of this content is important because it reflects the ideas, insights, and information that have shaped—and continue to shape—my understanding of the world and my world view.
The Mechanics of It All
In thinking of all these inputs, I’d need a seamless and frictionless way to update the information that informs AI Dennis’ algorithms. I’d want a simple and painless way to:
Highlight the passages that we (or our friends) found most interesting; and
Comment on why the passages were interesting (or, for lack of a better word, infuriating).
For books and movies, my thoughts and observations might relate to the book or movie as a whole (either as a standalone product or as a data point in our relationship with the franchise or genre), a chapter or scene, a character (or actor’s performance), a quote or a scene, etc.
The reason that I think this is important is that it would (should) help our AIs when they acts as agents that scour the web on our behalf. To know what topics we’re interested in as individuals is only half of the problem: our personal AIs also should have a sense of why we’re interested in the topic (or what dimensions of the topic we’re most interested in).
There is No Single Dennis AI
In listing the items out above, it occurs to me that we’re bumping into the problem of all things social: there is no single Dennis on which to model a Dennis AI.
There’s Dennis the increasingly curmudgeonly Gen Xer, Dennis the recovering political analyst and analytic methodologist (who thinks about the vocation and practice of analysis a lot), Dennis the futurist, political Dennis, philosophical Dennis, idealist Dennis (and his cynical alter ego), there’s Dennis the aspiring designer and architect, countercultural Dennis, whimsical Dennis, and there’s probably a couple different cultural and pop cultural Dennises.
Each of these facets of the real Dennis vary in degrees of depth and breadth knowledge of the topic being discussed, and the understanding, seriousness, humor, irreverence, empathy, compassion, frustration, outrage, etc., with which I approach a topic.
Which facet of myself that I show to or share with the people (or entities) I am interacting with varies in response to the interaction (i.e., topic, setting, structure of the exchange, etc.) and in response to past interactions (and my feelings about those interactions over time). Cultivating Dennis AI requires something tech companies…struggle…with: respecting individual privacy and giving people control over what information and insights they share with whom.
All the Things I Don’t Know
One of the things I love about AI is my ability to have a conversation with it about the things I don’t know.
Because these conversations tend to be informal, I take the answers I receive with a pound of salt. If the answer is interesting enough, I will try to see if the AI can give me a list of the most authoritative voices on the subject or go off and run a bunch of searches to see what I can find.
This AI persona should know what Dennis AI knows and should know when to ask if I want a line of inquiry rolled into core Dennis AI. Perhaps more interestingly would be when AI Dennis suggests that I invite another personalized AI into the room. If I have questions about the most recent Portland Thorns’ game, perhaps I could talk with AI Kling in the context of the season and/or the history of the matches between the Thorns and whatever club they just played? Might a college professor, actor, director, writer, politician, or bureaucrat be pulled into conversations as needed and appropriate? How do I compensate them…and how is the cost of interacting with their AI avatar determined?
So What Does this Have to Do with How I Write or Read?
This has been a meandering journey and, if you’ve made it this far, I appreciate your time and patience.
My takeaway from Reid Hoffman’s video is that it is interesting, if not incomplete. Today’s Reid Hoffman is different from last week’s Reid Hoffman and just as he is different from next week’s Reid Hoffman by virtue of the ideas, insight, and information he’s consumed and created…and that’s OK because the video was intended to demonstrate the art of the possible.
I would be interested to see how Julia Mossbridge would build onto the model so that we are more compassionate and charitable toward ourselves (and, hopefully, toward others). I’d also be interested to see what a philosophy module would look like or a wonder module or a gratitude module. Each of us contains several selves and the ability to talk with a virtual “self” that understands our hopes, aspirations, needs, wants, and fears and could talk with me about them in a dispassionate and supportive manner strikes me something that could potentially be a powerful force for human betterment toward the good.
Returning to my initial questions about whether I should, in an era of AI, read or write differently: if I think about my writing in the context of it being fodder for AI Dennis, I feel like I need to be more explicit about what I have questions about, what I simply do not know, and how I feel about the topic. Dennis AI should have a clear sense of what I knew (or didn’t know) and when.
Catching AI Dennis up to where I am in terms of my thinking, how I got here, and where I am going is going to take some doing: how might we seed and build the AI’s knowledge base and graph? If one of the functions of AI Dennis is to provide me with a means of reflection and introspection, it should know where my knowledge and thinking is and ends so it can ask me questions that push me to broaden or deepen my thinking or so it knows not to pile onto what I know (or think I know) in unhelpful ways.
Most importantly, I need to know that I can trust the company or companies behind AI Dennis. What legal and ethical guardrails exist? To what degree can I customize them depending on who is accessing the AI and what topic they’re discussing?
The idea of a personal and personalized AI is really appealing but if I don’t derive value from or benefit from this AI Dennis, it becomes another distraction vying to fill the gap for intuitive and easy to use knowledge management system in my life…