Road Trip to the Future: Exploring AI

A friend called me in April, early on my road trip, to find out what I thought of AI. I spent the last ten years of my career as a thought leader in predictive analytics for life insurance applications, so the question was not out of left field. Sadly, I had to admit I didn’t think of it much, only that it seemed that it had recently taken over all of the airwaves. It was a topic at the insurance conference I went to, it was popping up in my news feeds, it was all over podcasts, and it felt that with every week I heard the term ten times as often as the last. Given I was looking ahead at 4000 more miles of driving before the end of May, I decided it was a great opportunity to catch up.

Looking back, I had ignored AI for years. I thought of it as a catchier name for machine learning, which I thought of as telling a computer to search for the best model it could find within a given set of parameters. Essentially nothing more than what a human could do if given unlimited time and/or computing power. With that perspective, it didn’t sound particularly compelling for me to learn more.

After a month of catching up, that is not how I think of AI today.

I am still not an expert. I am writing this post to capture what I think of AI today, how I hope to start using it in my own life, and how I hope to keep up to date with new developments. I hope that this will help others who are currently where I stood a month ago, and that those who are further ahead in adopting, understanding, or developing it, will give me hints on how to go further.

AI today

Today AI is many things. The most common way I hear current AI capabilities described is the large language models that live behind chat bots that can help with many generative language tasks. For example, in response to a written prompt or request, they can write blog posts, emails, poems, or any of these in the style of Shakespeare (or your mom), or they can write code (in your mom’s style I assume, if your mom has a known style for that).

All the large language models are trained on text. Some of the models are trained on images as well, and those are therefore able to generate images, including videos. Those trained on music can generate music. Those trained on code can generate code (though really, code is text, and what is code but another language anyway?).

Sometimes its answers are great. Sometimes they’re terrible, or awkward, or not quite what you wanted, or outright untrue (that’s called hallucinating). The more text, images, or music that is fed into a model during training, the more the generated content will conform to how we expect a human might respond to the same request. With each new release, the answers improve.

What does that mean for the average human?

It means your kids (or your friends’ kids), from grade school through college, are going to ask ChatGPT to do their homework and to figure out how to talk to that cute boy. Don’t take it from me, watch South Park.

That’s today.

AI in the future

Tomorrow, with a little encouragement in the right direction, those same kids are going to understand how and why AI gave them those homework answers, ask the AI to improve its own answer, and then tweak the answers a bit to have a genuinely unique dissertation to submit to their doctoral thesis committee.

That will be the future for the kids who start using AI today with a growth mindset. They have a goal of reducing their workload so they can get out to play. They will try and try to ask the AI in different ways to finish their homework. They will learn the best ways to make their requests, and they’ll spend less and less time on the portions of homework they deem unimportant. We can look at that as cheating, or we can observe that they’re learning how to use a new tool. If they’re lucky, they will have teachers who foster this exploration. They will fail over and over and keep trying, sure that there is a solution they’ll find eventually. And they will find it.

How I plan to approach AI

AI isn’t going anywhere.

I am trying to approach today’s AI like a kid with a growth mindset. I believe I will find ways to use it to improve my life. I don’t know what they are, so I’m trying a whole lot of ideas and seeing what sticks. That turns out to require two paths of learning: exploring how others think about it and use it [intake], and using it myself [experimenting].

I hope to find more channels I enjoy for to keep tabs on the latest developments and new ideas to try and to keep experimenting. I will look for tasks I don’t like spending time on and see if a chat bot can take it off my plate, at least a little. If it works, great! Either way, I’ll look for the next thing I want off my plate.

These are some highlights of what I’ve found so far.

Intake

I’m listening to the podcast Possible to hear a positive take on how the world is changing, because without something positive I found I just stopped wanting to hear anything at all about it. I’m listening to Ezra Klein to hear interviews with key players in the field, and to at times get a cautionary spin. I’m following Zain Kahn on LinkedIn to see ideas for how to use ChatGPT, or to learn about apps that leverage LLMs to help with specific applications.

As I was driving, I queued up a bunch of content and let it play and wound up listening to an entire hour of a podcast that was just two dudes having a conversation about using one of the LLMs to help them plan a vacation in Turkey (they weren’t going to take the trip, it was purely a demonstration). It was horribly boring. But that content is out there. It highlighted the fact that certain tools now have access to the internet, and they have creative modes, and they can help you with more and more real-world tasks that you never thought you wanted an AI to help you with.

Finally, after hearing multiple recommendations for people to follow on Substack, I a) figured out what Substack was, b) signed up and followed two authors, and c) discovered that I could listen to any article written on Substack by clicking the headphones icon in the app on my phone. Even when I’m not on the road, I find that I’m more likely to make time to listen to a source than I am to sit down and read one. If my goal is to really learn the details, the best is if I listen and read at the same time, which is also possible with Substack. If I don’t need to file away all of the details, the audio alone is enough to get me up to date on concepts shared, and I can absorb it at that level while driving around town, cleaning the house, cooking, or falling asleep.

My confidence in my ability to stay up to date got a big boost when my first recommended Substack author (Ethan Mollick) was resent to me by a second friend, and then he appeared as an interviewee on the podcast Possible in the same week.

Experimenting

I’ve used ChatGPT and Bing successfully in first drafts for emails and blog posts, preparing for meetings, finding (old) articles online, and thinking about how to structure my time as an independent consultant. I’ve failed at getting it to translate code from one language to another, at getting meal recommendations based on what’s in my fridge, and at getting it to write a song about recent developments AI in the style of We Didn’t Start the Fire. I even asked AI (Bing in creative mode) to edit this blog post, but in the end it didn’t sound like me anymore so I passed on its edits.

A few words on what AI is not (today)

So far, there is no AI model that is an AGI, or Artificial General Intelligence. Per Wikipedia, that is something that can learn to accomplish any intellectual task that human beings or animals can perform.

It is not a paperclip making robot (today).

It is not going to take your job (today).

Conclusion

These are my thoughts on AI as of mid-July, 2023.

Thanks to Frank Chechel, Julia Romero, Joshua Rusch, Sarah Hoffman, Marshall Lagani, Zachary Stenberg, Liz Ray and many others for their thoughts and suggestions that helped me learn!