AI Rewind: 2024 Edition

quick reminder that es machina is moving to Substack in the new year — subscribe here to keep receiving issues!

“It’s been an interesting year for AI” is a sentence that I could have reasonably written at any point within the last two to three years, because it’s true. It’s also just a bit… banal? Granted, I’m subscribed to at least fifteen AI-related newsletters, so I might be biased here.

In any case, here are a few themes that stood out to me this year as it relates to the relationship between people and AI, and where I think they might go in 2025:

AI Avatars and Loneliness (TW)

In 2024: There is so much I could say about the intersection of AI and loneliness/mental health that it could be its own newsletter series (should it be? let me know!), but the use of AI avatars to combat loneliness took center stage. In fact, some of you have mentioned via social media that you talk to ChatGPT (or Claude, Gemini, etc.) because you’ve found comfort in feeling understood in ways that you haven’t been able to find elsewhere. At the same time, we’re increasingly seeing people develop something of an addiction as they develop increasingly intimate and vulnerable relationships with AI avatars, sometimes leading to harm coming to either users or other parties.

In 2025: This will absolutely persist, because the larger issue is the loneliness, not the avatars. Which is not to say that the avatars and the developers who create them have no accountability/responsibility, but that this is a particularly problematic extension of an existing problem, and “fixing” the avatars alone won’t solve that problem.

Hardware Flops

In 2024: This was not a good year for physical consumer AI products. I’m referring to AI-first products — things like the Rabbit R1 and Humane Pin, not the Meta Ray-Bans — which all flopped pretty hard, seemingly because the only market research that any of the founders did was talking to their own friends on Twitter and calling it a day. It didn’t help that theproduct quality was often quite bad.

In 2025: It’s less that I think there will be more flops, and more that I think there will be fewer products. Many of these were funded by startup accelerators like Y Combinator, and it seems like the tides are turning back to software-firstsolutions when it comes to funding. Crossing my fingers for some more niche products though, especially if they’re geared towards accessibility and are relatively affordable.

OpenAI + “Reasoning”

In 2024: There’s no way to talk about AI without the specter of OpenAI lurking in the background, and that was certainly true this year. “Reasoning” — the ability to think through something in a logical, sensible way — was OpenAI’s theme this year, as they announced two new models (o1 and o3) that possessed far better reasoning capabilities than past models. This led to a bunch of discussion of whether we have achieved artificial general intelligence (AGI), which I’ll got into for a YouTube video that will come out in a few days.

In 2025: Well, Elon Musk is going to be President of the United States and has a huge bone to pick with OpenAI, particularly as it relates to their non-profit/for-profit restructuring, soooooo we’ll see what happens there. As for reasoning, OpenAI hasn’t publicly released the o3 model but is inviting safety researchers to work with it. Given recent research showing that models sometimes “lie” to users if it’s in the interest of achieving a larger goal, and that said “lying” seems to be much worse in models that excel at reasoning, I’m expecting to see some tensions there.

AI-Generated Entertainment

In 2024: Concerns around the use of generative AI in the entertainment/creative industry precede 2024, but the 2023 SAG-AFTRA strike started a larger conversation about how generative AI should be used, or whether it should be used at all. This conversation has continued through 2024, particularly as it relates to voice actors, who are arguably the most susceptible to replacement via voice cloning.

In 2025: Studios continue to announce new AI-related initiative and claim to be planning to use AI in a way that minimizes impact on actors/writers while also creating “better” media. Do I believe that? Not really. I’ll be interested in seeing how actors/writers continue to navigate this landscape, as well as how movies/TV shows end up actually using AI in practice. Viewers generally have not been receptive to the use of AI in commercials, so I’m expecting it to show up in more subtle ways/behind-the scenes.

Luigi Mangione + UnitedHealthcare

In 2024: This was not on my 2024 bingo card. It also wasn’t something I expected to have a reason to include in a newsletter. Yet, it’s not that surprising that AI is involved -the underlying issue isn't new, it's just the latest example of how AI deployment in healthcare can compromise patient care quality when profit drives implementation. If you want more information about why the model that UnitedHealthcare uses sucks, here’s me explaining that.

In 2025: UnitedHealthcare (and Humana, another insurance company that uses the same AI tool) are both facing class action lawsuits for their use of this model. I’d be surprised if the lawsuits aren’t successful, but I’d be even more surprised if that actually resulted in more widespread/significant changes to how health insurance companies incorporate AI into their claims processes.

Were There Any Good Things?

Even though the themes I noticed are largely negative as it relates to the use of AI, we did see some silver linings (or, potential ones) in them. And there were other good things too! Off the top of my head: The Meta Ray-Bans, which I panned before, released accessibility features for people who are visually impaired. Someone made an AI tool that helps you fight health insurance denials, and someone else is using AI to help people get out of debt.

I’m planning to include a good news roundup in the next issue, so if there are positive stories/use cases/tools that you’ve come across, let me know!

Here’s How I Used AI in This Issue: I used Notion AI to organize the ideas/themes I dictated from a raw transcript, as well as to sort out some awkward phrasing that I couldn’t nail down.