Watching the hype cycle build around OpenAI’s ChatGPT, I can’t help but wonder when the first New York Times or Atlantic story comes out calling the top – declaring the whole thing just another busted Silicon Valley fantasy, this year’s version of crypto or the metaverse. Anything tagged as “the talk of Davos” is destined for a ritual media takedown, after all. We’re already seeing the hype start to fade, with stories reframing ChatGPT as a “co-pilot” that helps everyone from musicians to coders to regular folk create better work.
But I think there’s far more to the story. There’s something about ChatGPT that feels like a seminal moment in the history of tech – the launch of the Mac in 1984, for example, or the launch of the browser one decade later. Is this a fundamental, platform-level innovation that could unleash a new era in digital?
Possibly, but the simpler co-pilot concept also resonates. It reminds me of a conversation I had over the course of a year or so with Rob Reid, the author of the prescient 2017 novel After On. AI plays a central role in the novel, and Reid introduces the concept of “centaurs” – creatures that are part human, part AI – borrowed from Garry Kasparov, who imagined merging with Deep Blue’s chess AI back in 2014. More colloquially, Rob and I drained more than a few bourbons imagining how AI could be more like a smart friend or assistant, rather than an evil force hell bent on destroying humanity.
Centaur-like behavior is already emerging across the nerdosphere. I was on a call with a scholarly colleague just yesterday, and he showed me two applications that leveraged large-language models (LLMs) to both supplant and enhance human communications – both of them in multi-billion dollar markets that currently support millions of workers. I’ve no doubt that AI-enhanced models are just getting started, and we’ll likely see huge VC investment in the space this year.
What I’m interested in is the nature of those investments. While OpenAI is positioned as the alpha startup in the space – reportedly negotiating a $10 billion injection of capital from Microsoft at a $29 billion valuation – I find the ecosystem that’s developing around the APIs these large-language models enable to be far more fascinating. For AI to reach the historical status of the Mac/PC or the browser, it will have to spark a massive surge of new companies and economic activity. Six years ago, Kevin Kelly predicted an explosion of startups that would “just add AI” – similar to the surge of Web 2.0 startups in the mid 2000s that “just added AJAX” or mobile startups in the 2010s that “just added social.” He was wrong on timing – but I think he’s onto something in the long run.
What might an ecosystem leveraging ChatGPT-like functions look like, I wonder? What categories are poised for true disruption? Where might “ChatGPT as a service” re-imagine large industries? Here are a few that come to mind…
- Customer service. Computers and script-driven voice mail trees already dominate this space, but we still have to deal with outsourced customer service agents who feel … worlds away and not particularly good. With clever parameters and programming, LLMs could truly change the game here. This firm is already leveraging ChatGPT in automated customer service and creating “super agents,” which sound a lot like centaurs.
- Search. I’ve already written a fair bit about this (and so have many others), but search is a huge business, and the potential for “conversational search” is tantalizing.
- Law. A lot of the current grunt work in law – essentially, what paralegals do all day, billed at $400 an hour – is just recombination of a massive, interlinked reference set of cases and precedents. I imagine there’s plenty of opportunity to code that into parameters for a ChatGPT-driven platform that enables massive efficiencies in legal costs (which sounds good to me…).
- Healthcare. Ditto here. Of course, as with law (and pretty much every other case here) we’ve got to be careful, but LLM’s ability to find patterns and report them out could truly disrupt healthcare, from the ridiculous maze of paperwork and insurance bureaucracy to diagnosis, novel protein and molecule discovery, and more.
- Media. I don’t think any AI system will create the equivalent of a well thought out article (and I’ll be writing a piece on why soon), but AI is already being used for clearly structured pieces in finance, sports, and the like. I expect that trend to continue. When I was coming up as a tech reporter in the late 80s, my first assignments were to rewrite press releases. I stood out because I actually called the companies and asked follow up questions – but that wasn’t industry standard practice, and it still isn’t. ChatGPT, with the right guardrails, could probably do the same job – and for nearly no cost. I’m not sure that’s a huge, investable market, but it’s a leading indicator. Plus, tons of “real” writers and creators will use AI like centaurs – as muses and prompts. Media will certainly be changed forever by this technology.
- Coding. This is obvious, and well-reported, but like media, a lot of coding these days is repetitive and rote. Plus, just like with writers, coders are using AI to prompt their work to another level.
OK, I’m going to stop here and ask what you think might change thanks to ChatGPT and LLMs. Does ChatGPT mark a truly historic breakthrough, like the Mac/PC or the web browser? Let me know in comments, or email me direct – jbat at this domain (battellemedia.com). Happy prompting, folks…
Jonathan Glick from Mastodon says, aptly: “You didn’t mention Education, John. That, to me, seems like the most potentially interesting and impactful.”
As someone who works as a law clerk, I see a couple of potential problems with how ChatGPT could disrupt the law industry in the previously mentioned ways.
First, ChatGPT would need a more powerful search engine than legal research websites such as LexisNexis. It would need to understand specific details about the case that’s being tried and how a particular case is more appropriate to refer to as a precedent than another. There would need to be an algorithm that is able to weigh the intangible factors that are common in legal cases, such as determining a person’s intent, awareness, and level of responsibility regarding certain actions either they carried out or were affected by in order to relate one case to another. Some of these factors are very difficult to quantify, so that could be an obstacle. I think this problem would also come up when ChatGPT would refer to a precedent-setting case in a legal document as it would need to be able to identify and explain the implications of the precedent-setting case and how it relates to the case being tried.
Second, if law firms wanted to reduce the total cost of paralegals by using ChatGPT, they’d run into a few obstacles. They’d need someone who knows how to use ChatGPT in a legal context as well as read, edit, and revise what ChatGPT had written. That means they’d need to train paralegals in this skill, which could increase their hourly rate. It’d be interesting to see who would be the early adopters as well as how long it would take for them to get a return on their investment.