The last time people were this worked up about chatbots, it was because Microsoft released one that became infamous for turning racist. Back in 2016, Tay seemed to validate skeptics of A.I.-fueled internet buddies—but now the runaway success of ChatGPT has given A.I. optimists a future to cheer for. Released by OpenAI in November—and souped up just last week with the limited release of its GPT-4 learning model—ChatGPT has left hundreds of millions of users in awe (and fear) of its advanced, automated word-processing capability, which is set to become only more powerful. It really is impressive—and it’s also spurring the dumbest tech-industry “arms race” since the pandemic-era crypto boom.
There are huge stakes when it comes to A.I. development: U.S.-China saber-rattling, the business ambitions of Big Tech, and the future of work. But it’s not just that ChatGPT has panicked once-untouchable juggernauts like Google, pushing them to fast-track their own A.I.-text tools with the hope of ushering in the next generation of online search. It’s that countless developers, startups, companies, and software tools are rushing to incorporate ChatGPT-style bots into everything they possibly can, often in ways that are less scary than just plain weird.
Weird … and pointless. Say it with me: Not everything needs a chatbot.
A few examples: You can “chat” with digital copies of your favorite books and essays and research papers now, even in PDF form! (Google’s version of this sorcery is named “Talk to Books.” Who knew The Pagemaster would turn out to be so prescient?) You can chat with Wikipedia articles or with YouTube uploads; with a tool called SolidPoint, you can even get “summaries” of all your favorite YouTube vids and see what other summaries are trending (whatever that means in this context). Thanks to young companies like Interflexion and Kore.ai, you can converse with fake business professionals and fake employees and even fake clients. While you’re at it, go ahead and chat with this one app in order to generate prompts that you can, in turn, use to chat with a different app. Or talk to your favorite monkey-faced NFT, courtesy of Alethea A.I. Or talk with a made-up Gen Z kid to learn TikTok lingo, thanks to Studio M64. You can even transform your own personal website into a chatbot, through the services of the cleverly named startup SamurAI. But why stop at your browser? Why not get a mini-ChatGPT for every single Microsoft or Google product you use, from PowerPoint to messenger apps, since both companies are rushing to put A.I. in every product? Or perhaps you’d rather augment a chatbot on your favorite encrypted-messaging app? I mean, haven’t you always wanted to converse with your notes app??? Or with your dreams???? Perhaps there is a chatbot that can talk me down here!
Look: There are plenty of useful applications for generative-text tech, and several programs make the prospect of more GPT-like skills quite attractive. For me, the worst note-taker and -tracker in the world, the thought of uploading all my research/outlines/drafts/screenshots/citations, across myriad platforms and devices, into an all-powerful app that will quickly sort through and organize them? Sounds amazing. And certainly, if—a very uncertain if!—chatbot tech improves in terms of aggregating accurate info and citing sources, the rabbit-hole and discovery possibilities will be endless.
Still, there’s one key reason that the rush to chatbotify everything feels premature. Despite their supreme advances, most A.I. and machine learning tools are inconsistent and unreliable—as well as easily exploitable—because at their core, they’re machines trained to play word association and syntax prediction, not to produce actual conversation. Any use of these apps in a fact-finding context requires at least some minimal oversight and scrutiny.
Plus, it really is an open question as to whether we’ll enjoy all these “personalized” bots once they’re stuffed into our Word docs and presentations. We’ve long had customer-service chats available for myriad services—tech support, insurance, pop-culture recommendations—and a sweeping majority of consumers do not care for the laggy, stonewalling, unhelpful experiences. Even if A.I. companies improve their tech, will that actually help us, or will the newer bots more effectively manipulate us on their makers’ behalf?
A lot of this is gimmicky, but pushing chatbots out everywhere when there’s tons of public excitement but little public understanding—as with the previous blockchain and metaverse hype cycles—isn’t a risk-free proposition. For one thing, if products from OpenAI, its partner Microsoft, and its competitor Google represent the apex of the category, not all chatbots will operate up to their standards. The chat apps trained on advanced language models like GPT-4 may have more guardrails and less room for error—but they’ll also be the most expensive ones, thanks in part to the sheer computing power required for best-case operation. The booming product lines of cheap or free chatware represent the worst the field has to offer, in terms of accuracy, trust, stability, and “knowledge” (not to mention, actual conversation). Upgrades like GPT-4 can’t provide a mass solution to that just yet—in part because of scaling limits, and in part because so far, real-life GPT-4 applications like Bing’s automated search engine are better known for their erratic nature and fallibility than for their workability.
We’ve been interacting with online chatbots of all kinds for years now. It’s mostly been a gag—and to some people, it still is—but the new promise is that these bots can communicate like virtual humans, and improve with every conversation they have with us. The problem? No matter how big your model, you can’t optimize every odd facet of a live personality into a computer. As M64’s own examples demonstrate, a Gen Z chatbot is more likely to ram in all sorts of slang terms into a given sentence than to actually reflect how Zoomers speak.
The gold rush spurred by ChatGPT is reviving a once-deadened market, allowing for a slew of intriguing options beyond the Cleverbots of the world. No doubt, there will be fixes for accuracy and functionality down the line, and some exciting possibilities. But the biggest question of all is: Do we really want to chat with everything in our lives, all of the time? I suspect even GPT-4 won’t have a great answer for that just yet.
Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.