What makes a good AI name?

You’ve just spent millions of dollars, countless computing hours, and untold gallons of cool water steeping a neural network in the sum of internet knowledge. Now, just one question remains: What do you call this thing?

That’s where David Placek and his naming agency, Lexicon Branding, might come in. If you don’t know the agency’s own name, you definitely know some of the names it’s coined: Febreze, Dasani, BlackBerry, Sonos, the Subaru Outback.

Situated in the houseboat haven of Sausalito, California, the small agency has worked with most of the tech giants jostling over AI supremacy on the other side of the Golden Gate Bridge. And lately, Placek has been spending a lot of time thinking about what to call various chatbots.

After “numerous requests from companies to name new AI technology brands,” Lexicon recently polled around 350 consumers and developers in the US and Germany to try to understand what makes a good moniker for this new wave of generative AI models.

Based on the findings, Placek might encourage clients to use something that brings to mind a piece of the natural world, like Meta’s Llama models, as a way to “convey a sense of simplicity, calm, and ease,” according to the report. Words that are “relatable” and “capable,” and include a commonly used suffix or prefix, like Co-, as in Microsoft Copilot or Cohere, can connote “collaboration, competence, and a sense of being in control,” the findings suggest.

“It’s better to make these things approachable and able for the human mind to process in a really simple way,” Placek told Tech Brew. “Natural would be a good way to do it. Until that becomes a trend, and all of a sudden, there’s mountain this and sky that and ocean that.”

Stay away from “overengineered and overly inventive names” and think twice before tacking on a dot-ai domain suffix, lest you fail to deliver on that hype-laden appendage, the report advises. As for people names—think Anthropic’s Claude or Amazon’s Rufus—“our research shows that although some human names can carry human-like associations of being friendly, creative, or even visionary, they may also be less likely to be seen as reliable, secure, or responsible,” the report says.

NaM-E evolution: For much of the history of large language models (LLMs), researchers didn’t seem to give that much thought to model names as brands. Clunky acronyms and alternating cases abound—ELMo, LaMDA, even the GPT family.

But now that tech giants are starting to fold them into big-money commercial products, slicker, more consumer-friendly names are starting to emerge—Copilot, Claude, Gemini. Still, there’s progress to be made on bringing more human connection to these names, Placek said.

“We’ve seen in the research that there’s not a lot of emotion that consumers…are taking away from this,” he said. “We still don’t have that kind of Nike sort of emotional thing, self-expressive behavior here.”

Sonic vroom: Employees at Lexicon, as well as its network of dozens of linguists around the world, drill down on how sounds roll off the tongue and the emotions they can conjure. Many times, “sound symbolism” will transcend language, which can be useful for global companies, according to Placek.

“The interesting thing is that many of those sounds are global in nature,” Placek said. “I always give the example to take the letter V…Based on research we’ve done, V is the most alive, energetic sound in the English alphabet. You think of things like Corvette and Viagra.”

While Placek said he can’t yet talk about any of his agency’s AI naming projects, he did walk us through the thinking behind Intel’s Tiber platform, which Lexicon named and would “meet our standards for a very effective…AI name,” he said.

“There’s something natural. It’s a major river that flows through Rome. Rich in Italian history. You set that aside, and you have something that is from a structural standpoint—it starts with a consonant, back to a vowel, back to a consonant—really easy to pronounce,” Placek said. “You may not know the Tiber River and its history, but it’s…got that active ‘-er’ [sound]. T is one of the most reliable sounds, so all of a sudden we have something that’s very easy to pronounce, it’s simple, it’s short, and there’s a backstory about this river that comes into play.”

As generative AI continues to evolve, Lexicon recommends that tech players move beyond names that promise hype and zero in on those that tell more of a story. The company plans to field more studies over the next couple years, hoping to better understand more nuances of naming an AI.

“We really need to begin to develop our own insights into what people are thinking, what developers are thinking, what are our clients thinking, and how we can develop a body of knowledge that helps us develop better names in the future,” Placek said. “AI is not going away.”

—--

To give an insight into his process, Tech Brew asked Placek give quick initial thoughts to some AI names generated by ChatGPT:

MuseGen: “You have some wisdom in there, generating wisdom, something like that. Not a bad name because I recognize “muse” and “gen.” It’s just that the assembly there is a little bit awkward. So you know, it’s Muse and then Gen, so it’s start and then stop. So it doesn’t have a nice flow to it. Not an ideal name. But in a pinch, it’s probably usable.”

PrometheusAI: “OK, so Prometheus, the bringer of fire, right? If I remember mythology. So you’ve got AI there, you’ve got a real person with a story of Prometheus. The only problem is it’s a little bit long. Spelling will be a little bit difficult. But it’s better than MuseGen. So we’re making progress.”

DreamWeaver (this happens to be the name of a real Adobe product): “DreamWeaver is two well-recognized words that are meaningful to most humans—the notion of dreaming, right?…So I think it’s an expression that people might use, you know…‘He’s kind of more of a dream weaver.’ ‘He’s a dreamer.’ So, more on the human side, I think that’s probably easy to spell by an English speaker or someone with English as a second language. So I would rank that No. 1, Prometheus No. 2, and MuseGen No. 3.”