"That's like ChatGPT!" my dad recently shouted at the screen as we watched Commander Sisko talk to the station's computer on Star Trek: Deep Space Nine. But the best example of how to talk to ChatGPT comes from a different character on a different Star Trek show: Geordi La Forge of TNG.
Star Trek: The Next Generation was the first of the "legacy" Star Trek shows to presage the rise of large language models almost uncannily. But with the exception of Geordi La Forge, the characters on the Next Generation were comically bad at using the ship's LLM— what they called "computer"— to its fullest potential.
Picard and Most of His Crew Would Hate ChatGPT
The Enterprise D crew's ineptitude with generative AI is highlighted in this 7 minute supercut of the cast interacting with the ship's voice interface:
Watch the whole thing if you've got the itch. But the tl;dw here is that most of the characters in this video are clearly not having fun or getting what they want. The standout character, of course, is Geordi La Forge. At 2:54 in the video, he shines as a relaxed, confident expert 24th century user of AI. But what is he doing right? And how are his crewmates getting it so wrong?
Geordi Would be a Next-Level ChatGPT User. Here's why.
Even if you didn't watch the supercut above, you should definitely jump to this 2:54 timestamp and watch Lieutenant La Forge work his magic as the large language model whisperer.
Geordi treats the ship's computer as a partner.
What is LeVar Burton's iconic character doing right? Two words: thought partnership. Geordi approaches the computer in a way that is collaborative rather than pat and demanding. After a basic opening question about brain patterns that primes the computer to start thinking about the topic at hand, he says:
"OK, we're going to track down any possible cause of that brain pattern. First, what are the medical conditions that might account for it?"
Notice his use of "we." After initially priming the computer's LLM-style "brain," Lt. La Forge uses language that indicates he is collaborating with the computer. This teaches the computer to expect dynamic, back-and-forth conversation, to expect thought partnership. And Geordi keeps that up throughout his conversation—lots of "we" and "let's" to encourage the ship's computer—its LLM, if you will—to see him as a teammate.
Throughout, the tone he takes with the computer is collaborative and conversational. He uses natural-sounding transitional words like "OK" and "first" to introduce the computer to new concepts and approaches. He uses tentative words like "might" to encourage the computer to explore and analyze unconfirmed possibilities, rather than just looking for concrete "knowns."
Geordi patiently feeds the computer context.
And when Geordi encounters outputs that aren't helpful, instead of sighing heavily or squirming impatiently, he gives the computer another key ingredient of a successful interaction with a real, 21st-century large language model: guiding context.
When the computer suggests that "Erocene Syndrome" could be causing the brain wave, Geordi doesn't act frustrated or even show mild irritation. Instead, he gives the LLM context to help it look beyond Erocene Syndrome. He does this by saying:
"Right, and Doctor Crusher already eliminated that. So, how many non-medical forces or substances might cause the pattern?"
This context helps the computer understand that it should be looking beyond the realm of medical pathology.
Near the end of the "Geordi" portion of the YouTube supercut, things get interesting. Geordi finally gets a little impatient and frustrated. But here, he actually does something constructive with that frustration, unlike the rest of his colleagues on board the ship.
Geordi knows when-- and how-- to push back.
Geordi engages in pushback. Savvy users of ChatGPT-- and the one savvy user of the Enterprise D's computer-- know the value of this prompting tool. When an LLM gives you something odiously unhelpful, don't just dismiss the output and the LLM's ability. Instead, tell the LLM why it got things wrong, and tell it what you would like it to do differently. To quote the Lieutenant's wise words when the computer says it doesn't "routinely scan" for a substance he's looking for: "Well this isn't a routine situation. Scan for the nylazine!"
Words to live by! Metaphorically, at least.
This approach helped Geordi find a solution to the bad brain pattern he was trying to cure. And speaking of bad brain patterns....
To be like Geordi, turn off your Google Brain.
Frustrated ChatGPT users, like the Enterprise Crew, make the mistake of treating ChatGPT like a conventional search engine. I call this the "Google Brain" approach. And it leads to disappointment.
Let's go to a less enlightened, distinctly non-Geordi moment in the YouTube TNG supercut video I showed you near the beginning of this article. In fact, just jump back to the beginning, and check out the opening scene with Data and Troi.
Frustrated ChatGPT users, like the Enterprise Crew, make the mistake of treating ChatGPT like a conventional search engine. I call this the "Google Brain" approach. And it leads to disappointment.
Let's go to a less enlightened, distinctly non-Geordi moment in the YouTube TNG supercut video I showed you near the beginning of this article. In fact, just jump back to the beginning, and check out the opening scene with Data and Troi.
This scene is from a fan-favorite episode of The Next Generation: Darmok (S5E02). In that episode, the crew encounters a race of aliens who speak in metaphors and riddles that absolutely confound the universal language translators that aliens use to understand each other.
I love how prescient the episode is—we have an entire budding industry of AI translation companies! What I don't love is the way Counselor Troi and Lieutenant Data are talking to the ship's computer. Troi simply tells the computer, "Search for the term 'Darmok' in all linguistic databases." When that gets Troi and Data nowhere, Data, in a facepalm moment, tries pretty much the exact same thing, saying, "Search for the term 'Tanagra'; all databases."
The ship's Counselor may have sighed and rolled her eyes at the computer's outputs, but Troi and Data have both got me sighing and rolling my eyes at their approach. Don't talk to an LLM as if it's a search engine! This is basic, 21st-century Prompt Engineering 101, and this pair of 24th-century professionals should know that!
Instead, they should say at the very least:
"We encountered an alien race that speaks in phrases that don't make sense to us. One of the aliens said 'Darmok at Tanagra' to us. Can you analyze the audio of the alien's speech for emotion and attitude and speculate at any possible meaning?"
Additional context about the situation could help as well.
Make Your Own Ship's Computer Fall in Love With You
It's not just AI consultants like me that notice Geordi's special ability to connect with the ship's computer. In fact, there is a popular fan theory that the ship's computer is in love with Lt. La Forge.
Like the ship's computer, ChatGPT doesn't actually have emotions or a will of its own. It is, after all, generative AI—not the artificial general intelligence that Data possesses.
But I'm going to let you in on an open secret: generative AI is programmed and reprogrammed daily by a paid army of real human trainers—and I should know, because I work as one of those trainers as a side-hustle to my consultancy!
Trainers like me program AI to respond best to natural, conversational inputs—the kind Geordi La Forge is fantastic at. So the more you talk to AI like it's a person, the more it behaves like a real, human thought partner. And that's what you want, because humans are smart. When they're not failing to recognize the difference between ChatGPT and Google, that is!
Comments