Some of my early work with voice technology was as an assistant ADR editor on the film Galaxy Quest (1999). ADR stands for automated dialogue replacement, which is essentially recording additional voiceovers during post production. Working on the film was a high-note as I transitioned out of Hollywood and into the burgeoning field of experience design. Still, despite its well-deserved cult following, I don’t think about Galaxy Quest all that often. That changed when ChatGPT revealed to the world at large just how powerful a conversational interface can be.
I’ve been working with conversational AI for more than 20 years. In many ways, ChatGPT is something I’ve been waiting for a long time. It’s delivered a light-bulb moment where people everywhere are realizing how easy it can be to interact with machines using a conversational interface. This really is an inflection point in our relationship with technology. (Chat GPT amassed over a million users in its first five days, and it’s only grown from there). Why not take a moment to learn a few lessons about how to properly leverage generative AI from a screwball sci-fi classic.
1. Like Thermians, GPT Has a Severely Limited Understanding of the World
In Galaxy Quest, a group of actors from a syndicated series that’s a lot like Star Trek encounter an alien race with a deep connection to their work. The Thermians have been streaming television content from Earth for years, believing that it’s depicting real events. They’ve modeled their entire society around the show, but are unable to contextualize what they’ve seen.
“Since we first received transmission of your historical documents, we have studied every facet of your mission and strategies,” the Thermian leader explains to the fictional crew once they’ve been brought on board a fully functional recreation of their ship, the Protector. It quickly becomes evident, however, that Thermians are unable to understand human emotion and the contextual clues that other humans pick up on right away. They are also totally unaware that the source of all their inspiration is a work of fiction.
GPT is similar. It’s essentially been trained on all the knowledge available on the Internet, but it doesn’t really know anything. It’s a highly predictive best-guess machine, and any grasp it seems to have on the staggering amount of information it’s been fed is illusory at best. While plenty of users have unearthed novel uses for ChatGPT, its primary strength is as a conversational interface. It becomes infinitely more useful as a portal into an ecosystem where other technologies and data are being sequenced to automate sophisticated processes. Without that orchestration layer, ChatGPT, like Thermians, has a limited knowledge base.
2. Inefficiency and Bias Always Bubble Up, Even in Outer Space
The Protector is a state-of-the-art ship with a voice-activated computer system. Unfortunately, the system only responds to commands from Lt. Tawny Madison (Sigourney Weaver). The irony is thick. The inherent strength in conversational AI is that it enables anyone to communicate with technology, without any special training. Setting up the interface so that only one crew member can use it is almost surgically poor design.
As a result, Madison is forced to repeat everything the computer says to Commander Taggart (Tim Allen) and then repeat his response back to the computer. The reason it functions this way is so that an attractive female character has something to do. “Look, I have one job on this lousy ship,” Madison yells when one of the crew complains about her parroting behavior. “It’s stupid, but I’m going to do it!”
It’s a clever jab at the narrow ways women have been depicted in media for centuries, and points to the kinds of entrenched bias that could be extremely destructive inside of technology this powerful and pervasive.
As Davey Alba pointed out recently in Bloomberg, “like all AI products, [ChatGPT] has the potential to learn biases of the people training it and the potential to spit out some sexist, racist and otherwise offensive stuff … [OpenAI] has attempted to bake-in guardrails that “decline inappropriate requests” that have befallen similar programs run by artificial intelligence. It won’t, for example, offer up any merits to Nazi ideology, if asked.”
These are smart steps in the right direction, but we can’t put the onus on OpenAI to fix the societal biases and horrors that AI reflects back to us. There’s also the troubling news reported recently in Time that OpenAI’s efforts to detoxify their GPT technology were outsourced to a firm that paid Kenyan laborers less than $2 per hour. If AI is going to reach its potential as a great ally to all of humanity, we’ll all have to do better.
Other disruptive large language models (LLMs) are also bound to pop up in the coming months and years. At the same time, large numbers of individuals are going to unearth opportunities to train generative AI in novel ways. With conversational AI becoming a regular player in all of our lives moving forward, we need to find ways to work together to strip bias from our systems. That can take the form of end users reporting troubling behavior as well as fostering more diversity among the ranks of developers and programmers who are shaping this tech.
To borrow a catchphrase from the Spock-esque Dr. Lazarus (Alan Rickman): By Grabthar’s Hammer, by the Sons Of Warvan, uh, the time for this difficult work is now!
3. Outcomes Improve Once Humans Are Involved
The Thermains have recreated the Protector down to the most minute detail, but they can’t use it on their own. They need human help with that. When the crew is attempting to beam Taggart off of a hostile planet using the ship’s “digital conveyor,” a wrinkle appears. One of the Thermains reveals that “theoretically, the mechanism is fully operational, however, it was built to accommodate your anatomy, not ours.” Thus, Tech Sergeant Chen (Tony Shalhoub) has to take the reins of the dangerous and temperamental device to save his captain.
In a similar fashion, the scheme to defeat Sarris only unfurls once the crew puts aside their differences and works together as a team. Their ability to use creativity to solve problems is the secret sauce. All of the ship’s impressive technical abilities are useless without humans at the helm.
The same holds true for AI, which frankly can’t create its way out of a wet paper bag. Without human input, generative AI in particular has nothing to generate. This is a good thing. The key to solidifying AI as an ally (and not some dark overlord) is to always keep humans in the driver’s seat (or captain’s chair).
ChatGPT is an extraordinarily powerful interface, but it needs humans to tell it what to do. Like all AI, it needs humans to monitor its activity and make sure it isn’t heading into dangerous areas. It needs humans to act on the choices it presents.
Never give up! Never surrender!
Surprisingly, the public reaction to ChatGPT has been far more frenzied and colorful than our reaction to the U.S. military confirming the actual existence of hundreds of unidentified aerial phenomena (or UFOs, colloquially). It seems like the aliens have landed, but they’re not aliens.
Our unexpected visitor is a large language model that slurped up the internet and now looks back at us with gigantic eyes and no guile. Metaphorically, those eyes might as well be massive and almond-shaped, that head gigantic and teetering on a slender neck. We’re communing with an abstract reflection of humanity swollen with knowledge but lacking in understanding. It needs our help making forward progress.
The next logical step is putting ChatGPT to use as a front end to ecosystems built for creating and evolving elaborate process automation. This is how we pilot the mind-blowing spaceship our real-world Thermains have laid at our feet. There’s plenty of fear and uncertainty in the air, but don’t let it cloud a generational opportunity to push humanity into a new era. To quote Commander Taggart, “Never give up! Never surrender!”