Ask the Next Question: from ChatGPT to Star Trek (via Davos, HAL & Cave Dots)

After a week of Davos panels, furrowed-brow magazine articles and breathless podcasts, I have come to think that ChatGPT could be an ancestor of the Star Trek computer. Yes, this early iteration tells lies, “hallucinates” narratives and conflates facts, all without batting a single byte. ChatGPT is a sociopathic toddler, oblivious to — and also innocent of—the random damage it leaves in its wake.

But who wouldn’t want a Star Trek computer to assure them that all is “operating within normal parameters?” Or that would routinely connect the dots at warp speed to save the crew/ship/planet/galaxy from certain annihilation as the clock counts down? Or that could brew up a perfect cup of “Earl Grey tea, hot” any time of the day or night, conjured from the void of a “food replicator”?

The Star Trek computer. Tony Stark’s marvelous J.A.R.V.I.S. TARS from the film Interstellar. And, of course, who can forget HAL, the unhinged, if utterly logical, computer of Stanley Kubrick’s 2001: A Space Odyssey? “Daisy, Dasisy, Give me your answer do…” Indeed.

One way or another, smarty pants AI is standard issue for the future.

These, of course, are all examples of “general intelligence,” which doesn’t yet exist. They have “brains” that work more like ours, only better and faster. For better. Or worse.

ChatGPT is more of an idiot-savant, capable of unintentional brilliance delivered with unshakeable confidence.

“…You can predict the next word somebody is going to say by having an internet’s worth of data… Then you have outlier cases. So the problem with the driverless car is that things come up that there’s no data to represent. What happens if you summon a car across a jetway?

Humans are not driven by massive amounts of data. We always need some data, but we’re trying to have abstractions. We try to have a few examples that will tell us a lot, where the neural net approach is to get as many data points as you can and hope that the new thing is going to be close to one of them. I think it’s a limited paradigm. It’s not really working for truth and reason…”

— Gary Marcus, emeritus professor of psychology and neural science, N.Y.U: from The Ezra Klein podcast.

••••••••••••••••••••••••••••

“Prompt engineering.” Those two words quickly achieved mantra status at the 2023 World Economic Forum conference in Davos where an early morning panel on “AI and White Collar Jobs,” drew an SRO crowd of freshly insecure executives.

It is one thing for automation to go after blue collar jobs. That’s about increasing productivity and fattening up bottom lines, which is pretty much the point of WEF. But “intelligent automation” presents an existential threat, painting targets on the backs of all the white shirts beneath all those white collars. “AI is coming for all our jobs, too,” noted moderator Ina Fried, chief technology reporter for Axios.

“Prompt engineering,” which is a fancy way of saying, “Ask better questions,” provides a role for humans in this quickly shifting labor equation. ChatGPT cannot yet “prompt” itself. At least for now.

Panelist Mihir Shukla is the co-founder and CEO of Automation, Anywhere, Inc, a company that creates algorithmic bots for knowledge work (e.g, processing invoices and mortgage applications). He pointed out that “anywhere from 15% to 70% of all the work we do in front of a computer can be automated.” Then added, “I don’t know anybody who wakes up in the morning and says, ‘My mission is to process invoices.’”

Shukla spoke at length about how a mortgage application can now be processed in a matter of minutes, instead of the standard 30 days, opening the door to massive growth. He envisions “human / bot” partnerships with “digital co-workers.”

“‘You (ed. note, meaning the digital co-worker) process while I do something else. When the answer comes back, here’s the next set of work, while I do something else.’ That’s the future of work.”

What the “something else” human part the job might entail was left fuzzy. Could it, too, soon be fodder for intelligent automation? Will these new positions pay as well? And who will be left to apply for all those speedy mortgages as middle class and even upper middle class jobs evaporate?

“Reskilling!” became the second conference mantra.

But reskill for what?

Not media planning and buying, according to Sir Martin Sorrel, the serial advertising entrepreneur who founded WPP, a massive global holding company that owns ad agencies, marketing companies and public relations firms. Sorrel’s latest venture, S4 Capital, focuses on digital marketing.

“You will not be dependent as a client on a 25-year old media planner or a buyer, who has limited experience. You’ll be able to pool the data.”

His predictions for the creative side of the business are just as disruptive. The combination of image generators such as Dall-E, along with AI-boosted presentation deck generators (e.g. DesignerBot) and, of course, ChatGPT, put entire creative teams in the cross-hairs. Low-cost, “good enough” work will soon flood the market.

Which leaves humans with little to do other than find a way make a career out of being consumers.

••••••••••••••••••••••••••••••••••

That’s an awful future. So let’s try looking to the past: 20,000 years in the past. According to a study recently published in the Cambridge Archeology Journal, researchers may have finally solved the riddle of patterns of dots found on ancient cave paintings of animals. They propose that the dots were a notation system for documenting the breeding cycles of prey animals.

It is a conclusion not without critics, but for now let’s say it’s true. Once that system — that technology — existed, not only did it have valuable predictive value, it saved time and also freed up brain space to ask the next question: When is the best time to organize a hunt?

Every technology, from the wheel to the pen to the smartphone, extends human capabilities. We tend to think of technologies in terms of hardware, and more recently software, but it is also storytelling. In fact, storytelling, through words, dance, sounds, sculpture and yes, cave paintings, is the original technology.

ChatGPT is like the dots and lines on the cave walls: a way to get to next question faster.

•••••••••••••••••••••••••••••••••••

It took a high school intern working in the lab of a scientist I know about a nanosecond to figure out that ChatGPT is a fabulous tool when coupled with good “prompt engineering.” When my friend asked his intern about made up facts, the intern replied that, of course, you have to double check. But if you insert some facts into the prompt, it’s pretty good. Voilà. Instant essay.

Back at Davos, Ina Fried told the crowd about a professor who asks his students to have ChatGPT write their essays. Their assignment is to write about what ChatGPT got wrong.

Brilliant. That’s critical thinking.

The dangers of a sociopathic toddler are not to be underestimated: Fake news at scale, big lies on turbocharge. But the path to the Star Trek was never going to be easy.

So, with a hat tip to sci-fi writer Theodore Sturgeon: Who has the next (prompt engineered) question? The future of ChatGPT’s descendent, the AI of the Starship Enterprise, depends on it.

In the meantime, there is Promptbase, a marketplace where for as little as $1.99, one can buy “prompts.” It is clever side-hustle, at least until, inevitably, there will be a bot for that, too.