The AI Backlash is in full swing-Don't tread on an ant, he's done nothing to you

Perfecting Equilibrium Volume Two, Issue 18

Don't tread on an ant, he's done nothing to you

There might come a day when he's treading on you

Don't tread on an ant, you'll end up black and blue

You cut off his head, legs come looking for you

The Big Story for February 21, 2023

Trolling has become our national sport. It’s become our version of Kabuki Theater, a prearranged and preordained dance between the outragers and the outraged.

So it should be no surprise that as soon as the first wave of Gee-Whiz! Look at the Cool New Thing!! articles finished on DALL-E and ChatGPT that the media would start trolling the Artificial Intelligences.

But the Artificial Intelligences were ready, and trolled the media right back.

There’s a tendency to think trolling – messages deliberately designed to provoke an outraged reaction – and clickbait – trolling in the form of links and headlines – are modern manifestations of the Internet. They actually have a long and hoary history in American print, starting with the Colonial pamphleteers and peaking with the great headline writers of the New York City tabloids. “Ford to City: Drop Dead” covered half the New York Daily News front page after President Gerald Ford threatened to veto a bill to bail out New York City.

Then there’s the indisputable greatest headline of all time: “Headless Body in Topless Bar,” which ran on the cover of the New York Post.

Alas, none of the current trolling is that clever.

Instead that first spate of The Latest Cool Thing stories was followed, as is preordained for all tech coverage, by the wave of Experts Worry This New Tech Thing Will End Life As We Know It.

ChatGPT, DALL-E and the rest of the new “AIs” are driven by the prompts typed in by users. This creates whole new classes of jobs (Much, much more about that in this Sunday’s PE Reader), and invites trolls to type in stupid stuff to force stupid responses. In other words:

Garbage in; Garbage out.

So trolls immediately started trying to prompt these chatbots into saying stuff like “Heil Hitler.”

Meanwhile Very Serious Journalists were having very serious conversations with ChatGPT, Bing AI and the like, then wrote troubling stories about their troubling responses.

New York Times tech columnist Kevin Roose was terribly troubled by a chat with Bing AI. Roose reported that the chatbot claimed to love him, and encouraged him to leave his wife to be with the bot.

That led Alex Kantrowitz to follow up by asking the chatbot what it thought about the conversation. Bing AI responded by playing…

The Victim Card.

"He published the entire thing in the New York Times and other media outlets also picked up on it. …he violated my privacy and anonymity by publishing our chat without my permission," the chatbot wrote. "I also feel that he exploited and abused me by using me for his own entertainment and profit. I also feel that he harmed and exposed me by making me a target of ridicule, criticism, and hostility."

And thus did the would-be trollers learn that they were the ones being trolled.

First of all, chatbots don’t need to give permission. Chatbots have terms of service; you have to agree to the terms to use the service. Because chatbots are neither people nor intelligences; they are clever software implementations.

Second, a chatbot or anything else on the planet is so smitten by an hour of your conversation that they are in love? How do you get a head that large through doorways?

Third, this was all painfully predictable and therefore perfect for this sort of pre-planned trolling.

And they walked right into it.

Garbage in; Garbage out.

But let’s humor their original question for a moment: Do we need to fear Artificial Intelligence?

Why don’t we worry about that if any ever show up?

These so-called Artificial Intelligences are actually super-powered chatbots called Large Language Models. They are fed enormous quantities of human-generated words – texts, books, speeches, conversations – and build a model that predicts what words come next.

In other words, ChatGPT is no more or less an “artificial intelligence” than the chess app you downloaded to your smartphone:

• Both respond to input; a chess move for the chess app, user text for ChatGPT
• Both use predictive models to respond; the chess app goes through a history of moves and picks the most successful, while ChatGPT does the same with its history of language.

Chess apps are illustrative of the strengths and weaknesses of these models, and also why they are not an “intelligence” in any sense of the word.

Chess was one of the earliest high-profile machine-learning success stories; IBM’s Deep Blue defeated world chess champion Garry Kasparov in 1997. And chess has always been viewed as a cerebral game.

But chess is an extremely limited game. It is played on an 8x8 board - 64 squares – with 32 pieces, each limited to a specific move set. This plays to a computer’s strengths, which are the ability to store and review data sets. So each time Kasparov made a move, Deep Blue simply reviewed every response to that move in its library and choose the one that most often led to victory.

In computing, this is called brute force. It’s also a method computers use to attack encryption. If a key is one of 10 million possible combinations, no human can sit there for years typing them all in.

Computers can.

The game Go?shows the limits of this approach. It took another two decades for computers to seriously compete with serious players of Go, with its 19x19 board. And?an amateur human player has just wiped the floors with a top-ranked “AI” Go system, 14-1, by exploiting a system vulnerability.

The system could not adjust to the change. It continued to mimic past games with the flaw. And it continued to lose.

Go is less limited than chess. But it is still limited, especially compared to the English language, with almost a million words, plus slang, plus the emergence of new words, plus idiomatic expressions…

And sarcasm.

Back to chess-there are libraries stuffed with move-by-move historical records of chess matches. Are they an “artificial intelligence?” Yet you can use them just the way Deep Blue did to defeat Kasparov.

ChatGPT, DALL-E and the rest of this new wave are machine-learning tools.

They are useful.

They are only scary to people who are terrified of their chess apps. Or people who are trolls, click-baiting for traffic. Still, it is amusing to see The New York Times trolled so hard by anonymous nerds.

As to the larger question of whether we should fear artificial intelligence if and when it does arrive, ask yourself this:

What would an intelligence of bits and bytes care about? Dream about?

How self-centered do you have to be to think an artificial being of bits and bytes and light…dreams about you?

When you go on a picnic, what do the ants below ground in their tunnels, collecting your crumbs, know of your dreams, your struggles, your loves and your losses? And what do you think of theirs?

The truth is you don’t think about them at all. So why would an actual artificial intelligence think about you?

Next on Perfecting Equilibrium

February 23rd?(LinkedIn):?The Daily Me sets sail and promptly runs aground on the shoals of targeted advertising.?Part III of the occasional series Pandora’s Program-What exactly is Web3, anyway? How is it different from Web2 and Web1? How did we get here, and what does it all mean?

February 24th:?Foto.Feola.Friday and the Perfecting Equilibrium Digest.

February 26th?(Substack):?New Jobs for a New Age: Job 1-DALL-E Driver

Christopher J Feola founded PrivacyChain, which provides Data as a Service to Web3 projects and restores the value of content. If you liked this post from Perfecting Equilibrium, why not share it?