Skip to content

Things are Changing Very Fast

2022-12-09 · letter vlog · Hank · 3:59 · watch on YouTube

Topic: Internet & YouTube · mode: Personal essay · tags: timely hook

Following John's AI video, Hank says new technologies push people into camps and it will take years to know even the questions (such as whether AI image models plagiarize); his big worry is that a new AI wrench arrives before society has absorbed the last communication revolution, and that AI will be good at making people fear and dislike each other.

What is said

The parts of the video and what is said in each (★ — how much it matters to the video).

00:00:00 — Worried about AI too; new things send people to their camps, and groups form around AI now

  • ★ Hank says he too has been worried about AI this week (week of 5 December 2022), which is not weird since they are actual intelligences anchored to different interests and worries. — 00:00:00
  • ★★★ Hank says he wrote two books about how, when new things arrive (a pandemic, social change, a powerful technology, an alien), people retreat to the camps where they feel safest, the more so the more confused they are. — 00:00:00 · An Absolutely Remarkable Thing, A Beautifully Foolish Endeavor
  • ★★ Hank says each camp asks questions about its own concerns, the powerful camps dominate the conversation, and it all ends in two internally divided groups butting heads. — 00:00:29
  • ★★★ Hank says groups around AI are forming in real time now, and it will probably be five or ten years before we know what questions we are asking and what the dominant perspectives are. (hedged) — 00:00:29

00:00:58 — How AI models learn and sometimes copy: Midjourney and the National Geographic image; open legal and ethical questions

  • ★★ Hank says he doesn't know what those perspectives will be, what questions need asking, or their answers. (unsure) — 00:00:58
  • ★★ Hank explains that AI models are not meant to copy: they learn how many texts or pictures are structured and output things with that structure, but their procedures sometimes make them copy. — 00:00:58
  • ★★ Hank asked Midjourney to imagine an Afghan woman with green eyes and shows what it gave him. — 00:00:58
  • ★★★ Hank says Midjourney's model does not contain the famous 1980s National Geographic cover image but was trained on it, so its output plagiarizes it. — 00:01:27
  • ★★★ Hank lists open questions: can you sue an AI for this, is training on copyrighted works without consent okay, should artists opt out, is it illegal or a call for new law, who will it help or hurt; he doesn't know. (unsure) — 00:01:27
  • ★ Hank says people are going to disagree about all this like crazy. — 00:01:27

00:01:56 — Big concern: society has not absorbed the last communication revolution; ChatGPT is bad at facts but good at making people dislike each other; sign-off

  • ★★★ Hank's big concern: we have not yet got through the last revolutionary shift in human communication and are still figuring out how to live with disruptive tools given to us 10 or 20 years ago. — 00:01:56 · social media
  • ★★★ Hank says we are just realizing we live much of our lives in places that are not democracies, where we get no vote and leadership can change with whoever has the money to buy them. — 00:01:56 · social media
  • ★★ Hank notes many of these platforms have never been through a recession and have grown on lots of money; now at a ceiling, nobody knows what they will do to break through. — 00:02:26
  • ★★★ Hank says it now feels like another weird, entirely new wrench is about to be thrown into human communication. — 00:02:26
  • ★★★ Hank says ChatGPT is not good at replacing Wikipedia: it confidently states true and untrue things alike, and only fact-checking tells them apart. — 00:02:55 · Wikipedia
  • ★★★ Hank thinks such AI will be great at easy problems, like getting people to dislike each other more. — 00:02:55
  • ★★ Hank says many people, external and internal, want people more afraid of and hostile to each other; much get-out-the-vote work is about fear and anger. — 00:02:55
  • ★★★ Hank says making people afraid and angry at each other is a trivial problem for AI: it only has to emulate a human at their worst, and at our worst we are not complex. — 00:03:24
  • ★★ Hank says fast change in people and societies is usually not fun; he wouldn't mind things getting less interesting for a while, but we don't get to choose. — 00:03:49

In the description and comments

  • ★★ The description disagrees with a TikTok argument that we have a small window to decide what to do with AI: there is no window (the genie is out) and also one open for a very long time. — video description · TikTok
  • ★★★ The description says we learn new technologies' powers and dangers slowly, then develop tools, mostly taboos, for their use; taboos cannot be formed before the negative outcomes are experienced. — video description
  • ★★ The description bets there will be many negative outcomes and that systems to handle them may take a generation or two, as with the internet. (hedged) — video description
  • ★ In a channel comment Hank says he, not ChatGPT, wrote this video, and admits he is also a little stressed. — video description

Threads in this video

Where the stories that moved here went next

Mentioned

Central: artificial intelligence

Also: Works: A Beautifully Foolish Endeavor · Afghan Girl · An Absolutely Remarkable Thing

Organizations: National Geographic · Wikipedia

Concepts: copyright · Plagiarism · recession · social media · tribalism

Products: ChatGPT · Midjourney

Only in the description or credits: TikTok

Also centred on artificial intelligence: Google in your Brain · Question Tuesday Except I'm an AI · Asking an AI Image Generator for Pizza Johns · AI tells me if AI will write novels. · After 12 hours in LA I'm a stand-up comic doing ASMR with a side gig as a music promoter · The Coming AI Invasion

Replies to

← Previous: AI tells me if AI will write novels. · Next → What I Would Have Tweeted