Memories: Artificial Intelligence at Stanford in the 70s | Hacker News Reader

Memories: Artificial Intelligence at Stanford in the 70s

lawrencecpaulson.github.io

72 pointsfurcyd14 comments

From the Shameless Self-Promotion Dep't:

My book https://www.albertcory.io/inventing-the-future has a chapter set in 1979 or so, at the Dutch Goose, where AI Winter 1.0 is fully underway:

===== So, Grant brought up one more thing before they left the topic of work. “What do you guys think about artificial intelligence? Is that ever going to go anywhere?”

Porter chuckled, “There’s a paper from a few years back called ‘Artificial Intelligence Meets Natural Stupidity.’ You might want to read that.” The other guys laughed.

Patrick announced, “I’ll never forget his opening line, ‘As a field, artificial intelligence has always been on the border of respectability, and therefore on the border of crackpottery.’”

Porter sat back with a wide grin, “I think they crossed that border several years ago. That’s why they can’t get any more funding.”

Ray said, “Hey, the big breakthrough is only 10 years away. And always will be!”

Grant had the strong impression from their approving looks that they weren’t impressed with AI. He wasn’t ready to give up quite yet, “Didn’t they do some cool stuff, like Blocks World? I loved how you could say ‘pick up a big red block,’ and it did it.”

Ray interrupted, “Winograd’s around all the time. You could talk to him.”

> Early robots were put through their paces.

Including, if the rumours are to believed, early exercises in teledildonics.

(search engines are not confirming at the moment; it would've been late-60's early-70's and the way I heard the story, the geeks at SAIL were shocked —but not that shocked— to discover that two coeds who answered an ad seeking "open minded individuals" turned out to be even freer of hangups than had been hoped... Or have my aging neurons misattributed these antics from a different AI lab?)

The "farewell address" from the WAITS machine included a reference to the event.

WAITS has the true details; good to see my memory was not any more unreliable than usual.

> The first character alphabet that was programmed for my plotter was Elvish rather than Latin.

To be fair, tengwar is so much more regular than latin, that it only stands to reason one might wish to debug with it, parametrically* generating letterforms, before committing the effort to digitising a full latin alphabet, letter by painstaking letter.

(in our days of cheap bandwidth, it's even easier to download preexisting alphabets; eg https://emergent.unpythonic.net/software/hershey )

* cf https://i.imgflip.com/7757j9.jpg

Before I look, do I want this in my search history?

MIT's PDP-10 OS, ITS, was unrelated to TOPS-10 and in fact predated it (it was the first OS on the PDP-6). AFAIK, SAIL's PDP-10 OS, WAITS, was also written from scratch, as it started on the '6 as well. I DOUBT DEC every provided a PDP-6 OS anyway as it was a research machine manufactured in small volume mainly designed for LISP.

ITS's name, Incompatible Timeshare System, came from a prior MIT OS, CTSS ("Compatible Time Sharing System"). I never learned what it might have been compatible with as it was long gone by the time I showed up at MIT.

WAITS supposedly got its name as a joke on ITS, but you'd have to ask Les Earnest for the real story.

IIRC DEC provided a very, very limited OS for PDP-6 called "MONITOR", which later survived long enough to be used as origin point for TOPS-10. The version as arrived with PDP-6 was very, very limited, which didn't help PDP-6's abysmal sales (especially coupled with infamaous unreliability).

...and then came the first AI winter. Which will certainly happen again, this time, only a few years later ;-)

Why do you think there'll be an AI winter? And in what form — stagnation of neural-network based technologies, a change in the overall paradigm of learning-from-data, or something else altogether?

Because the realization that "no, an AI is not artificial life" will hit, everyone will realize that all an AI does is running a function with millions of unknown state variables, and that this is not helpful when you're tyring to understand why a network does (or doesn't) do something.

Sure, the applications we have (and then some) will stay, for cases where using AI brings real benefits (just like after the first AI winter).

The next AI winter will be caused by very needy physical robots walking around the world acting like little kids, trying to train their models.

They will ask "what are you doing?" and then the dreaded "why?" about everything. They will annoy the crap out of everyone to the point where we just pack the whole thing up for twenty years.

“Why?” To which any Animaniacs fan will respond, “OK, love you, bye-bye!”.

Bullshit. People have wrongly predicted an AI winter for the last decade. It didn't happen in 2018, it didn't happen in 2022, and it sure as hell won't happen in 2026.

I'm hoping the current crop of ML/AI yields sufficiently commercially viable blindsight behavior [1] to fund continuing research into AGI. Generative LLM seems sufficiently sophisticated to write ad copy and general political speech of all kinds, so if that's the direction we go I can see all kinds of interesting possible implications. I'd like to see baking in positive optimism and collaboration into the DNA of solutions delivered by such blindsight systems.

[1] https://en.wikipedia.org/wiki/Blindsight_(Watts_novel)#Major...

Reply on news.ycombinator.com