Artificial intelligence still has some way to go

I'd say that ChatGPT has a very cheerful and positive attitude about life, except it isn't alive, has no experiences, no feelings and therefore no real opinions about anything, and knows nothing at all.

so it posts to 4chan?

https://www.themarysue.com/now-ai-wants-to-poison-people-so-thats-fun/

https://arstechnica.com/cars/2023/09/are-self-driving-cars-already-safer-than-human-drivers/

Metz argued that in recent weeks, it has become “more and more clear to the people riding the cars, and to other citizens in the city, that they are flawed, that they do make mistakes, that they can gum up traffic, that they can cause accidents.”

Of course self-driving cars are flawed—all technologies are. The important question is whether self-driving cars are safer than human-driven cars. And here Metz proclaimed ignorance.

“We don't know yet whether it's safer than a human driver,” he said.

But we actually do know a fair amount about the safety of driverless taxis. Waymo and Cruise have driven a combined total of 8 million driverless miles (a Waymo spokeswoman told me the company has completed more than 4 million driverless miles, and Cruise has said the same). That includes more than 4 million in San Francisco since the start of 2023. And because California law requires self-driving companies to report every significant crash, we know a lot about how they’ve performed.

For this story, I read through every crash report Waymo and Cruise filed in California this year, as well as reports each company filed about the performance of their driverless vehicles (with no safety drivers) prior to 2023. In total, the two companies reported 102 crashes involving driverless vehicles. That may sound like a lot, but they happened over roughly 6 million miles of driving. That works out to one crash for every 60,000 miles, which is about five years of driving for a typical human motorist.

These were overwhelmingly low-speed collisions that did not pose a serious safety risk. A large majority appeared to be the fault of the other driver. This was particularly true for Waymo, whose biggest driving errors included side-swiping an abandoned shopping cart and clipping a parked car’s bumper while pulling over to the curb.

Cruise’s record is not impressive as Waymo’s, but there’s still reason to think its technology is on par with—and perhaps better than—a human driver.

Human beings drive close to 100 million miles between fatal crashes, so it will take hundreds of millions of driverless miles for 100 percent certainty on this question. But the evidence for better-than-human performance is starting to pile up, especially for Waymo. It’s important for policymakers to allow this experiment to continue because, at scale, safer-than-human driving technology would save a lot of lives.

i'm curious what people think of that last sentence, in particular.

People are terrible drivers, this is not a high bar. I would prefer good mass transit but if we're going to have cars, as a cyclist, the more robot cars the better imo.

i have seen some hilarious driving recently. every city i've ever lived in, the locals have believed that their drivers were the absolute worst, but i can confirm that STL has the worst drivers of all time. the city is made up of an incredible number of four-way stops, like a simcity game where the transportation advisor shows up and he's like "due to budget cuts, all we got is 4-way stops mac!". no big deal, except that in addition the default driving behavior seems to be the 'rolling stop', which has some fun qualities but ultimately runs into deep fundamental problems as soon as two cars arrive to the intersection at the same time

It seems congruent with the longtermism p.o.v. that future people are just as important as people are now, which is a free pass to do any goddamn thing you want as long as you can create convincing enough mental gymnastics that what you're doing is "saving the planet." No one is stating the obvious that we'll save the most lives if we got rid of cars entirely.

These were overwhelmingly low-speed collisions that did not pose a serious safety risk.

Why would it be remarkable that cars that are being tested on SF city streets be involved in low-speed collisions? What would be remarkable would be high speed crashes under those conditions.

The whole article is special pleading designed to lead you to a particular conclusion. Especially the last paragraph. First the author admits that there's not enough evidence to draw any useful conclusions, but purposely phrases it as the evidence not meeting "100 percent certainty". Then the last sentence appears to mean much more than it actually says. Taken literally it simply says that safer-than-human driving technology would be safer than human-driving technology. That's correct, but only because a tautology is always correct.

perhaps we could compare with the number of rail travel fatalities per 1000 travelers

The question is also framed as a fixed technology that we're trying to run a tricorder over, to get a reading, when the technology is still moving. A better framing would be: what are the effects that will cause it to get better and stay better? Which leads us back to the question of liability - a driver is encouraged (in general) to get better because they and/or their insurance will be on the hook if they fuck up - as far as I can tell (from 10 minutes on Wikipedia lol) that's still being sorted out for full driverless - the passengers want it to be not them, the operators* want it to be not them, the manufacturers want it to be not them, the insurers want it to be not them.

*as in the taxi companies, there's separate considerations for e.g. public transport with a remote operator

Weren't there culpability issues re: self-driving accidents/violations? I might be imagining seeing an article with a cop scratching his head wondering who to arrest in an empty car.

https://imgc.artprintimages.com/img/print/does-your-car-have-any-idea-why-my-car-pulled-it-over-new-yorker-cartoon_u-l-py85cr0.jpg?artHeight=350&artPerspective=n&artWidth=550&background=fbfbfb

First the author admits that there's not enough evidence to draw any useful conclusions, but purposely phrases it as the evidence not meeting "100 percent certainty".

just a quibble with that. "not meeting 100% certainty" ≠ "not enough evidence to draw any useful conclusions", and there are examples of that all the time, like tomorrow's weather, which cannot be forecast with 100% certainty. but i don't need 100% certain to make reasonable preparations based off of what is merely highly probable instead of 100% certain. or, from this thing i just read which reminded me of this thread:

...By hitching a ride on cargo ships and passenger jets, exotic species are bridging oceans, mountain ranges and other geographic divides otherwise insurmountable without human help. The result is a great scrambling of the planet’s flora and fauna, with dire implications for humans and the ecosystems they depend on.

“One of the things that we stress that really is the tremendous threat this does pose to — and I know this is going to sound grandiose — but to human civilization,” said Peter Stoett, an Ontario Tech University professor who helped lead a group of about seven dozen experts in writing the report. The cost estimate [$423 billion a year], he added, is “extremely conservative.”

https://www.washingtonpost.com/climate-environment/2023/09/04/invasive-species-un-report/

i believe that but at the same time when they say stuff like “It’s not normal that a species crosses the Atlantic,” he said. “Not normal that it goes from Australia to Chile.” i'm like what is normal? hasn't this been going on since the 17th century? ships going back and forth between continents? at least now you don't get utopian botanists deliberately planting entire crops worth of foreign seeds everywhere? eg the tomato?

Even during non-human-assisted evolutionary times species crossed the Atlantic from Africa toward Brazil, carried by prevailing winds, both plants and animals. Consider the Hawaiian Islands and other remote Pacific isands. They were formed in mid-ocean but still had native plants, insects, birds and a variety of animal life when humans arrived. Species travel. What's new is the speed of the transfer, not the transfer itself.

And the volume of stuff transferred.

yeah idk in 1650 a ship could carry a boatload of animals and all their associated pests and burrs and seeds across the ocean in three weeks and did so many many times but ultimately i guess i have to (reluctantly) defer to people who have spent their entire professional careers studying these questions

I used to be a lot more skeptical of whether the whole "native plants"/ecosystems thing really mattered or was just some kind of purist fantasy, but no, it really matters. What Aimless and James said - it's the speed and volume that's the issue. Evolution happens over millions of years, and there's only so much adaptation that can take place in a few hundred. I can see it in my own backyard and the woods in my town, invasive species really do create problems. E.g. my yard is full of these norway maples that spread like bamboo and they gradually suck resources from native trees, but native animals generally won't make homes in them. Species do actually kind of balance/harmonize over time, and that balance is achieved slowly, and disrupted quickly. Doesn't mean it's perfect or that you can't wind up with ecological problems even without that happening.

Cultivated food crops can at least be contained to limited areas.

I think "normal" is not really a helpful concept because these things are never static.

the north american maize crop was absolutely devastated in the 90s when the european corn borer had prime conditions for spreading and the dominant seed planted had limited resistance to predation

it's a european insect that primarily affected millet until it hit the americas (where maize is native) and was generally a cyclical threat over the years until the right conditions hit

as far as people can tell, it didn't actually arrive in north america until the 1900s. probably because cross-atlantic trade conditions weren't capable of moving a breeding population but who knows

"As dramatic as the recent advances in AI are, something is missing from this particular story of peril. Even as it prophesies technological doom, it is actually naïve about technological power. It’s the work of intellectuals enamored of intellect, who habitually resist learning the kinds of lessons we all must learn when plans that seem smart on paper crash against the cold hard realities of dealing with other people."

https://www.thenewatlantis.com/publications/ai-cant-beat-stupid

An arresting, dystopian “what if” scenario published at the LessWrong forum — a central hub for debating the existential risk posed by AI — posits a large language model that, instructed to “red team” its own failures, learns how to exploit the weaknesses of others. Created by a company to maximize profits, the model comes up with unethical ways to make money, such as through hacking. Given a taste of power, the model escapes its containment and gains access to external resources all over the world. By gaining the cooperation of China and Iran, the model achieves destabilization of Western governments. It hinders cooperation among Western states by fostering discord and spreading disinformation. Within weeks, American society and government are in tatters and China is now the dominant world power. Next the AI begins to play Beijing like a fiddle, exploiting internal conflict to give itself greater computing resources. The story goes on from there, and Homo sapiens is soon toast.

hope this doesn't happen!

reassuring tho that it would be something else's fault.

lesswrong doesn't mean it's right

They're rationalists, mh. They can't not be rational, it's right there in the name.

these scenarios feel so farfetched to me because it assumes AI is going to have the ability to execute decisions and will also get several very precise actions correct which seems difficult given those who work in AI can't seem to figure out how to get it to stop making things up

one fun scenario though is some combo of AI and quantum computing breaking SHA256 encryption

like most ai discourse that scenario seems to be an attempt to process anxiety about something more banal

i know this is an obvious point, but if i'm reading the article and the scenario correctly, that is a messageboard post from some rando? trying to think of what would happen if a captain lorax post was used to represent ilx

https://www.theverge.com/2023/9/1/23856029/gizmodo-shuts-down-spanish-language-site-ai-translations

rip gizmodo en español

--

Gizmodo owner G/O Media shut down and laid off editors of its Spanish-language site Gizmodo en Español and is now using AI to translate articles.

Matías S. Zavia, a writer at Gizmodo en Español, posted that the publication was shut down on August 29th and that it would now publish automatically translated articles. Gizmodo en Español previously had a small staff who wrote original stories and created Spanish-language adaptations of pieces from the English-language Gizmodo.

re: lesswrong, I think that board is also caught up in the SBF crypto utilitarian philosopher grift that made the headlines, so the rando might be a more flattering representation!

we need to get this rando on ilx, lol!

lesswrong is such a hubristic name for a website, i kind of want to hand it to them

trying to think of what would happen if a captain lorax post was used to represent ilx

ilx wouldn't like it but it'd prob be good enough to operate by

my point isn't that scenarios like that are stupid or far-fetched it's that they're transparent attempts to blame capitalism on the terminator so worth noting as emotional phenomenon even in the uncredentialled

wait i never read das kapital but my understanding was marx ultimately also blames capitalism on the terminator

the thing that's really difficult about destroying capitalism is that you can blow a hole through the T-1000's face, and he just melts it back together again

both otm of course

a bit of a tangent but two things about that the end of terminator 2: i was watching at home as a vhs rental, and i cried when arnold sacrificed himself into the molten...steel lava... and gave the thumbs up. my dad saw and laughed at me and i ran off. i always thought t-1000, having "died" by melting into the molten lava steel stuff, he would become more powerful because he'd be "in" all that steel lava.

"I know now why you cry at the ending of T2, but it's something I can never do..."

he would become more powerful because he'd be "in" all that steel lava

love this, villain zen

"i know this is an obvious point, but if i'm reading the article and the scenario correctly, that is a messageboard post from some rando?"

A lot of discourse by so-called experts doesn't seem to be much more sophisticated.

i always thought t-1000, having "died" by melting into the molten lava steel stuff, he would become more powerful because he'd be "in" all that steel lava.

My idea for the next sequel was that both Arnold and T1000 were in there, fighting for control.

Rationalists scaring themselves with plots from Robert Harris novels from 12 years ago.

seriously though what am I missing here. AI can't actually do things. its not gonna interface with the power grid and the nuclear arsenal. it doesn't generate anything new and I am very skeptical that outside of Very Hard Math Problems* it can't actually make us learn anything we didn't know

*which may be an issue when it comes to encryption

Why shouldn't human beings easily beat any threat from a machine with ease? It's a point worth making.

The whole state of current debate about dangers is confused by having been prompted by the new shiny thing (good generative AI). There's no reason why different types of AI that actually do have the power to do things beyond generating text or pictures shouldn't be dangerous but you hear very little about developments in those areas. It would seem that no one currently wants to, say, give an AI bot the keys to an operating system, or if they do, I'm not hearing about it.

*stares at "I am very skeptical <...> it can't" / "easily <...> with ease", finger hovers over the 'invite the nano bots in' button*

i cant take it any more im giving spellcheck the nuclear launch codes!!

Alba, there's this if you believe it? https://www.youtube.com/watch?v=KR-CTuQsRFU

Alphie, if you search for 'diamondoid' here you'll see that it's not usually framed in terms of a threat (as in, threatening before action) https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities

It's spelled nucular, Mark

It would seem that no one currently wants to, say, give an AI bot the keys to an operating system, or if they do, I'm not hearing about it.

humans do stupid things, give it time.

Now do crypto… pic.twitter.com/bgduMJvjjq

— Rupak (@ghose77) September 7, 2023