Can we please put off worrying about dangerous AI for a couple hundred years? Chat GPT is a chatbot and not much else.. it is not the harbinger of an era of dangerous AI
Oh. Oh dear.
Of course, there is already a problem, as you point out. Humans shouldn't have access to these markets either!
But yeah to specifically prevent electron-on-atom violence we need to limit AI physical degrees-of-freedom. by limiting marketplaces. National/global security, not personal morality, should guide these new regulations.
We need to end all drive-by-wire automobiles, and electronic locks. Too many services are habituated to act on electronic signals without human confirmation - particularly the police. There needs to be an unbroken verbal chain between the person who saw an event and the person doing the law enforcement. Breaks in the human chain should be treated very seriously -they should be treated as firing offenses, at least. There are many other similar changes we're going to need to make.
Some folks aren't gonna like this. Regulations are inherently evil, they say. Maybe the mental model should be more like we're the big bad gorilla in the cage. But now there's a tiger in the cage. Regulation restrains the tiger. Also, some folks aren't gonna like it no matter what change you need. The fact of not liking it doesn't mean we don't need it, and it doesn't mean it won't get done. We have to trust that our leaders don't want to die.
And besides, the world will adapt. It always does. AI isn't optional, there's no putting the genie back in the bottle - and personally I don't want to. But I also don't want to be stupid about the stakes. Getting our whole species killed for lack of foresight would be deeply, deeply embarrassing.
The tool is too attractive not to use. The tool is too fun not to use. The tool is too dangerous to let out of the box, but that is exactly why we'll do it.
We're curious little monkeys, after all. "What do you think will happen" absolutely is a survival strategy for our species. The problem is when we encounter something that is so much more advanced than us, even if that advance portion is just access to multiple systems of our own creation.
To summarize: I think you make a good point, but I think we're fucked eventually anyways.
I can't wait for the inevitable "does my AI have the right to freedom" case in the supreme court when I'm in my 90's.
HN itself has been spammed relentlessly with people hooking it up to everything they can think of in an attempt to get a worthless reward (karma)
now imagine there's money, power or territory up for grabs instead
we are completely fucked
Leaving everything else aside, how would this look in practice? I think these conversations would need to be in person, since voice can already be faked. Would I need to run to the police station when I need help?
If that is too much trouble to ask, then is justice even possible?
Another scenario: I'm a police officer and I'm on patrol. My dispatcher had someone come to the police station to tell them that they think their neighbor is experiencing a home invasion. Does the dispatcher just page me and I now drive back to the police station to verify and then drive back out to the home invasion?
Lord, give me patience.
Call 911. The dispatcher broadcasts the problem over the radio, and a LEO responds. The dispatcher is a relay that verifies probable cause. The chain of human contact is unbroken between you, the 911 dispatcher, the LEO taking the call. The chain is not broken.
Compare this to a machine that spits out warrants, which are distributed to officers, who never even speak to anyone about the case, do not know the subject of the warrant, and simply execute the warrants.
We are also probably days away from video being trivial to fake.
Progress in AI doesn't imply that we're dangerously close to AGI, just because people at any given time are amazed by individual breakthroughs they witness.
Why do you think this?
There are loads of tests of their performance. Common one right now is to give LLMs the same exams we put humans through, leading to e.g. the graph on page 6: https://arxiv.org/pdf/2303.08774.pdf
Are they the best tests? Probably not! But they are definitely empirical.
Give students concurrent access to the internet and I'm sure they can pass all sorts of tests.
First, students only get good after studying — education is not some magic spell cast by the teacher that only operates on a human's immortal soul. As we should not dismiss what students learn just because we could look it up, it is strange to dismiss what GPT has learned just because it could be looked up.
Second, the GPT-3 (and presumably also GPT-4) training set is about 500e9 tokens, which is what? Something like just a few terabytes?
We've been able to store that in a pocket for years now without being able to do almost any of the things that GPT can do — arbitrary natural language synthesis let alone arbitrary natural language queries — on a computer, even when we programmed the rules, and in this case the program learned the rules from the content.
Even just a few years ago, SOTA NLP was basically just "count up how many good words and bad words are in the text, the sentiment score is total good minus total bad."
That difference is what these test scores are showing.
Let's say: adversarial Turing test + MMLU + coding competence (e.g. AAPS or Leetcode) + ARC (IQ-type test) + Montezuma's Revenge and other games like Stratego or Diplomacy + USMLE (medical exam) + IMO (math) + self driving + ...
You can even make it harder: have human judges blindly evaluate new scientific papers in math or theoretical physics for acceptance, see if AI can create highly-rated new apps, write a highly-rated book, compose a hit song...
I would measure using something similar to Yudkowsky's challenge: "What is the *least* impressive feat that you would bet big money at 9-1 odds *cannot possibly* be done in 2 years?" [1]
Pay a panel of experts to list their predictions each year, including an incentive to get it right, and then measure the percentage of those predictions that fail anyway.
[1] https://twitter.com/ESYudkowsky/status/910566159249899520
However, human beings will treat it like it is, so you will experience non-sensical happenings like “your new credit score, brought to you by AI”. When you dispute this, the credit score company will shrug and say “you’ll have to speak the AI directly to make a claim, it is objective”. Meanwhile the AI isn’t that much better than ChatGPT now
We have no idea what is happening behind the curtain, and to assume that private industry or the might of the US military is not decades ahead of what is available for consumers is just naive at best.
What is it about chatbots that trigger people? Probably the importance and centrality of language in human consciousness, and the manifest credularity of the general population (and many readers of HN).
Unfortunately it's unlikely this will stop, and it'll probably get worse. The final irony will be that when some "real AI" is created, no-one will believe it, having been exhausted by 200 years of hype.
In fact, I think the hard part of the whole thing would be trying to make the network traffic and source IPs look legitimate to the site owners/admins and it will be interesting to see how the battle unfolds to start authenticating content posted in communities as having come from an actual human.