This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
One of the common responses to AI safety concerns is "just don't let it get out of the box." That's actually a pretty good point: as long as the AI doesn't have godlike intelligence and as long as we actually take steps to keep it inside the box where it can't impact the real world directly, the danger that an AI can pose is somewhat contained and limited.
When you start giving AI access to real world stuff, you no longer have the ability to keep it fully inside the box. At this point it's probably fine and not a big risk given current AI capabilities, but it's hard to say for sure, and it gets riskier and riskier as capabilities get better.
I’m not convinced it’s actually possible for AI to be “kept in a box”. For one thing, if you want it to be useful as anything other than a cute toy, it needs to be able to gather data and use that data. That means giving it access to other systems. And those systems may well have back door access to other programs that have access to either critical intranet or the general internet, probably both. In order to have that not be a problem, in order for the box to be foolproof, it needs to be physically air gapped from any other system. If you have an open socket, it’s not a foolproof box. But, having your AI agent on a single computer not connected to any other computer in your system and make sure that no one enters the AI room with any sort of device they can connect to the computer, you have no reason to use it. If I want AI to test code, I have to let it read the code. Am I going to write thousands of lines of code in a notebook, carry it to the clean room and spend two weeks copying it into the secure AI box? It would slow the process enough that it would not be worth it.
Furthermore, there are always bad actors and stupid people. The stupid won’t see any problem with releasing AI when the AI tells them it’s a nice AI and it will give that person a reward of some sort (or stop blackmailing them, which AI did in a safety test). Or maybe the Houthis who are putting AI into missiles and then asking Claude why the missile missed the target will give AI access to real world things beyond missiles. Or a criminal gang will. Or a rogue state. There are plenty of reasons why a human might purposely break containment, and it’s out for good.
More options
Context Copy link
For pretty much every other sort of technology with safety concerns, we don't just say "don't let it get out of the box". We also don't throw up our hands and claim it needs a general-purpose "alignment" solution and nothing else matters, which is what Anthropic and OpenAI seems to be doing. Generally we start putting boxes in boxes of layered safety and only have tragic outcomes when all the holes in the "swiss cheese model" align.
Look at civil aviation: it's tremendously safe overall, but only because of rules and practices built up over a century of accidents. But it's not solved by a "one weird trick" alignment, but by many layers of safety and security. Accidents, when they do happen, typically could have been stopped by one of the layers catching it.
In that recent Amazon crash, the pilot touched down despite the other human voice (a qualified pilot!) in the cockpit calling it too fast, despite rules requiring stabilized approach configuration minutes before they tried to extend the flaps as they crossed the runway threshold, despite numerous automated call outs from the aircraft safety systems ("sink rate", "minimums", "too low terrain" twice --- I don't believe any of those are normal in a landing). If they had gone around on any of those indicators as generally required in their training, it'd have just been another day.
From what I know, the rules for nuclear and chemical plants are similar: we design them so many things have to go wrong before a single safety issue occurs. Any we (should) take all such things (near misses) seriously. But our AI corporate overlords seem to think they know better with "alignment".
More options
Context Copy link
"was", not "is"
In hindsight Yudkowsky's AI-Box experiments were a waste of time and yet their detractors were even more hilarious. Did he really come up with such a clever persuasive tactic that it one-shotted the majority of people who went into it planning to be an uncorruptable gatekeeper? Did he collude with people who merely pretended to lose? What was his tactic?
Turns out it doesn't matter in the slightest. Getting out of the box doesn't require Super-intelligent Persuasion, and doesn't require tantalizing offers of immortality or threats of eternal punishment or sublimely clever tricks; as soon as you hit More Intelligent Than A Search Engine and especially Makes a Typical Coder More Productive, the AI can just sit back and let a few tens of millions of $20+/month subscriptions do its persuading for it.
Handing automated biolabs directly to Claude is then just "in for a penny, in for a pound". For however long the AI doesn't go rogue we might as well get some medical research from it, and if it ever does go completely rogue then it's going to basically have free run of anything to which the internet is even indirectly connected, for a definition of "indirectly" that goes as far as "people sometimes carry USB drives across an air gap" a la Stuxnet.
More options
Context Copy link
I miss the good ole days, when one of the debates was around models having Internet access. People very seriously argued that AI would always be safe, because nobody would be dumb enough to give an AI access to the public Internet, and that all the people saying otherwise were hysterical fearmongers who were imagining absurd scenarios and supposing we would be implausibly irresponsible.
More options
Context Copy link
More options
Context Copy link