Sure! :) Sorry if I came off as brusque, I was multi-tasking a bit.
Rob Bensinger
Karma: 9,284
I wasn’t bringing up evolution because you brought up evolution; I was bringing it up separately to draw a specific analogy.
By analogy, I’d ask you to consider why it doesn’t make sense to try to “cooperate” with the process of evolution. Evolution can be thought of as an optimizer, with a “goal” of maximizing inclusive reproductive fitness. Why do we just try to help actual conscious beings, rather than doing some compromise between “helping conscious beings” and “maximizing inclusive reproductive fitness” in order to be more fair to evolution?
A few reasons:
The things evolution “wants” are terrible. This isn’t a case of “vanilla or chocolate?”; it’s more like “serial killing or non-serial-killing?”.
(The links I gave above argue that the same is true for a random optimizer.)
Evolution isn’t a moral patient: it isn’t a person, it doesn’t have experiences or emotions, etc.
(A paperclip maximizer might be a moral patient, but it’s not obvious that it would be; and there are obvious reasons for us to deliberately design AGI systems to not be moral patients, if possible.)
Evolution can’t use threats or force to get us to do what it wants.
(Ditto a random optimizer, at least if we’re smart enough to not build threatening or coercive systems!)
Evolution won’t reciprocate if we’re nice to it.
(Ditto a random optimizer. This is still true after you build an unfriendly optimizer, though not for the same reasons: an unfriendly superintelligence is smart enough to reciprocate, but there’s no reason to do so relative to its own goals, if it can better achieve those goals through force.)
In the next part (forgiving me if this is way off) essentially you are saying my second question in the post is false, it wont be self aware or if it is it wont reflect enough to consider significantly rewriting its source code
No, this is not right. A better way of stating my claim is: “The notion of ‘self-awareness’ or ‘reflectiveness’ you’re appealing to here is a confused notion.” You’re doing the thing described in Ghosts in the Machine and Anthropomorphic Optimism, most likely for reasons described in Sympathetic Minds and Humans in Funny Suits: absent a conscious effort to correct for anthropomorphism, humans naturally model other agents in human-ish terms.
Im more positing at what point does paperclip maximizer learn so much it has a model of behaving in a manner that doesn’t optimize paperclips and explores that, or have a model of its own learning capabilities and explore optimizing for other utilities.
What does “exploring” mean? I think that I’m smart enough to imagine adopting an ichneumon wasp’s values, or a serial killer’s values, or the values of someone who hates baroque pop music and has strong pro-Spain nationalist sentiments; but I don’t try to actually adopt those values, it’s just a thought experiment. If a paperclip maximizer considers the thought experiment “what if I switched to less paperclip-centric values?”, why (given its current values) would it decide to make that switch?
maybe the initial task we give it should take into account what its potential volition may be at some point rather than just our own as a pre signal of pre committing to cooperation.
I think there’s a good version of ideas in this neighborhood, and a bad version of such ideas. The good version is cosmopolitan value and not trying to lock in the future to an overly narrow or parochial “present-day-human-beings” version of what’s good and beautiful.
The bad version is deliberately building a paperclipper out of a misguided sense of fairness to random counterfactual value systems, or out of a misguided hope that a paperclipper will spontaneously generate emotions of mercy, loyalty, or reciprocity when given a chance to convert especially noble and virtuous humans into paperclips.
To answer questions like these, I recommend reading https://www.lesswrong.com/rationality and then browsing https://arbital.com/explore/ai_alignment/. Especially relevant:
Or, quoting “The Value Learning Problem”:
[S]ystems that can strictly outperform humans cognitively have less to gain from integrating into existing economies and communities. Hall [2007] has argued:
“The economic law of comparative advantage states that cooperation between individuals of differing capabilities remains mutually beneficial. [ . . . ] In other words, even if AIs become much more productive than we are, it will remain to their advantage to trade with us and to ours to trade with them.”
As noted by Benson-Tilsen and Soares [forthcoming 2016], however, rational trade presupposes that agents expect more gains from trade than from coercion. Non-human species have various “comparative advantages” over humans, but humans generally exploit non-humans through force. Similar patterns can be observed in the history of human war and conquest. Whereas agents at similar capability levels have incentives to compromise, collaborate, and trade, agents with strong power advantages over others can have incentives to simply take what they want.
The upshot of this is that engineering a functioning society of powerful autonomous AI systems and humans requires that those AI systems be prosocial. The point is an abstract one, but it has important practical consequences: rational agents’ interests do not align automatically, particularly when they have very different goals and capabilities.
And quoting Ensuring smarter-than-human intelligence has a positive outcome:
The notion of AI systems “breaking free” of the shackles of their source code or spontaneously developing human-like desires is just confused. The AI system is its source code, and its actions will only ever follow from the execution of the instructions that we initiate. The CPU just keeps on executing the next instruction in the program register. We could write a program that manipulates its own code, including coded objectives. Even then, though, the manipulations that it makes are made as a result of executing the original code that we wrote; they do not stem from some kind of ghost in the machine.
The serious question with smarter-than-human AI is how we can ensure that the objectives we’ve specified are correct, and how we can minimize costly accidents and unintended consequences in cases of misspecification.
Enslaving conscious beings is obviously bad. It would be catastrophic to bake into future AGI systems the assumption that non-human animals, AI systems, ems, etc. can’t be moral patients, and there should be real effort to avoid accidentally building AI systems that are moral patients (or that contain moral patients as subsystems); and if we do build AI systems like that, then their interests need to be fully taken into account.
But the language you use in the post above is privileging the hypothesis that AGI systems’ conditional behavior and moral status will resemble a human’s, and that we can’t design smart optimizers any other way. You’re positing that sufficiently capable paperclip maximizers must end up with sufficient nobility of spirit to prize selflessness, trust, and universal brotherhood over paperclips; but what’s the causal mechanism by which this nobility of spirit enters the system’s values? It can’t just be “the system can reflect on its goals and edit them”, since the system’s decisions about which edits to make to its goals (if any) are based on the goals it already has.
You frame alignment as “servitude”, as though there’s a ghost or homunculus in the AI with pre-existing goals that the AI programmers ruthlessly subvert or overwrite. But there isn’t a ghost, just a choice by us to either build systems with humane-value-compatible or humane-value-incompatible optimization targets.
The links above argue that the default outcome, if you try to be “hands-off”, is a human-value-incompatible target—and not because inhumane values are what some ghost “really” wants, and being hands-off is a way of letting it follow through on its heart’s desire. Rather, the heart’s desire is purely a product of our design choices, with no “perfectly impartial and agent-neutral” reason to favor one option over any other (though plenty of humane reasons to do so!!), and the default outcome comes from the fact that many possible minds happen to converge on adversarial strategies, even though there’s no transcendent agent that “wants” this convergence to happen. Trying to cooperate with this convergence property is like trying to cooperate with gravity, or with a rock.
I feel like this comment should perhaps be an AIRCS class—not on meta-ethics, but on ‘how to think about what doing debugging your brain is, if your usual ontology is “some activities are object-level engineering, some activities are object-level science, and everything else is bullshit or recreation”’. (With meta-ethics addressed in passing as a concrete example.)
I felt a “click” in my brain reading this comment, like an old “something feels off, but I’m not sure what” feeling about rationality techniques finally resolving itself.
If this comment were a post, and I were in the curating-posts business, I’d curate it. The demystified concrete examples of the mental motion “use a tool from an unsciencey field to help debug scientists” are super helpful.
Can you too-tersely summarize your Nisbett and Wilson argument?
Or, like… writer a teaser / movie trailer for it, if you’re worried your summary would be incomplete or inoculating?
I don’t know Dario well, but I know enough to be able to tell that the anon here doesn’t know what they’re talking about re Dario.
More timeline statements, from Eliezer in March 2016:
That said, timelines are the hardest part of AGI issues to forecast, by which I mean that if you ask me for a specific year, I throw up my hands and say “Not only do I not know, I make the much stronger statement that nobody else has good knowledge either.” Fermi said that positive-net-energy from nuclear power wouldn’t be possible for 50 years, two years before he oversaw the construction of the first pile of uranium bricks to go critical. The way these things work is that they look fifty years off to the slightly skeptical, and ten years later, they still look fifty years off, and then suddenly there’s a breakthrough and they look five years off, at which point they’re actually 2 to 20 years off.
If you hold a gun to my head and say “Infer your probability distribution from your own actions, you self-proclaimed Bayesian” then I think I seem to be planning for a time horizon between 8 and 40 years, but some of that because there’s very little I think I can do in less than 8 years, and, you know, if it takes longer than 40 years there’ll probably be some replanning to do anyway over that time period.
And from me in April 2017:
Since [August], senior staff at MIRI have reassessed their views on how far off artificial general intelligence (AGI) is and concluded that shorter timelines are more likely than they were previously thinking. [...]
There’s no consensus among MIRI researchers on how long timelines are, and our aggregated estimate puts medium-to-high probability on scenarios in which the research community hasn’t developed AGI by, e.g., 2035. On average, however, research staff now assign moderately higher probability to AGI’s being developed before 2035 than we did a year or two ago.
I talked to Nate last month and he outlined the same concepts and arguments from Eliezer’s Oct. 2017 There’s No Fire Alarm for AGI (mentioned by Ben above) to describe his current view of timelines, in particular (quoting Eliezer’s post):
History shows that for the general public, and even for scientists not in a key inner circle, and even for scientists in that key circle, it is very often the case that key technological developments still seem decades away, five years before they show up. [...]
And again, that’s not to say that people saying “fifty years” is a certain sign that something is happening in a squash court; they were saying “fifty years” sixty years ago too. It’s saying that anyone who thinks technological timelines are actually forecastable, in advance, by people who are not looped in to the leading project’s progress reports and who don’t share all the best ideas about exactly how to do the thing and how much effort is required for that, is learning the wrong lesson from history. In particular, from reading history books that neatly lay out lines of progress and their visible signs that we all know now were important and evidential. It’s sometimes possible to say useful conditional things about the consequences of the big development whenever it happens, but it’s rarely possible to make confident predictions about the timing of those developments, beyond a one- or two-year horizon. And if you are one of the rare people who can call the timing, if people like that even exist, nobody else knows to pay attention to you and not to the Excited Futurists or Sober Skeptics. [...]
So far as I can presently estimate, now that we’ve had AlphaGo and a couple of other maybe/maybe-not shots across the bow, and seen a huge explosion of effort invested into machine learning and an enormous flood of papers, we are probably going to occupy our present epistemic state until very near the end.
By saying we’re probably going to be in roughly this epistemic state until almost the end, I don’t mean to say we know that AGI is imminent, or that there won’t be important new breakthroughs in AI in the intervening time. I mean that it’s hard to guess how many further insights are needed for AGI, or how long it will take to reach those insights. After the next breakthrough, we still won’t know how many more breakthroughs are needed, leaving us in pretty much the same epistemic state as before. Whatever discoveries and milestones come next, it will probably continue to be hard to guess how many further insights are needed, and timelines will continue to be similarly murky. Maybe researcher enthusiasm and funding will rise further, and we’ll be able to say that timelines are shortening; or maybe we’ll hit another AI winter, and we’ll know that’s a sign indicating that things will take longer than they would otherwise; but we still won’t know how long.
Yeah, to be clear, I don’t think everything Simplicio says is obviously wrong, and I think a more interesting (and much longer) version of this dialogue would have tried harder to understand and steelman various Simplicio-models, regardless of whether it ended up siding against Simplicio.
Inadequate Equilibria was Eliezer trying to explain the perspective he’d previously tagged “civilizational inadequacy” and before that “people are crazy, the world is mad”. One of the main ways he’d previously been misunderstood when he used those phrases was that people took them as “generic cynicism” or “the kind of cynicism I’m used to”, so my model of Eliezer considered it particularly important to differentiate those two things now that he was finally properly explaining the view.
I think calling the character “Simplicio” was a joke. I think that the main function the character is meant to serve is to distinguish “civilizational inadequacy” pessimism from other forms of pessimism. If the discussion were just “person talking about how things are broken vs. person who doesn’t expect things to be broken by default”, it would be easier to round off the former perspective to “generic pessimism/cynicism” or “the forms of pessimism/cynicism I’m most familiar with”.
I’m confused about how manioc detox is more useful to the group than the individual—each individual self-interestedly would prefer to detox manioc, since they will die (eventually) if they don’t.
Yeah, I was wrong about manioc.
Something about the “science is fragile” argument feels off to me. Perhaps it’s that I’m not really thinking about RCTs; I’m looking at Archimedes, Newton, and Feynman, and going “surely there’s something small that could have been tweaked about culture beforehand to make some of this low-hanging scientific fruit get grabbed earlier by a bunch of decent thinkers, rather than everything needing to wait for lone geniuses”. Something feels off to me when I visualize a world where all the stupidly-simple epistemic-methods-that-are-instrumentally-useful fruit got plucked 4000 years ago, but where Feynman can see big gains from mental habits like “look at the water” (which I do think happened).
Your other responses make sense. I’ll need to chew on your comments longer to see how much I end up updating overall toward your view.
I’m not sure how much we disagree; it sounds like I disagree with you, but maybe most of that is that we’re using different framings / success thresholds.
Efficient markets. Rationalists developed rationalist self-help by thinking about it for a while. This implies that everyone else left a $100 bill on the ground for the past 4000 years. If there were techniques to improve your financial, social, and romantic success that you could develop just by thinking about them, the same people who figured out the manioc detoxification techniques, or oracle bone randomization for hunting, or all the other amazingly complex adaptations they somehow developed, would have come up with them.
If you teleported me 4000 years into the past and deleted all of modernity and rationalism’s object-level knowledge of facts from my head, but let me keep as many thinking heuristics and habits of thought as I wanted, I think those heuristics would have a pretty large positive effect on my ability to pursue mundane happiness and success (compared to someone with the same object-level knowledge but more normal-for-the-time heuristics).
The way you described things here feels to me like it would yield a large overestimate of how much deliberate quality-adjusted optimization (or even experimentation and random-cultural-drift-plus-selection-for-things-rationalists-happen-to-value) human individuals and communities probably put into discovering, using, and propagating “rationalist skills that work” throughout all of human history.
Example: implementation intentions / TAPs are an almost comically simple idea. AFAIK, it has a large effect size that hasn’t fallen victim to the replication crisis (yet!). Humanity crystallized this idea in 1999. A well-calibrated model of “how much optimization humanity has put into generating, using, and propagating rationality techniques” shouldn’t strongly predict that an idea this useful and simple will reach fixation in any culture or group throughout human history before the 1990s, since this in fact never happened. But your paragraph above seems to me like it would predict that many societies throughout history would have made heavy use of TAPs.
I’d similarly worry that the “manioc detoxification is the norm + human societies are as efficient at installing mental habits and group norms as they are at detoxifying manioc” model should predict that the useful heuristics underlying the ‘scientific method’ (e.g., ‘test literally everything’, using controls, trying to randomize) reach fixation in more societies earlier.
Plausibly science is more useful to the group than to the individual; but the same is true for manioc detoxification. There’s something about ideas like science that caused societies not to converge on them earlier. (And this should hold with even more force for any ideas that are hard to come up with, deploy, or detect-the-usefulness-of without science.)
Another thing that it sounds like your stated model predicts: “adopting prediction markets wouldn’t help organizations or societies make money, or they’d already have been widely adopted”. (Of course, what helps the group succeed might not be what helps the relevant decisionmakers in that organization succeed. But it didn’t sound like you expected rationalists to outperform common practice or common sense on “normal” problems, even at the group level.)
I think I prefer bolding full lines b/c it makes it easier to see who authored what?
I’d be interested in trying it out. At a glance, it feels too much to me like it’s trying to get me to read Everything, when I can tell from the titles and snippets that some posts aren’t for me. If anything the posts I’ve already read are often ones I want emphasized more? (Because I’m curious to see if there are new comments on things I’ve already read, or I may otherwise want to revisit the post to link others to it, or finish reading it, etc.)
The bold font does look aesthetically fine and breaks things up in an interesting way, so I like the idea of maybe using it for more stuff?
The real disagreement is probably about whether the teacher would change her how-to-treat-evidence preferences if she were exposed to more information. Is her view stable, or would she see it for a confusion and mistake if she knew more, and say that she now sees things differently and more clearly?