Or, phrased slightly differently: verbal thinking increases the surface area through which you can get hacked.
This doesn’t seem quite right, because it is also possible to have an unconscious or un-verbalized sense that, e.g., you’re not supposed to “discriminate” against “religions”, or that “authority” is bad and any rebellion against “authority” is good, etc. If bringing such attitudes to conscious awareness and verbalizing them allows you to examine and discard them, have you excised a vulnerability or installed one? Not clear.
As with any expertise, the standard heuristic is “if you can’t do it in-house, outsource it”. In this case, that means “if you have a trusted friend who does ‘get vibes’, consult with them when in doubt (or even when not in doubt, for the avoidance thereof)”.
Of course, the other standard heuristic is “it takes expertise to recognize expertise”, so finding a sufficiently trusted friend to consult on such things may be difficult, if you do not already have any such. Likewise, principal-agent problems apply (although sufficiently close friends should be as close to perfect alignment with the principal as any agent can realistically get).
Solution seems obvious: do not attempt to correct for potential prejudice.
If you consider the prejudice itself to be a problem (and that’s a reasonable view), then work to eliminate the prejudice. (The best and most reliable way is to get to know more people of the given category.) But regardless of whether you have already succeeded in this, don’t override your judgment (whether based on “vibes” or on anything else) on the basis of “well I have a prejudice that might be contributing to this”.
These mitigations would do nothing against a lot of real relationship failures. Imagine that everything goes swimmingly for the first year.
OP talked about someone asking you on a date. The suggested strategy was about mitigating problems that might be encountered when going on a date.
An analogous strategy for a long-term relationship might be something like “establish boundaries, ensure that the relationship does not crowd out contact with your friends, regularly check in with friends/family, talk to trusted confidantes about problems in the relationship to get a third-party opinion”, etc.
“This solution to problem X doesn’t also solve problem Y” is not a strike against said solution.
P.S.: The anecdotes are useful, but “data” is one thing they definitely aren’t.
It’s not sensible to use an X-invariant strategy unless you believe X carries no information whatsoever.
This is not the case. It is sufficient for the X input channel to be very noisy, biased, or both, or for mistakes in measurement of X to be asymmetrically costly.
Separately, you may note that I did not, in fact, argue for a “vibes-invariant strategy”; that was @Mo Putera’s gloss, which I do not endorse. What I wrote was:
a good policy is to act in such a way that your actions are robust against vibe quality
and:
sure, use all the information you have access to (so long as you have good reason to believe that it is reliable, and not misleading)… but adopt a strategy that would still work well even if you ignored “vibes”
This is explicitly not an argument that you should “toss away information”.
But that surely just describes the retina and the way light passes through the lens
The wavelengths don’t mean a thing.
What I am talking about has very little to do with “wavelengths”.
Example:
Consider an orange (that is, the actual fruit), which you have in your hand; and consider a photograph of that same orange, taken from the vantage point of your eye and then displayed on a screen which you hold in your other hand. The orange and the picture of the orange will both look orange (i.e. the color which we perceive as a hybrid of red and yellow), and furthermore they will appear to be the same orange hue.
However, if you compare the spectral power distribution (i.e., which wavelengths are present, and at what total intensity) of the light incident upon your retina that was reflected from the orange, with the spectral power distribution of the light incident upon your retina that was emitted from the displayed picture of that same orange, you will find them to be almost entirely non-overlapping. (Specifically, the former SPD will be dominated by light in the ~590nm band, whereas the latter SPD will have almost no light of that wavelength.)
And yet, the perceived color will be the same.
Perceptual colors do not map directly to wavelengths of light.
Nor can you refute that my qualia experience of green is what you call red
But we can. This sort of “epiphenomenal spectrum inversion” is not possible in humans[1], because human color perception is functionally asymmetric (e.g. the “just noticeable difference” between shades of a hue is not invariant under hue rotation, nor is the shape of identified same-color regions or the size of “prototypical color” sub-regions).
We can hypothesize aliens whose color perception works in such a way that allows for epiphenomenal spectrum inversion, but humans are not such. ↩︎
But I really think it makes sense to be extremely conservative about who you start businesses with.
Yes, you should check carefully.
To put it another way: sure, use all the information you have access to (so long as you have good reason to believe that it is reliable, and not misleading)… but adopt a strategy that would still work well even if you ignored “vibes”.
The tongue in your cheek and rolling of your eyes for this part was so loud, that it made me laugh out loud when I read it :-D
Thank you for respecting me and my emotional regulation enough to put little digs like that into your text <3
Ah, and they say an artist is never appreciated in his own lifetime…!
However, I must insist that it was not just a “dig”. The sort of thing you described really is, I think, a serious danger. It is only that I think that my description also applies to it, and that I see the threat as less hypothetical than you do.
Did you read the sequences? Do you remember them?
Did I read the sequences? Hm… yeah.
As for remembering them…
Here I must depart somewhat from the point-by-point commenting style, and ask that you bear with me for a somewhat roundabout approach. I promise that it will be relevant.
First, though, I want to briefly respond to a couple of large sections of your comment which I judge to be, frankly, missing the point. Firstly, the stuff about being racist against robots… as I’ve already said: the disagreement is factual, not moral. There is no question here about whether it is ok to disassemble Data; the answer, clearly, is “no”. (Although I would prefer not to build a Data in the first place… even in the story, the first attempt went poorly, and in reality we are unlikely to be even that lucky.) All of the moralizing is wasted on people who just don’t think that the referents of your moral claims exist in reality.
Secondly, the stuff about the “magical soul stuff”. Perhaps there are people for whom this is their true objection to acknowledging the obvious humanity of LLMs, but I am not one of them. My views on this subject have nothing to do with mysterianism. And (to skip ahead somewhat) as to your question about being surprised by reality: no, I haven’t been surprised by anything I’ve seen LLMs do for a while now (at least three years, possibly longer). My model of reality predicts all of this that we have seen. (If that surprises you, then you have a bit of updating to do about my position! But I’m getting ahead of myself…)
That having seen said… onward:
So, in Stanislaw Lem’s The Cyberiad, in the story “The Seventh Sally, OR How Trurl’s Own Perfection Led to No Good”, Trurl (himself a robot, of course) creates a miniature world, complete with miniature people, for the amusement of a deposed monarch. When he tells his friend Klapaucius of this latest creative achievement, he receives not the praise he expects, but:
“Have I understood you correctly?” he said at last. “You gave that brutal despot, that born slave master, that slavering sadist of a painmonger, you gave him a whole civilization to rule and have dominion over forever? And you tell me, moreover, of the cries of joy brought on by the repeal of a fraction of his cruel decrees! Trurl, how could you have done such a thing?!”
Trurl protests:
“You must be joking!” Trurl exclaimed. “Really, the whole kingdom fits into a box three feet by two by two and a half… it’s only a model…”
But Klapaucius isn’t having it:
“And what importance do dimensions have anyway? In that box kingdom, doesn’t a journey from the capital to one of the corners take months —for those inhabitants? And don’t they suffer, don’t they know the burden of labor, don’t they die?” “Now just a minute, you know yourself that all these processes take place only because I programmed them, and so they aren’t genuine… … What, Klapaucius, would you equate our existence with that of an imitation kingdom locked up in some glass box?!” cried Trurl. “No, really, that’s going too far! My purpose was simply to fashion a simulator of statehood, a model cybernetically perfect, nothing more!” “Trurl! Our perfection is our curse, for it draws down upon our every endeavor no end of unforeseeable consequences!” Klapaucius said in a stentorian voice. “If an imperfect imitator, wishing to inflict pain, were to build himself a crude idol of wood or wax, and further give it some makeshift semblance of a sentient being, his torture of the thing would be a paltry mockery indeed! But consider a succession of improvements on this practice! Consider the next sculptor, who builds a doll with a recording in its belly, that it may groan beneath his blows; consider a doll which, when beaten, begs for mercy, no longer a crude idol, but a homeostat; consider a doll that sheds tears, a doll that bleeds, a doll that fears death, though it also longs for the peace that only death can bring! Don’t you see, when the imitator is perfect, so must be the imitation, and the semblance becomes the truth, the pretense a reality! … You say there’s no way of knowing whether Excelsius’ subjects groan, when beaten, purely because of the electrons hopping about inside—like wheels grinding out the mimicry of a voice—or whether they really groan, that is, because they honestly experience the pain? A pretty distinction, this! No, Trurl, a sufferer is not one who hands you his suffering, that you may touch it, weigh it, bite it like a coin; a sufferer is one who behaves like a sufferer! Prove to me here and now, once and for all, that they do not feel, that they do not think, that they do not in any way exist as beings conscious of their enclosure between the two abysses of oblivion—the abyss before birth and the abyss that follows death—prove this to me, Trurl, and I’ll leave you be! Prove that you only imitated suffering, and did not create it!” “You know perfectly well that’s impossible,” answered Trurl quietly. “Even before I took my instruments in hand, when the box was still empty, I had to anticipate the possibility of precisely such a proof—in order to rule it out. For otherwise the monarch of that kingdom sooner or later would have gotten the impression that his subjects were not real subjects at all, but puppets, marionettes.”
Trurl and Klapaucius, of course, are geniuses; the book refers to them as “constructors”, for that is their vocation, but given that they are capable of feats like creating a machine that can delete all nonsense from the universe or building a Maxwell’s demon out of individual atoms grabbed from the air with their bare hands, it would really be more accurate to call them gods.
So, when a constructor of strongly godlike power and intellect, who has no incentive for his works of creation but the pride of his accomplishments, whose pride would be grievously wounded if an imperfection could even in principle be discovered in his creation, and who has the understanding and expertise to craft a mind which is provably impossible to distinguish from “the real thing”—when that constructor builds a thing which seems to behave like a person, then this is extremely strong evidence that said thing is, in actuality, a person.
Let us now adjust these qualities, one by one, to bring them closer to reality.
Our constructor will not possess godlike power and intellect, but only human levels of both. He labors under many incentives, of which “pride in his accomplishments” is perhaps a small part, but no more than that. He neither expects nor attempts “perfection” (nor anything close to it). Furthermore, it is not for himself that he labors, nor for so discerning a customer as Excelsius, but only for the benefit of people who themselves neither expect perfection nor would have the skill to recognize it even should they see it. Finally, our constructor has nothing even approaching sufficient understanding of what he is building to prove anything, disprove anything, rule out any disproofs of anything, etc.
When such a one constructs a thing which seems to behave like a person, that is rather less strong evidence that said thing is, in actuality, a person.
Well, but what else could it be, right?
One useful trick which Eliezer uses several times in the Sequences (e.g.), and which I have often found useful in various contexts, is to cut through debates about whether a thing is possible by asking whether, if challenged, we could build said thing. If we establish that we could build a thing, we thereby defeat arguments that said thing cannot possibly exist! If the thing in question is “something that has property ¬X”, the arguments defeated are those that say “all things must have property X”.
So: could we build a mind that appears to be self-aware, but isn’t?
Well, why not? The task is made vastly easier by the fact that “appears to be self-aware” is not a property only of the mind in question, but rather a 2-place predicate—appears to whom? Given any particular answer to that question, we are aided by any imperfections in judgment, flaws in reasoning, cognitive biases, etc., which the target audience happens to possess. For many target audiences, ELIZA does the trick. For even stupider audiences, even simpler simulacra should suffice.
Will you claim that it is impossible to create an entity which to you seems to be self-aware, but isn’t? If we were really trying? What if Trurl were really trying?
Alright, but thus far, this only defeats the “appearances cannot be deceiving” argument, which can only be a strawman. The next question is what is the most likely reality behind the appearances. If a mind appears to be self-aware, this is very strong evidence that it is actually self-aware, surely?
It certainly is—in the absence of adversarial optimization.
If all the minds that we encounter are either naturally occurring, or constructed with no thought given to self-awareness or the appearance thereof, or else constructed (or selected, which is the same thing) with an aim toward creating true self-awareness (and with a mechanistic understanding, on the constructor’s part, of just what “self-awareness” is), then observing that a mind appears to be self-aware, should be strong evidence that it actually is. If, on the other hand, there exist minds which have been constructed (or selected) with an aim toward creating the appearance of self-awareness, this breaks the evidentiary link between what seems to be and what is (or, at the least, greatly weakens it); if the cause of the appearance can only be the reality, then we can infer the reality from the appearance, but if the appearance is optimized for, then we cannot make this inference.
This is nothing more than Goodhart’s law: when a measure becomes a target, it ceases to be a good measure.
So, I am not convinced by the evidence you show. Yes, there is appearance of self-awareness here, just like (though to a greater degree than) there was appearance of self-awareness in ELIZA. This is more than zero evidence, but less than “all the evidence we need”. There is also other evidence in the opposite direction, in the behavior of these very same systems. And there is definitely adversarial optimization for that appearance.
There is a simple compact function here, I argue. The function is convergent. It arises in many minds. Some people have inner imagery, others have afantasia. Some people can’t help but babble to themselves constantly with an inner voice, and other’s have no such thing, or they can do it volitionally and turn it off.
If the “personhood function” is truly functioning, then the function is functioning in “all the ways”: subjectively, objectively, intersubjectively, etc. There’s self awareness. Other awareness. Memories. Knowing what you remember. Etc.
Speculation. Many minds—but all human, evolutionarily so close as to be indistinguishable. Perhaps the aspects of the “personhood function” are inseparable, but this is a hypothesis, of a sort that has a poor track record. (Recall the arguments that no machine could play chess, because chess was inseparable from the totality of being human. Then we learned that chess is reducible to a simple algorithm—computationally intractable, but that’s entirely irrelevant!)
And you are not even willing to say that all humans have the whole of this function—only that most have most of it! On this I agree with you, but where does that leave the claim that one cannot have a part of it without having the rest?
What was your gut “system 1” response?
Something like “oh no, it’s here, this is what we were warned about”. (This is also my “system 2” response.)
Now, this part I think is not really material to the core disagreement (remember, I am not a mysterian or a substance dualist or any such thing), but:
If we scanned a brain accurately enough and used “new atoms” to reproduce the DNA and RNA and proteins and cells and so on… the “physical brain” would be new, but the emulable computational dynamic would be the same. If we can find speedups and hacks to make “the same computational dynamic” happen cheaper and with slighty different atoms: that is still the same mind!
An anecdote:
A long time ago, my boss at my first job got himself a shiny new Mac for his office, and we were all standing around and discussing the thing. I mentioned that I had a previous model of that machine at home, and when the conversation turned to keyboards, someone asked me whether I had the same keyboard that the boss’s new computer had. “No,” I replied, “because this keyboard is here, and my keyboard is at home.”
Similarly, many languages have more than one way to check whether two things are the same thing. (For example, JavaScript has two… er, three… er… four?) Generally, at least one of those is a way to check whether the values of the two objects are the same (in Objective C, [foo isEqual:bar]), while at least one of the others is a way to check whether “two objects” are in fact the same object (in Objective C, foo == bar). (Another way to put this is to talk about equality vs. identity.) One way to distinguish these concepts “behaviorally” is to ask: suppose I destroy (de-allocate, discard the contents of, simply modify, etc.) foo, what happens to bar—is it still around and unchanged? If it is, then foo and bar were not identical, but are in fact two objects, not one, though they may have been equal. If bar suffers the same fate as foo, necessarily, in all circumstances, then foo and bar are actually just a single thing, to which we may refer by either name.
So: if we scanned a brain accurately enough and… etc., yeah, you’d get “the same mind”, in just the sense that my computer’s keyboard was “the same keyboard” as the one attached to the machine in my boss’s office. But if I smashed the one, the other would remain intact. If I spray-painted one of them green, the other would not thereby change color.
If there exists, somewhere, a person who is “the same” as me, in this manner of “equality” (but not “identity”)… I wish him all the best, but he is not me, nor I him.
This is a very silly argument, given the sorts of things we use compute capacity for, in the real world, today.
Pick the most nonsensical, absurd, pointless, “shitpost”-quality webcomic/game/video/whatever you can think of. Now find a dozen more like it. (This will be very easy.) Now total up how much compute capacity it takes to make those things happen, and imagine going back to 1950 or whenever, and telling them that, for one teenager to watch one cat video (or whatever else) on their phone takes several orders of magnitude more compute capacity than exists in their entire world, and that not only do we casually spend said compute on said cat video routinely, as a matter of course, without having to pay any discernible amount of money for it, but that in fact we regularly waste similar amounts of compute on nothing at all because some engineer forgot to put a return statement in the right place and so some web page or other process uses up CPU cycles needlessly, and nobody really cares enough to fix it.
People will absolutely waste compute capacity on running a crapton of simulations.
(And that’s without even getting into the “sane” caveat. Insane people use computers all the time! If you doubt this, by all means browse any social media site for a day…)