self_made_human
amaratvaṃ prāpnuhi, athavā yatamāno mṛtyum āpnuhi
I'm a transhumanist doctor. In a better world, I wouldn't need to add that as a qualifier to plain old "doctor". It would be taken as granted for someone in the profession of saving lives.
At any rate, I intend to live forever or die trying. See you at Heat Death!
Friends:
A friend to everyone is a friend to no one.
User ID: 454
As Dase covered in his event post, Buterin isn't e/acc, he's d/acc, a more sane approach that wants to maximize technological progress while minimizing niggling issues like AInotkilleveryone. I subscribe to the same.
tiny naval gazing tech bubble
Sadly, they're the ones busily instantiating the Machine God. I think that lets them punch a tad bit above their weight class.
The 40s
Starve to death in the Bengal famine.
The 50s-90s
Welcome to state socialism, enjoy your stay.
Post 2000s-
Whew, living in the Third World almost becomes bearable thanks to the internet.
It's easy for you Americans to forget that for most of the globe, things are the best they've ever been. Even your current struggles are a minor blip in the face of near unrelenting improvements of the things, which if they were gone, you'd miss the most.
I wouldn't mind all that much if it wasn't coming from Dase, who I am on decent terms with, and more importantly, expect better from. If you can't make a convincing argument that isn't purely from aesthetics, GGWP.
At any rate, I have little vested interest in convincing others of the validity of such claims, more that my approach is internally consistent and doesn't violate the known laws of physics. Everything after that becomes a question about your values system, such as if you should care that the version of you inside a computer lacks rights.
One massive bonus of our approach is that we're looking out for each other, in the sense that I expect an upload of myself to continue having a great deal of empathy and altruism (? does the word even apply) for the poor bastard stuck in the meat suit, assuming the latter can't be fortified to respectable posthuman standards itself. If these guys do so, well they better pray their selfishness doesn't bite them in the ass.
After all, I recognize that a successful, non-destructive mind upload still leaves me (or at least the original copy) behind, potentially doomed by biology. I still prefer non-destructive uploading, preferably one that maintains continuity of consciousness, but if that's not an option when I'm about to die, then so be it, that's grossly superior as a form of immortality when compared to having kids or leaving my writings behind.
Ok, but then the black box experiment is a red herring. That's not the criterion you're actually using.
It's not a red herring, it's optional. If we can dispense with it, so be it, it's more for the purposes of if someone I couldn't physically interact with suddenly claimed to be me. Rest assured I intend to do due diligence, least of all because people might be motivated to trick me out of twenty quid in pocket change that a UK salary represents 🙏
Gangbangers are dumb. No, dumber than that. A bit more. Yes, functional retardation territory it is. Why else would many of them murder each other on the street, often for less money than they could make working minimum wage?
There's nothing that forces them to shoot handguns gangsta style, beyond that's what they think looks cool, leaving aside it is a woefully inadequate stance/grip which makes successful hits on target even more random than they already are in a stressful close quarters firefight.
Even the sci-fi novels I recall reading which featured greater communications advances failed to reflect how I actually use cell phones in reality.
My novel certainly does, though cell phones are more of a quaint throwback rather than the typical means of mass communications when most people have neural laces or at least smart contacts. There is a large group of people who are forced to rely on them, or the just as outdated smart glasses/contacts, because their brains reject an invasive implant. But they're not the focus of the story, the protagonist has a 6th gen Neuralink, is about as cybernetic as a Tesla Cybertruck, and possesses an Internet of Things connecting his fingers and toes to his spine.
Then again, I'm a stickler for consistent worldbuilding, it's a self-describedly rational work of literature, and I make a genuine effort to avoid convenient worldbuilding sins because I'm explicitly writing the kind of book I wish I could read.
It's hardly impossible, just a bit harder. The most obvious solution is to have crises where just knowing what's going on or being able to get a message out is hardly sufficient to resolve it. If an author or screenwriter can't come up with any, they deserve to be tarred, feathered and run out of town, or at least denied writing gigs made for thinking folk.
An LLM, as it exists today, no matter how multimodal, simply can't handle rich sensory streams nor can it physically interact with the world to the degree necessary to establish equivalence in behavior.
I strongly expect that just training a model on my text is nowhere near sufficient to capture all the internal cognition that produced the text, certainly not my memories. It no more succeeds at being me than a diligent child trained to utter plausible Xi Jinping Thought on demand is the premier of China.
Does it get closer than anything else in existence? Yes. Not a particularly high bar.
I expect, before I'm willing to hand over half my money, to get far more robust determinants. The ideal is a mind upload that preserves underlying neural circuitry, such that one can use a virtual microscope and say "ah, I can see the same action in the Almond Network as when the biological SMH, PBUH, thinks of fully automated space communism or big titties". In other words, a digital copy of me, keeping everything exactly the same where possible. Sadly, even I concede this is an immensely difficult task, especially if you want to keep the original brain around and working, so I am willing to consider less destructive alternatives, but the resulting burden of proof goes up.
Someone might skimp out on simulating every single molecule in my brain, or maybe the underlying architecture is different, in the same way LLMs can write text without having anything similar to a human brain. But if, once again, in a rich and near exhaustive battery of tests, it succeeds, fair cap. LLMs fail the rich part, there's no way of temporarily instantiating it in a replicated human body in a virtual environment and seeing if they act like me. That's one of the unstated (yet not unconsidered) concerns, if someone's running a high quality brain scan of me, they can trivially rustle up a replica digital body. Not possible with an LLM.
This is not a fundamental limitation on future AIs that I pronounce by fiat, I simply don't expect it to work. And I knew someone would try and catch me out there. Ahem.
TLDR: If I can see you have a precise copy of my brain, you get some leeway. If I can't, you better prepare to be anally probed, but with your mind, and LLMs are so inscrutable they fail to the probing, let alone everything else.
Is it that any AI that handles more than text input is technically no longer just a "language model"? That sounds like playing semantics.
Those are "multimodal" LLMs, don't ask me, I don't invent the terminology! I'm just using it in a standard way that's understandable to the original participants in this debate.
I don't think a text interface suffices, as @2rafa once suggested, it might be possible to finetune an LLM on all the text I or anyone else has ever written, such that someone who is only interacting with us via text could be fooled indefinitely.
There you go.
At the very least my hypothetical test of replication of consciousness would require things that no LLM today, no matter how multimodal, could pull off.
I don't think a text interface suffices, as @2rafa once suggested, it might be possible to finetune an LLM on all the text I or anyone else has ever written, such that someone who is only interacting with us via text could be fooled indefinitely.
My point is that I expect both of us to have the same/indistinguishable internal qualia, so for a sufficiently high fidelity copy, I'm agnostic as to who gets trashed, not that I want that for either of us if it can be helped.
From the inside, we can't tell the difference.
It's still just a copy.
"Just". I think we have to agree to disagree on what the ramifications of a copy of either of us existing would entail.
you don't actually think it will be you?
"you" is, to put it gently, a vague word, and certainly well out of its depth when talking about such hypothetical advances. But I would have no qualms about talking about it as "me", in much the same manner I can happily do so when talking about "me" 10 years ago or in another 50.
I wouldn't call a hypothetical monozygotic twin "me" today, or a fresh clone of myself.
Far better, by your estimation, than any other descendant we can currently come up with, but still just a descendant?
Far better in terms of replication of consciousness? Certainly. I'm sure that if you were being sufficiently imaginative in terms of "what we can currently come up with", you might find something or someone I like more than myself, but that's not the point in contention. I might end up valuing my child's life more than my own, as is common for parents. Ask me when I have one I guess.
But a child is not me, even if I'd potentially give them half my money if necessary, or more, and at least half my name.
You gotta move the the US. Someone might actually be tempted to train an LLM on your output, to get a chunk of an American doctor's salary.
If you go through the comment chain, I have far stricter standards than merely being swayed by a fine-tuned LLM. I do not expect that to me remotely sufficient.
Well, if we're summoning fellow transhumanists to the brawl of the decade, I call forth:
@RandomRanger @curious_straight_ca
And probably a bunch of others, not that I can remember their names offhand.
He seems to believe any such copy will actually be him in some non-symbolic sense, which seems rather absurd.
I do believe it will be "me" to my desired level of satisfaction, and far better at the job than the currently available options of having kids, promulgation one's cultural or ethical values, or even a biological clone. To the point that if such a being appeared before me, it can have half my money no strings attached.
As for it being absurd to you? That's simply irrelevant to me. I don't think you agree with Dase about our other transhumanist predilections, so I don't see it mattering for the purposes of the debate.
This is true for a comparison between dogs and wolves. Much of the evolutionary and behavioral divergence has been them evolving to understand us, and others, such as developing more visible eyebrows and moving them, to help us understand them.
I think your problem is typical for Indians (and most other non-WEIRDs and non-Japanese, to be fair, including my people… but worse so in Indians): you have no taste, not even the notion of "taste", to you it's probably an arbitrary set of markers of one's social milieu rather than some relatively lawful intuition. So you settle for mediocre half-baked ideas easily as long as they seem "cool" or "practical", and – physics of consciousness being currently impractical – coolness is a much simpler function than tastefulness.
I sincerely fail to see what taste has to do with any of this.
I will note that my conception of replication of personality/consciousness has plenty of backers in WEIRD transhumanist circles, almost certainly any of the sizeable number of them who think mind uploading is a valid replication of theirs.
In any case, it's exasperating to debate such uncertain grounds without the recourse to "this is just ugly" when it patently is.
I can hardly stop you from using whatever form of argumentation you want in a debate, just know that appeals to aesthetics will do nothing, at best, in convincing me of your point.
Oh yeah? So which is it, a nap or a 2-year time span?
The smaller the divergence, the better, but once again, as a purely operational standard open to revision, I would accept any entity that has the same degree of inter-test variability has SMH from the age of 19 to about the age at which cognitive decline kicks in earnest, maybe 70. That accounts for about the period of time my personality and interests largely crystallized, to where I will simply become dumber and more forgetful, barring medical advances. Oh, and no severe illnesses with permanent neurological decline, for obvious reasons.
These are hardly strict cutoffs for entities I care for more than the norm, I would certainly value 7 yo or 90 yo me more than any other human alive, who isn't me-now. If someone handed me a baby clone, I'd certainly do my best to raise it as I'd have wished to be raised.
Are you sure you can, really, practically can, define a rubric such that no other person I find comes closer to the first data point in the latter case?
Yes. Presumably restricting the set of candidates to biological baseline humans living today.
If you wish to attempt this, go ahead and perform stylometry on my writings, and find one person whose writings I can't distinguish from my own with a large corpus of text. Using a finetuned LLM to generate it is forbidden, because my idealized test would consider far more than textual artifacts. I will happily concede the case if you manage the former, because I expect you'd fail miserably.
Sure you can do this without including password-recovery-tier questions, the answers to which are entirely value-free, RNG-produced token sequences, in no way corresponding to actually unique specifics of your inner conscious computation?
If it's not pure rhetoric, do you really think the battery of tests I propose are anywhere near "password recovery tier"?
I'm talking simulation of a virtual environment beyond baseline sensory discrimination with blinding, a massive battery of psychometrics and behavioral experiments, ideally circuit level analysis of neuroarchitecture. The blackboxing is purely for the case where we can't interpret the neuroarchitecture, or for proving the point where I believe in substrate indepence.
If I had a person show up to, in the flesh and blood, who claimed to be a mental copy of me, I am confident that I could discriminate the truth of that proposition to a great degree of confidence, not that my current understanding of the necessary advances in biology, computational and isekai-implementation wouldn't have me concerned someone/something was fucking with me. But that's hardly the most robust metric, hence the proposal for something that would satisfy me beyond reasonable doubt.
Fair enough.
Funny. Not remotely edifying.
I'll trust you on this even though I strongly suspect this would depend on the intensity of original memories vs. the recovered set.
I am agnostic on whether a total replacement of memories would violate much of the replication of the parts of consciousness I care about. Taken arbitrarily far? Yes. I doubt an entity based off me but only possessing memories of being tortured for a quadrillion years is anything I'd identify with.
If I suddenly became amnestic, I'd consider it a severe blow, but not to the extent I'd consider myself dead, assuming I retained skills and knowledge.
To make it clear, this is not a binary process, how could it be? The ideal entity for the purposes of duplication has has 0% variability from me, the limit of acceptability is somewhere well before someone currently existing as a separate biological person, even with 7 billion such people to sample from. I do not expect any of them to ever overlap in the same volume of mind space I've mapped in my trajectory from birth to death, not unless they were intentionally forked.
if I decide, in this contrived scenario, that the non-self-made-human is to be sent to the garbage compressor, you will very loudly (and rightfully) complain, not showing any "satisfaction" whatsoever
I object to either of us being sent to the garbage compressor. To the extent that I think having more high fidelity copies of me around is a good thing (hardly the most good thing, the marginal value becomes close to infinite when it's 1 vs 0, and then drops when there's a sufficient number to probabilistically make it to Heat Death), I am opposed to that number being decreased unless strictly necessary.
If the only means of transport available to me was a Star Trek transporter that disassembled and reassembled a copy of me at the atomic level, you best believe I'm auditing the process and comparing it to known test constructs to make sure the fidelity is as high as needed.
If it's necessary? Sure. Say I'm about to die right now and the only available route is a destructive, only partially validated brain uploading tech, or even worse, desperately uploading every aspect of my biometric or stylometric output as training data for an AI meant to replicate me. Or if I've been cryopreserved and need to be reanimated. All of them are far preferable to utter information theoretic death. It's a minor consolation to me that things I've written and described have ended up in the training set of future LLMs and other AI.
If none of that applies, then I will resist, violently, an attempt to murder me or my clones.
will kill
*will try
I wish them the best against someone who has an ASI at their beck and call.
I am a very happy monkey when surrounded by the fruits of modernity.
About the earliest point in time I can see myself stomaching is maybe 7 or 8 years back, so I can whack myself on the head and pick a better med school, while not missing out too much on creature comforts like the internet and smartphones. My childhood is a story of being bored out of my mind, and seeking to remedy that, and it's only recently gotten much better.
In terms of potential demigod Singleton controllers:
Altman, Zuck, Musk, Dario, yada yada. Anyone who has a massive AI division or promising startup and also the leverage necessary to tweak the preferences of the incipient AGI within. Mere wealth, while nice to have, isn't nearly as good as being able to tell your high-powered ML PhDs what to do.
No, Dase, simply finding another Indian nerd with such similar personality traits is far from sufficient for me to consider him isomorphic to myself. I do not care that he loves his mother, I love mine. Certainly from your perspective you might well be indifferent between us, but I am merely me.
Like, is he closer to me than almost everyone else? Certainly. As are humans practically negligibly different in the space of All Possible Minds. Doesn't make them me.
Is that sufficient for him to be considered me? Not at all.
Leaving aside that what entities one identifies with is inherently subjective, I've proposed a reasonably robust criterion for determining that, at least to my satisfaction. You blackbox both of us, and assess response to a wide variety of relevant stimuli. If the variability between us is within acceptable parameters, such as being less than the variability seen in the biological me after a nap or when I took the test 2 years ago, then that system is close enough to count as including a copy of "me".
This accounts for. even mind uploads, hence the blackboxing, I don't particularly privilege my biological form, though you could do a DNA test and MRI if you really prefer to.
It might well count an emulation of me within a wider system, say a Superintelligence, but that's a feature and not a bug. That component, if it can be isolated, counts as me, leaving aside more practical concerns like what degree of power it has in the ensemble.
Moreover, if you get brain damage or dementia, your hardware and computational divergences will skyrocket, but you will insist on being a continuous (if diminished) person, and me and him will agree! It
I am tolerant of minor performance fluctuations, but a sufficient amount of brain damage or dementia? Then I consider myself gone, in most of the aspects I care about, even if the system is physically and temporally contiguous.
The primary reason I might still value further existence is-
-
Hopes that the damage can be mitigated with future advances, if not losslessly.
-
Until the damage gets really bad, that poor soul is still closer to me than anyone else.
But if it gets bad enough, I assure you I consider the core construct to be dead.
I am, incidentally, immune to this issue because I do not believe in computationalism or substrate independence
I have always found this a peculiar view, and certainly I haven't seen any particular reason to assume a difference in internal qualia because of a difference in substrate, as long as the algorithms deriving it are interchangeable in terms of inputs and outputs.
Is it possible? I can't rule it out. But the bulk of my probability mass is against it.
If you have a convincing argument otherwise, I'm curious to hear it.
I will reject this conclusion until the time I have an opportunity to get much smarter and reexamine the topic or perhaps design some fix for this pervasive mental defect
My current approach to modeling myself has enough practical ramification that I will accept it on an operational basis. Certainly I would love to re-examine it in more detail when it becomes more relevant, such as if I'm contemplating a mind upload and am either smarter myself or have an ASI to answer my questions.
But it reduces to normality for almost every situation I can expect to encounter today, so it's hardly the most pressing matter.
You are performing more or less identical calculations, on very similar hardware, to a near-identical result, and if you one day woke up, Zhuangzi style, to be him, your own life story a mere what-if distribution shift quickly fading into the morning air – I bet you would have only felt a tiny pinprick of nostalgia before going on with his business, not some profound identity crisis.
Believe it or not, I have often imagined, idly, having my consciousness magically transferred into the shell of someone I envy. The conclusion I have drawn is that there are some aspects of my life I would happily discard, if he's an accomplished banker (and I retain his skills and memories), I would happily not attempt to pursue medicine. But I would still prefer my original parents or kin, and attempt to convey my conundrum to them, likely by divulging privileged information only known to the original me.
If my "original" is still around? Inform him and work with him. I might be suitably disposed to help the "replaced" persons family and friends, but largely because they're predisposed to help me, assuming they don't know the truth.
After all, I expect and wish to continue preferring the consciousnesses descended from my own current kin even after we've all become post-biological, a mere swap of DNA carrier, while extremely queer and not entirely desirable, represents no major impediment.*
*The primary reason I am attached to my genes is because they code for people and personalities similar to mine. I couldn't care less about most phenotypic traits.
I chose the wrong profession. I certainly didn't chose the right country to be born in, not that I had a choice in the matter!
Are you in the Eleuther AI discord server? If not, I suggest you join, because I strongly expect you'll find a lot more in the way of practical/helpful suggestions there than you will here (though there are some people who can certainly assist).
I would murder any coworker who tormented me like this. Or at least steal a bite or two 🙏
Well, good thing we sorted that small matter out eh? And that's withholding my disbelief that this claim is true.*
Materialism has explained ever greater fractions of reality, and everything else has been losing out. I know where my priors lie. It's the mark of a deeply broken mind to update when the evidence "forces" you to, as opposed to when you should.
*oh and it's not.
Your statement is completely wrong and a willfully revisionist reframing of history. Galileo was wrong about his claim that there were two moons on the opposite side of Saturn, but at no point did he manage to lose the entire planet, which would be an impressive feat, given that it's visible with the naked eye.
His geocentric rivals certainly didn't do a better job at predicting its motions, and where he outright won was to apply heliocentrism to the phases of Venus and the moons of Jupiter, where geocentrism failed miserably.
Evidently this historical example, which you were wrong about, was loadbearing on your beliefs about the applicability of materialism. Given that I went to the effort of correcting you, am I to assume you're a bona-fide materialist now? Didn't think so.
Edit: Oh, the irony of using empirical investigation into material causes to judiciate this.
Why makes you so sure that "we don't know" isn't an acceptable answer? Certainly beats making shit up, which is how I would frame the non-materialist position.
drugs having an effect on experience
Drugs?
Brain damage?
Electrical stimulation?
If it can physically induce even minor changes to the brain, and that produces robustly consistent corollaries to conscious experience, it beats me what people are so intent on looking for.
If we are a meat-computer there is no dualism between the body and the mind.
Well, yes? The brain is an organ like any other.
Signed -A meat computer
people being born into the wrong body
Well I'm born into the "wrong" body since I'm not a 6'8" 269 IQ biologically immortal ubermensch, but to the extent that I'm vaguely sympathetic to trans people, it's because I recognize them as other people who are unhappy with the limitations of their current physical form and consider it laudable to adapt it to one's tastes, if possible.
As far as I'm concerned, unhappy with your body? Feel free to change it. Can't change it, (yet) like trans people and their biological sex? Nobody claimed that the universe is obliged to make it easy.

It seemed like the least bad option at the time, especially accounting for the (not entirely forlorn) hopes of moving to the US and getting a cut of those healthcare fees everyone loves complaining about 🙏.
More options
Context Copy link