I generally work on internal-use Python codebases. I would much rather do some basic validation and fail loudly if anything I didn’t account for happens.
One of the issues I run into is that agents are predisposed to write extremely defensive, odd-case-handling code, and that makes me recoil when I have to look into it: the SNR ratio is very low. You get a spaghetti that is unlikely to crash, but really hard to extract the gist of. And finding more global bugs can be difficult because what should be structural impossibilities can be coerced into silent skips.
The article made me think that maybe what I’d like is a language where validation doesn’t triple the lines of code you have to write.
If OP said the important thing is to be appropriately generous with your blamelessly earned wealth as you safeguard it against future perils, I would not have replied.
A society where the only people with stability are those who parasitize on a increasingly emaciated wealth-producing population surely doesn't seem like a stable society, nor one that would be pleasant to live in.
There's no true stability in the world, but if the only option for "safety" we see is to be one of the few who snatch self-perpetuating generational wealth, aren't we just speeding up the unraveling of the very system off of which we desire to subsist on?
Forgive me if I ramble for too long. I've been seeing a lot of comments in this vein and the thoughts have accumulated.
Tacit in your question is the notion that the inquiries that are important are those that can result in predictive models of phenomena encountered in the world — hence feelings, intentions & perceptions turn into a shorthand for reported accounts of the same — and that given enough reports (data), we could build a dictionary that maps a bundle of reports to a(n equivalence class) of physical system(s).
But when we speak of having feelings, or acting on intentions, most often we are not using these as stand-ins for our failure to pin down the current state of our physical system to another. If I am exposed to fire, I want to get away — I am unconcerned with how well I could translate my report of the pain to a patterns of neural activations. The reality of pain for me is unaffected by the fidelity of my "experience report dictionary". And it is there whether it's a brush fire or a neuralink streaming fire bits to my cortex.
If you decide that primacy ought to always be given to things as they can be modeled, you can choose to elevate the "experience report dictionary" and make the reality of experience a second-class citizen. Then you end up with an eliminativist ontology where indeed, we can rightly be called a mechanism.
But that is a "world-making" decision, a value judgement: "this is how things should be seen". It might be sponsored by our recent history, where we got high on the fruits of applied scientific modelling, nursed by the education which taught us that being a good engineer can have us continue in line with that, and pushed on us by impoverished modern eschatologies promising eternal youth, experience machines and what-not at this point. And it might seem preferable or more dependable than whatever equally impoverished, inhumane eschatologies we may have been presented with before.
It doesn't mean there isn't a whole world of places where we can go instead. But in general, we don't change our value judgements until the current one seems inadequate for some reason.
> If we created a molecule by molecule synthesis of a human being, you'd agree it is conscious and the same thing as a human created via typical reproduction, right?
If the question is what is there about us that's not covered by the body, we can mention things like: feelings, intentions, perceptions, acts of consciousness.
Or however else you want to divide up things that have to do with the mind.
Eliminativists/illusionists may completely deny such things. The rest can fall into many camps, some of them religious.
It's not like there are any surprising new parts. It's about how one chooses to interpret/conceive those we are familiar with.
> Which leads me to conclude that it wouldn't necessarily matter whether an AI like SkyNet has subjective awareness or not. What matters is what kind of long-term plans it could pull off and how much it could reshape the world.
In discussions like this, we're always going to bottom out at certain assumptions we bring with us, so I agree.
One reason I like bringing up examples like this (the xkcd in sister reply is also good) is that it makes really visible what our assumptions are. The scales are big both in space and time in order to emphasize what weight is given to functional equivalence.
I feel pretty confident most people wouldn't presume that doing a bunch of math by hand on paper can create glacial ephiphenomenal experiences (though I like the term).
Another thing that's interesting to me is that the converse assumption, i.e. one with a strong allegiance to functionalism, ends up feeling far more idealistic than you might expect. A box of gas, left on its own for long enough, will engage in a pattern of collisions that in a certain interpretative framework correspond to an LLM forward pass. In another, it can be a game of minesweeper.
The individual particles of course, couldn't care less whether you see them as part of one or the other. Yet your ability to see them in light of the first one is perhaps enough for the lights to truly turn on, if transiently, in some mind somewhere.
When we speak of the “despair vectors”, we speak of patterns in the algorithm we can tweak that correspond to output that we recognize as despairing language.
You could implement the forward pass of an LLM with pen & paper given enough people and enough time, and collate the results into the same generated text that a GPU cluster would produce. You could then ask the humans to modulate the despair vector during their calculations, and collate the results into more or less despairing variants of the text.
I trust none of us would presume that the decentralized labor of pen & paper calculations somehow instantiated a “psychology” in the sense of a mind experiencing various levels of despair — such as might be needed to consider something a sentient being who might experience pleasure and pain.
However, to your point, I do think that there is an ethics to working with agents, in the same sense that there is an ethics of how you should hold yourself in general. You don’t want to — in a burst of anger — throw your hammer because you cannot figure out how to put together a piece of furniture. It reinforces unpleasant, negative patterns in yourself, doesn’t lead to your goal (a nice piece of furniture), doesn’t look good to others (or you, once you’ve cooled off), and might actually cause physical damage in the process.
With agents, it’s much easier to break into demeaning, cruel speech, perhaps exactly because you might feel justified they’re not landing on anyone’s ears. But you still reinforce patterns that you wouldn’t want to see in yourself and others, and quite possibly might leak into your words aimed at ears who might actually suffer for it. In that sense, it’s not that different from fantasizing about being cruel to imaginary interlocutors.
Fortunately, as you mention in your last sentence, stress is introspectable.
How exactly stress corresponds to biomarkers doesn’t matter if your desire is to lower it.
The issue is that many of us don’t pay attention to how we keep our body & mind throughout the day, or do so on a very superficial level. So strain on the body can accumulate for a long time.
“Stress management” is a lifetime skill. It doesn’t come in bulletpoints, it’s as broad as “living happily”.
Edit: That said, this can make the advice “be less stressed” a bit vacuous.
But people do get scared when random health issues flare up and become more conscious of how they deal with stress in life.
So it’s not bad to keep reminding people either :)
It would be really cool if it could highlight the parts of the speech that gave you away your accent. It guesses mine correctly most of the time (though not the first time I tried), but also lets me know my accent is pretty light.
What strikes me as interesting about the idea that there is a class of computations that, however implemented, would result in consciousness, is that is is in some way really idealistic.
There's no unique way to implement a computation, and there's no single way to interpret what computation is even happening in a given system. The notion of what some physical system is computing always requires an interpretation on part of the observer of said system.
You could implement a simulation of the human body on common x86-64 hardware, water pistons, or a fleet of spaceships exchanging sticky notes between colonies in different parts of the galaxy.
None of these scenarios physically resemble each other, yet a human can draw a functional equivalence by interpreting them in a particular way. If consciousness is a result of functional equivalence to some known conscious standard (i.e. alive human being), then there is nothing materially grounding it, other than the possibility of being interpreted in a particular way. Random events in nature, without any human intercession, could be construed as a veritable moment of understanding French or feeling heartbreak, on the basis of being able to draw an equivalence to a computation surmised from a conscious standard.
When I think along these lines, it easy to sympathize with the criticism of functionalism a la Chinese Room.
What I find a practical, related advice is “If you want to get good at something, you have to make yourself glad that you’re doing it.”
This involves reminding yourself why it is that you want to get better at it, perceiving the process of learning as an interesting challenge, and in general generating interest.
There is a lot of creativity in how you actually do this. It is a skill in itself, and a very useful one, especially for skills where you find yourself lacking patience and motivation.
One of the issues I run into is that agents are predisposed to write extremely defensive, odd-case-handling code, and that makes me recoil when I have to look into it: the SNR ratio is very low. You get a spaghetti that is unlikely to crash, but really hard to extract the gist of. And finding more global bugs can be difficult because what should be structural impossibilities can be coerced into silent skips.
The article made me think that maybe what I’d like is a language where validation doesn’t triple the lines of code you have to write.