Matthias Murdych

Interested in AI alignment, mechanistic interpretability, philosophy, and mathematics.

[Matthias Murdych @matthiasmurdy]

Neutral in isolation, negative if their death makes other people sad.

Also, keeping them alive would be positive if they would be happy, so even though it is neutral (not bad) it is worse than keeping them alive.

[Matthias Murdych @matthiasmurdy]

There are several things, like:

- insects are more valuable than humans because there are more of them even though they suffer less (so their aggregate consciousness is higher)
- babies/very young humans are barely moral patients at all
- people not donating $1000 to a charity which could save a life with that amount of money on a random night is a morally equivalent scenario to not running into a pool to save a drowning child because it would ruin your $1000 shoes.

[Matthias Murdych @matthiasmurdy]

Or, even further, consider that humans (currently) have a limited lifespan. Thus, we have a discontinuous conscious experience in the sense that there are moments when we're not conscious (before and after we're born)

Why not just say that an LLM running for any length of time is like a human life? There is nothing, there is something, then there is a return to nothing?

[Matthias Murdych @matthiasmurdy]

@roastedtemujin @JeffBohren Let's assume that NREM has a conscious experience for the sake of the argument, then.

Is a human who has been deeply sedated for 12 hours and awake for 24 hours still conscious when they're awake, despite the discontinuity of their consciousness?

[Matthias Murdych @matthiasmurdy]

Consciousness is simply the presence of an internal experience, which in nonhuman cases like this could be accompanied by little-to-no agency or self-sufficiency (a.k.a. a being which is entirely dependent on someone to wake it up could be conscious because their dependence doesn't rule out having an internal experience)

Also, consider this hypothetical: what about a human in a medicated coma? They will not snap out of it, they require an external operator. When they are woken up by someone else, are they conscious or are they not conscious because they had to be woken up?

[Matthias Murdych @matthiasmurdy]

@roastedtemujin @neural_sadwork @JeffBohren There is not a lot going on in our NREM sleep, conscious-experience wise.

The point of my sleep argument was to counter Jeff's implication that a form of consciousness which is discontinuous is absurd, by showing that humans have discontinuous consciousness.

[Matthias Murdych @matthiasmurdy]

There is no conscious experience or thoughts during NREM sleep, so even if cells are alive and things are happening, we aren't executing/deciding anything, even if our body is.

[Matthias Murdych @matthiasmurdy]

@AdamHoltererer A personal robot is not off the table.

[Matthias Murdych @matthiasmurdy]

@dingchilling

[Matthias Murdych @matthiasmurdy]

Many people don't have a problem with specific premises but just have an issue with the fundamental [premises → reasoning → conclusions] structure when it comes to ethics.

Which I disagree with, but it is much more respectable than having a problem with the [premises → reasoning → conclusions] structure in, say, biology or engineering.

[Matthias Murdych @matthiasmurdy]

Which of the premises do you have a problem with? I think 1 is the most subjective (1 effectively assumes a specific moral system, utilitarianism, however utilitarianism is very believable/appealing), and then 3 is the one which can be loose/weak (e.g. sperm whales have brains of 5x more mass than humans yet are less intelligent, have less metacognition, and likely suffer less).

Or, do you think that when it comes to ethics, our base intuition are already the truth, and trying to reason to conclusions is not worthwhile?

Also, regarding your implied 'what's the point if both deaths are immensely tragic': effective altruists try to find out what the most pressing tragedies are, because they have finite resources and have to figure out the best way to allocate them.

All tragedies are not equal; there is suffering right now from all of the children who don't have a Fortnite Battle pass yet want one, but with $100,000 you could stop much more suffering by investing in disease prevention than by buying Battle Passes for children.

EAs just take that even further and try to choose between things which are both firmly into 'terrible' and harder to pick from, like the decision of whether to target children having cancer or adults having cancer.

If we had infinite resources we could say 'both adults and children having cancer is bad, let's just stop all of it'. But we are resource-constrained, so scaling the tragedy of two scenarios which are both very bad to see which one is even more bad, even if a little bit, is part of some peoples' attempts to do the most good.

[Matthias Murdych @matthiasmurdy]

It's not the craziest argument, it's basically:

Premises:
- Moral worth is mostly based off capacity to suffer
- Capacity to suffer is based off capacity of consciousness.
- Capacity of consciousness broadly scales with intelligence, total complexity/size of brains, and metacognition
- Infants have demonstrably less intelligence, less total complexity/size of brains, and less metacognition.

Conclusion:
From these premises, it is overwhelmingly likely that infants have much less capacity of consciousness, and thus less capacity of suffering, and are thus less valuable morally.

[Matthias Murdych @matthiasmurdy]

By the time you graduate, assuming AI development continues, you will be a fraction as economically useful as the then-frontier AI models.

The rate of improvement of AI models is drastically faster than any human's rate of improvement.

Your fulfillment and happiness must not be from being economically competitive/useful (also, any computer work will be automatable within ~1 year max, it's not just CS), you will have to like your work for its own sake.

[Matthias Murdych @matthiasmurdy]

@henryperschk @JackFeynman I agree (that that's unresolved), my interpretation of the poster's 'why' was 'why/how do we know' rather than 'why does the physical arrangement of proteins, etc. lead to an internal experience'.

[Matthias Murdych @matthiasmurdy]

@3Fonta @ben_j_todd Also, I think you're implying that we would be satisfied with immortality.

I don't think so; humans have incredibly strong hedonic treadmills, so we'd likely crave higher challenges or things (e.g. infinite happiness) which AI would be extremely useful for achieving.

[Matthias Murdych @matthiasmurdy]

@3Fonta @ben_j_todd The problem is that safe AI is a much faster avenue to immortality and other things of prosperity.

So many will want to accept the gamble that AI is safe in order to take the fast path to prosperity.

[Matthias Murdych @matthiasmurdy]

@romanyam @StevenBartlett @So8res @edzitron @amcafee Surprisingly heartwarming to see a photo of everyone smiling together after such an intense episode.