Thinking specifically about AI here: if a process does not give a consistent or predictable output (and cannot reliably replace work done by humans) then can it really be considered “automation”?
and cannot reliably replace work done by humans
See, that’s the crux of it for me. Something with stochastic elements could totally count as automation, but it has to actually replaces some manual work.
LLMs could be made deterministic, by the way. They just produce the nth best token sometimes instead of the 1st best for the sake of naturalness.
LLMs could be made deterministic
Good reminder that LLM output could be made deterministic!
Though correct me if I’m wrong, their training is, with few exceptions, very much going to be stochastic? Ofc it’s not an actual requirement, but under real world efficiency & resource constraints, it’s very very often going to be stochastic?
Personally, I’m not sure I’d argue automation can’t be stochastic. But either way, OP asks a good question for us to ponder! The short answer imo: it depends what you mean by “automation” :)
Uhh, that actually raises some questions about the definition of determinism one is using. If the order in which it sees training material is generated by a single known seed, for example, does that count? What if it’s a really bad RNG algorithm, or literally just a complex but obvious pattern?
They are a black box, so in a sense the way they’re constructed might as well be totally random.
good points on the training order!
i was mostly thinking of intentionally introduced stochastic processes during training, eg. quantisation noise which is pretty broadband when uncorrelated, and even correlated from real-world datasets will inevitably contain non-determinism, though some contraints re. language “rules” could possibly shape that in interesting ways for LLMs.
and especially the use of stochastic functions for convergence & stochastic rounding in quantisation etc. not to mention intentionally introduced randomisation in training set augmentation. so i think for most purposes, and with few exceptions they are mathematically definable as stochastic processes.
where that overlaps with true theoretical determinism certainly becomes fuzzy without an exact context. afaict most kernel backed random seeds on x86 since 2015 with the RDSEED instruction, will have an asynchronous thermal noise based NIST 800-90B approved entropy source within the silicon and a NIST 800-90C Non-deterministic Random Bit Generator (NRBG).
on other more probable architectures (GPU/TPU) I think that is going to be alot rarer and from a cryptographic perspective hardware implementations of even stochastic rounding are going to be a deterministic circuit under the hood for a while yet.
but given the combination of overwhelming complexity, trade secrets and classical high entropy sources, I think most serious attempts at formal proofs would have to resign to stochastic terms in their formulation for some time yet.
there may be some very specific and non-general exceptions, and i do believe this is going to change in the future as both extremes (highly formal AI models, and non-deterministic hardware backed instructions) are further developed. and ofc overcoming the computational resource hurdles for training could lead to relaxing some of the current practical requirements for stochastic processes during training.
this is ofc only afaict, i don’t work in LLM field.
In practice there really is no incentive to avoid stochastic or pseudorandom elements, so don’t hold your breath, haha. It’s a pretty academic question if you could theoretically train an LLM without any randomness.
Thanks for writing that up, I learned a few things.
Consider this example:
You have a road that forks and joins up again. You need to reach the end of this road and have a vehicle that takes you there without your input. At the fork, it will flip a coin and choose to either take the left fork or the right fork depending on the results. This agent is therefore stochastic. But no matter what it chooses, it’ll end up at the same place at the same time. Do you consider this to be automation?This is a crisp answer, nice one.
Automation is just using technology to replace human labor, so yes. The exact mechanism doesn’t change that. “AI” is a buzzword but LLMs have replaced human labor already in various ways even though most of the applications are hype / BS. For example, it has certainly taken a bite out of stock images and product graphic design.
Individual capitalists must seek out automation because reducing labor cost without decreasing productivity means a higher profit for them. Capital in aggregate seeks automation because it disciplines labor, means you can threaten and mistreat labor more easily. In that sense “AI” is serving the same purpose as historical automation even when it fails to substitute labor as a productive aspect. Companies can threaten their employees with “AI” that doesn’t work and they can rebrand firings as layoffs using media discourse that overhypes “AI” on their behalf, it is part of the PR universe.
If it’s “good enough” for the task, yes.
Many tasks have loose success parameters, and an acceptable failure rate. If the automation fits in those, and it simplifies my day, then it’s reasonable automation.
Depends on the use-case. AI isn’t a panacea, and the excessively pro-AI camp is deeply unserious, but it does have some cases it can function fairly well at, like stock image creation, that doesn’t need to have backdrops, props, actors, etc for every random idea. That’s the extent of it, really.
I mean, it can’t really do ‘every random idea’ though, right? Any output is limited to the way the system was trained to approximate certain stylistic and aesthetic features of imagery. For example, the banner image here follows a stereotypically AI-type texture, lighting, etc. This shows us that the system has at least as much control as the user.
In other words: it is incredibly easy to spot AI-generated imagery, so if the output is obviously AI, then can we really say that the AI generated a “stock image”, or did it generate something different in kind?
“Every random idea” meaning AI can take the place of some stock photos, and not all, as in we don’t need to do the traditional stock photo process for every random idea, AI can replace some of them. As for the quality of the output, that’s something that varies from case to case, and further the idea isn’t to replace human art in general, but to exist alongside in instances where a human artisinally producing it isn’t the purpose, but the traditional means to an end. Therefore, it doesn’t actually matter if we can tell or not, the goal isn’t to decieve, but even that is getting blurrier and blurrier as AI improves.
Essentially, if an AI image can fulfill the same purpose as a stock image, then the act of creating the stock image through traditional means is just unnecessary expenditure of effort. We don’t traditionally appreciate stock images for their artistic merit, but for a visual function, be it to convey information or otherwise, not because our goal is to appreciate and understand the artistic process a human went through to create it.
If you can tell it was produced in a certain way by the way it looks, then that means it cannot be materially equivalent to the non-AI stock image, no?
These are distinct hypotheticals.
In the first case, the question is if it is equivalent, does the use-value change? The answer is no.
In the second case, the question is “if we can tell, does it matter?” And the answer is yes in some cases, no in others. If the reason we want a painting is for its artisinal creation, but it turns out it was AI-generated, then this fundamentally cannot satisfy the use of an image for its appretiation due to artisinally being generated. If the reason we want an image is to convey an idea, such that it would be faster, easier, and higher quality than an amateur sketch, but in no way needs to be appreciated for its artisinal creation, then it does not matter if we can tell or not.
Another way of looking at it is a mass-produced chair vs a hand-crafted one. If I want a chair that lets me sit, then it doesn’t matter to me which chair I have, both are equivalent in that they both satisfy the same need. If I have a specific vision and a desire for the chair as it exists artisinally, say, by being created in a historical way, then they cannot be equivalent use-values for me.
This argument strikes me as a tautology. “If we don’t care if it’s different, then it doesn’t matter to us”.
But that ship has sailed. We do care.
We care because the use of AI says something about our view of ourselves as human beings. We care because these systems represent a new serfdom in so many ways. We care because AI is flooding our information environment with slop and enabling fascism.
And I don’t believe it’s possible for us to go back to a state of not-caring about whether or not something is AI-generated. Like it or not, ideas and symbols matter.
“We” in this moment is you, right now. If the end product is the same, then it is the same. If the process is the use-value then it matters, but if not, it doesn’t.
Ideas and symbols matter, sure, but not because of any metaphysical value you ascribe them, but the ideas they convey.
First you said “it doesn’t matter if we can tell or not”, which I responded to.
So I’m confused by your reply here.
NNs do give a consistent output. If you put the same inputs and seed in you get the exact same output every time.
The algorithm represented by the NN is fully deterministic, but way too large for a human to wrap their head around it.
All machines are automation, i think?
Ask marketing. They will tell you that everything is AI.