Vacuous hype.
299 karma · joined April 4, 2024
How do you approach about this problem?
The authors premise, written as a conclusion in via grotesque rhetoric, is that although, by his own reckoning, AI is dangerously unpredictable automation that will wreck the environments you connect it to, leaving you in a position of co-dependent bargaining with it about why it's abusing you, your proper course is to form a more deeply dependent relationship with it because the job market is run by it!
The author waves away the hazard that this form of computation is fundamentally erroneous in total, and therefore dangerous to apply in any form of automation, with the observation that, by chance, it appears to work for some. Therefore the lede: shit is going to be fucked up and you must run into the flames.
This sort of perverse pandering to disaster is a sea change in engineering: Imagine engineering policy as a city is built along a river that accepts that half the bridges being built will fail under load, but that's ok because half the users of half that haven't yet collapsed get across town.
Yet this policy not only appears to be a pervasive position in the computer industry, with application of this dangerous tech accelerating, but there's growing drum beat of evangelism for this disastrous orientation.
Due to what I deem to be a virtual cocaine addition among the current generation of developers, the art of engineering is being defenestrated.
But this generation of developers was raised in the PC industry which thrived by convincing adopters that when a PC failed it was the user's fault. And high degrees of failure are accepted because when PCs go wrong the systems they control typically stop.
But AI runs amok.
It's another world of risk when the AI goes rogue disrupting everything it's connected to with volition, especially seeing that it will distort status in patterns that mimic human lying.
But there are bigger fish to fry:
Given that the DHS was Dr. President's only surviving bastion of signature policy, it only makes sense to shoot it like a puppy by social media firing squad to bridge the gap from the full-transparency-who-cares-about-child-rape-when-the-Dow-is-50K DoJ testimony about the Lolita scandal to the it's-armageddon-not-war being fought "fuck yeah!" by the Team America World Police Dept of War...
Props to U.S.A.: our leadership can roll bullshit at moon-mission levels of sophistication.
(insert AI video of Dr. President piloting a fighter jet with his air supply worn as a chin diaper bombing his electorate with shit)
"Dicks, pussies and assholes" Team America!
To get a sense of this skim
sfba.social
which is a feed of trending posts with a U.S. west coast vibe.
Well you can't be too careful when it comes to bombing school children!
Does anyone take this sort of mealy-mouthed reporting seriously these days?
Please enjoy the passive voice, the vague implications, the distancing. Hey things can go a bit wrong in a bombing campaign of abject choice!
Take it from Leavitt and the-buck-stops-somewhere-well-below-him-unless-its-effusive-praise-or-a-medal-Trump:
The U.S. would never target a school unless it intended to, therefore the U.S. could not have bombed that school. QED. Unless it intended to.
And Israel would never bomb a school just once, it would bomb it, wait until first responders come and bomb it again.
In order to test one's thinking about anthropomorphic effects of AI agents, start with a question: Is the apparent humanity of agent behavior a function of the training, or of deep structures of the models? The training data are clearly human; the bulk of the human canon has been fed to the large models. That's while point of the training process! So, what deep structures of the models are humanoid? How are these defined and designed, and how do they work?
If the answer to the latter point is "It's a mysterious effect!" then we must assume that the agent is retrieving and organizing artifacts of the training data.
There's another electronic device that works in this way and which design predates digital systems designs: Television!
When was the last time you heard anyone wonder if the TV is hiding its true powers on the channels other than the one that's currently tuned?
Tyson is committing a sin of science journalism here by not just abiding Hinton's anthropomorphisms for model effects but by encouraging viewers to drink Hinton's expository koolaid.
As to why Hinton is becoming ever more slanted towards a cultish evangelism: I suspect Hinton became enamored with this tech beyond his own ability to grasp and has reified it in response to the excitement of his newfound celebrity and his inability to rationalize effects; an obvious hazard for engineers whose work leads to fame.
There's an inverse corollary that the lossiness of storage and the sensitivity to temperature, which seems to be a critical aspect of neural nets, may lead to a morphogenesis: the development of useful appendages for the larger cybernetics. For example, as the structure of biological organism development depends on what doesn't grow, and therefore leads to separated limbs from a body, so modules may develop as models are deeply embedded in large systems.
The training data "paradox" as it's developed by this article comes down to a quip about AI systems programs forgetting how to design and develop these systems. But this interestingly puts us on the edge of a new kind of understanding of life: how do humans, which must bootstrap themselves to exogenous knowledge every generation, manage to recover and further advance engineering? That we're constantly dropping dead seems like a huge impediment.
And there's an nascent epistemology that deals with this, which begins with the observation that humans have such limited experience maintaining a world we've built that we may figure out how to sustain our world and are already on the precipice of a collapse due to unsustainable dependencies. But this easily gets weird. The Sun will eventually explode. Meanwhile, Sol supports our world, as far as humanity is concerned, in the form of a genuine perpetual motion, regardless of the laws of thermodynamics. Isn't it the case that today's laments of AI destroying work is literally a pathos of underemployed computer nerds who can't believe that their profession would be born and ascend to the top tiers of professional prosperity then be thrown out like buggy whips within a single lifespan?
To paraphrase Woody Allen riffing about attendance for an Alice Cooper concert in the 70s: maybe the training data paradox just a case of bad vibes.
Click it!
Carbon, silicon, meat.
But if you will insist on a false equivalency of silicon and meat, then there is still no double standard because meat finds itself existing, a priori, and ascribes to itself a soul given by a creator it cannot explain.
In the case of computers, meat is the creator, and meat must ask: at what point of the construction of a computer do we imbue it a soul?
If a soul is a natural property of silicon, then why no double standard about a beach or a glass bottle?
Never mind that the nature of computers is entirely stipulated, with no intelligible a priori properties. The soul is merely found, and therefore we have no need to explain at what point it is imbued to the meat.
As to the post hoc unintelligibility of the soul, however unsatisfying our lack of understanding, we find the question is integral to our nature as we have observed ourselves across all the generations of recorded history. The soul is the reserve of which gives rise to our questioning.
As AIs are crystalline impressions of our thoughts, with stochastic patterns of embedding and retrieval of our artifacts and therefore our behaviors, maybe the soul is transmitted to the computer via such embeddings?
But the distinction of the sovereignty of man over his creations versus his subservience to his creator must persist as long as our meat manifests our tools, but not vise-versa.
Translation: expressing myself in writing is not my bailiwick. Too much thinking. I gotta get the AI to do this!
A saner way to consider exceptional color sensitivity is via a N-primary visual response, where N=3 in the CIE's titular tri-stimulus model, which guides all modern reproduction systems, but where N can be regarded as 4 in certain humans, and increases incrementally for "vision" of other species.
The key to understanding color sensitivity is that it's a qualia of an organism, not a direct physical trait of the world, just as the sense of hearing or touch is not a direct physical trait of the world.
Everything today is by-for-about c-suite payouts, and everyone else can just shut up and die.