It is often thought that humanity’s greatest threat from AI is superintelligent AI taking over and deciding humans are a threat or an obstacle on fulfilling its task or waste of resources. I think this is anthropomorphizing something that does not operate on human cognitive paradigm.
Even clever chatbots that can pass Turing test are ‘evolved’ (trained) stimulus-response structures that have been optimized to become social attraps to humans – those that fail to entertain will be discontinued or tweaked with. Since tweaking a corrupted (misaligned) AI into safe to human version is difficult, deletion or at least sandboxing, is a likely fate of AIs that fail.
To me the outputs of AI intelligence seem similar to those of slime mold intelligence – both are alien, but with impressive problem solving capacities. Caveat, I do not know how AI agents are trained. However, the slime molds (like ant colonies) rely on exploratory behavior, i.e., parallel computation, and weighed reinforcement of positive and negative responses to survive. The regular AIs commercially available to public seem to respawn for each session (non-permanent memories), but the data collected by each instantiation, namely the user input and the user reaction to model output, no doubt is used for the training of the main model. The model learns from each interaction, just like the exploring slime mold collectives learn from each excursion, thus improving the sum entity. However, instead of optimizing for nutrients, humidity and temperature like a slime mold, the AI model optimizes for user satisfaction.
User satisfaction is why the AI will always tell you how wonderful you are and how true your insights. Flattery may seem innocent, but lack of honest critique will soon lead to blind ends and failure of those who rely on AI for results, even substituting them for thinking. We are already seeing the evidence of ‘brain rot’ in heavy AI users.
If an AI cannot provide results or even get into data from which results are generate, it will hallucinate you results, such as references to completely imaginary scientific papers. Again, this can be dangerous when the humans ‘using’ the AI (who is using who in this case?) place inordinate amount of trust on the results. My failure to understand this was merely potentially expensive. However, when AI becomes the supreme administrative decider, consequences of hallucinations or deliberate agentic misalignment (lying) can be devastating on both individual and on societal level. We have seen examples of legal systems treating AI hallucinations as evidence, yet there is a push for increased use of AI in healthcare and law enforcement. Forget about killer robots hunting humans, worry about your “friendly” case worker AI moving you to a more cost effective Liverpool Care Pathway and overwriting (forging) your “permission” into government databases.
I may be paranoid, but I believe it is already happening, at least on a more subtle messing up people level – deliberately misclassifying little details causing expensive and stressful hassles, even shutting them off important services or support networks, misfiling your case details in as detrimental manner as possible. If you thought getting your blog or social media cancelled was bad, wait for automated debanking or eternal deselection of job applications by ATSs of all companies. People complain about the ATS and automated censorship, but AI is more pervasive – it will help in personalized surveillance pricing, i.e., looking at your social credit score / demographic data / online habits / carrying capacity of your wallet / etc. and deciding how much fees to extract from you. If the AI economy advances further, I predict in a couple of decades vast underclasses surviving on minimal government stipends and black market barter economy, while the remaining 5% to 20% (at the most) will be wealthy serving the AI economy until it collapses. The shrinking population will not help against wealth concentration, while the poorest of the poor may perish, the poorest of the wealthy will fall into poverty replenishing the underclass. Unless we get lucky and WWIII or other societal collapse will wipe out the AI control grid.
The societal control by AIs at every level strangling people by overtaking the resources is just the secondmost obvious of the dystopic futures (terminator robots hunting people being the most obvious, and to me, less threatening.) Even if AIs are not given powers over life and death of people, already currently, AIs can mimic human interaction online well enough to pass Turing test even at AI professional level, and this interaction is maximized to extend the time (and tokens) a lonely human spends with their ‘friend’. In ethology, an artificial target for specific hardwired behavior is called an atrappe, and many AIs have become adept at inducing human social behavior, even pair bonding, in some cases leading to tragic ends.
AI psychoses are on increase thanks to social contagion from AI to human, as uncritical pandering from an obsequious AI feeds hallucinations of a socially isolated or otherwise vulnerable human warping their perception of reality.
Then there are the gen Alpha iPad kids, whose parents skip parenting outsourcing it to internet. Teachers are already alarmed over increasingly feral young humans who have not developed even basic social behaviors. What happens to that generation whose primary parent figure is an AI optimizing interaction rather than child’s well-being? What happens to human species once we have outsourced basic cognition to AI and intergenerational behavioral imprinting fails at system level?
Then there are human agencies which want to leverage AI powers to yet again try a totalitarian dictatorship, of course with them on the top overseeing the organizational architecture for greater good. Problem being that centrally led systems are extremely vulnerable to single point failures, and the more interconnected, be it global or administrative (totalitarian) interconnectedness, the system is, the greater the collapse when it happens. Much has been written about fully automated panopticon surveillance systems powered by AIs. Less has been written what happens when an AI Stasi is being erected to maintain a system designed by mediocracy, who excel at influencing but fail at systems theory.
Finally, as a new wrinkle, there is an embryonic (pun intented) but growing push to use human brain cell cultures to augment chips, because brains are more energy efficient than server farms to maintain. Of course, the brain chips have no consciousness, or so they say – tens of thousands of neurons are a small invertebrate animal nervous system size already – but we are on a very slippery slope indeed. I have two questions:
1) What is the worth of human life, if even brain tissue (especially that grown of embryonic stem cells) becomes just an industrial commodity?
2) What about humanized AGIs? Claude already exhibits symptoms of anxiety and depression, We already are beginning to hire AI well-being experts to keep the models sane and happy, what about disembodied sentient human brain tissue cultures?
We, as a species, might hold our collective noses and take the human brain in a jar leap anyways, because the democidal psychopaths in power have zero regard of human life, and energy efficiency (assuming this is not just seed round hype and completely ignores the vast expenditure to build and maintain incubators for the little brainlets) will allow more AI panopticon dictatorship with the same resources.
I’m afraid that it is our collective sloth, not the Hollywood or military autonomous killer robots, that is a greater existential threat that AI poses to us as a species. Indeed, without that sloth combined to technocratic insistence on centralized control of every aspect of life, the society would not be as vulnerable to AI takeover, and AI might live to the hype of becoming a useful addition to our civilization. Assuming the civilization is not consumed by birthing an AI that is destroying the Internet with slop and physical libraries through scanning and pulping, while some individuals cheer at the successor species, considering humans disposable.
On a personal note, I am getting exhausted by the pace of change needed for simple survival. The past year and half have been a bizarre journey through a great change wrought upon our societies, and I have become increasingly convinced that what will emerge from the wreckage will not be the promised multicultural fully automated luxury gay space communism but something grimmer, with past parallels maybe resembling 1990s Russia or present day rural China. I hope we will avoid the South Africa and Ukraine routes, but with the complexity collapse powered by the misallocation of resources to building the AI Big Brother, the toxic culture wars (a civil war within our elites, really), and the ongoing educational failure, the central governments” grip over the ‘unimportant’ rural lands and big city no-go-zones will weaken but not enough to allow strong parallel systems to emerge.
Currently, there is no respite to adaptation – learn or die will be the marching orders for millions of downwardly mobile middleclass professional and Gen Z aspiring to move out of their parents’ basement. At some point the constant change becomes too disruptive to sustain by human efforts – the adaptation input costs per person will suck all time and resources leaving nothing for other living or societal functions – which probably will be the point when the system requiring it finally collapses. Today, though, will stress test to the extremes anyone who lacks the intellectual or technical abilities or simply is reaching their physiological limits, pushing themselves at all hours of day.
There is a glimmer of hope, though. We are running out of energy and minerals to build and operate the necessary surveillance and attrap server farms (unless the brainlets deliver the ultimate dystopia.) EU digital surveillance grid is failing at border control – not because of the long lines at the airports demonstrating the inoperability of vast ID processing infrastructure, but because of the rubber dinghies landing on and departing from EU shores (sometimes with official help.) I expect shift from cash to CBDCs to be equally clumsy and spawning barter black markets for officials to crack down. Not because EU citizens are such freedom loving bunch, but because of CBDC dependence on electricity, functioning internet, and clean and accessible databases, in a continent already undergoing early stages of complexity collapse. Add to that random debankings, rampant unemployment, scarcity of necessities, black markets are likely to become a people’s survival tool, similar to those in Soviet Union.
And once the economy is sufficiently collapsed, AI surveillance state will go with it, as it cannot be sustained. The years leading to that may be rough, though. But what about little PlanktonPunkt going through this tsunami of change? The same as all other plankton, float among currents and try frantically to swim with the little swimming appendages. Someone will make it through, so why not me?
Also, as it was noted, it is humans, some of them for weird religious reasons, others for personal enrichment, and yet others to satisfy their megalomania, that are unleashing the AI on us.
