Blogs & Articles: !!!Is AGI around the corner??? š 52 weeks ago
- Category: Blogs & Articles | Aleksandar Svetski on Medium
- Author(s): Aleksandar Svetski
- Published: 21st April 2023 13:56
Or is everyone just deeply confused?
This article was originally published on my Substack here
Subscribe there for more & tip someĀ Sats!
Forgive the click-bait titleāāāI had to do it. In this short essay, Iāll present a short case for why AGI is a red-herring wrapped in modern hysteria.
Part of how Iāll do this is by exploring what is meant by āGeneral Intelligence,ā why true āConsciousnessā exists beyond the realm of pure computation, and that most AI people are just deeply confused.
It wonāt be an exhaustive argument, but the first of a series of essays Iāll be writing on this topic to (a) help elucidate this viewpoint further, and (b) tell you something more useful than the kind of cringe that 99% of Ai newsletters are producing.
Blind from theĀ start
When most people talk about AGI, or even AI, theyāre completely unclear on whether theyāre conceptualising some form of Intelligent entity or a new sentient or Conscious being.
In fact, I donāt think they even realise there is a difference between the two. To make matters more confusing, there is very little consensus on whatās mean by intelligence in the first place, which alone is multi-variantāāālet alone theĀ latter.
And lastly, assuming they do mean Intelligence, then nobody is clear on specifically what theyāre afraid of. And if itās something more āconsciousā that theyāre envisioning, then Iāll argue there is very little for us to be concerned about (for two reasons).
Letās exploreā¦
Intelligence
To begin with, letās look at Intelligence.
Not only are there many forms of it, but Intelligence itself is nested within the broader concept of cognition, which in turn is nested well inside sentience and ultimately consciousness.
Weāll explore consciousness next, and more in subsequent essays. For now start with the incredibly diverse thing that is Intelligence.
What is it? When you begin to dig into this question, you find that itās more complicated than just āproblem solving, pattern recognition, language or learningā. To define intelligence, you need to actually include the kinds of intelligence(s).
In humans for example, we have of course the Braināāāwhich computational reductionists believe is the mechanistic hardware for the software of mind. It is split into left & right hemispheres each endowed with their own kind of sub-intelligence. The right hemisphere is intuitive and tends to provide immediate, holistic gestalts and insights that are difficult to articulate. The left hemisphere specializes in systematic, step-by-step reasoning, can only see the parts, not the whole, and is focused more on proofs and dealing with the tangible. When they work together, typically, the right hemisphere generates a solution, and the left hemisphere verifies that solution by analyzing the underlying relationships between its constituent parts.
You then have this emergent property where their combination is an intelligence thatās greater than the sum of their parts (I recognise thatās hard to measure).
Then of course, connected to the brain, we have the rest of the central nervous system (CNS), along with the peripheral nervous system (PNS), each with their own complex and intelligent network of cells that allow for communication and coordination throughout the body. The CNS consists of the brain and spinal cord, which work together to process information and generate responses to stimuli. The PNS consists of nerves that connect the CNS to the rest of the body, allowing for the transmission of sensory and motorĀ signals.
These are then related to the broader intelligence of the body. We have neuronās all throughout or bodies, even in our digestive tracts and muscles, and while we * think* this is where body intelligence comes from, because we associate Neuronās alone to thinking, the truth may even be deeper. It may be that body intelligence is more fundamental and integral in how it works. Either way, itās also a unique and interrelated form of the broader general intelligence that makes upĀ man.
āThere is more wisdom in your body than in your deepest philosophy.ā
- A twitter frog (unknown)
Then of course we have emotional intelligence, instinct and intuition, all which are properties or āintelligencesā emergent from or of the interrelation between the nervous system, brain, body, and experiences with our environments.
These are all unique intelligences, intertwined and interrelated to make something more complex, ie; a human being. They ultimately reflect a unique way of understanding and interacting with the world, and together form the ever-enigmatic thing we may like to call theĀ āmind.ā
While some like to think of the mind as the software program running on the hardware that is the brain, in my opinion, the reality is obviously far more complex. As is clear, the brain is not alone when it comes to thinking. Furthermore, while there is influence from the brain on what we think/feel/perceive, the awareness of self, the individual agency and the interconnectedness of it all these intelligences makes for a much more complex and emergent phenomenon than a program running on hardware.
Embodied cognition advocates such as John Searle, attempt to define intelligence and consciousness as something that emerges from the body, and at the very least is a more holistic view that tries to take into account the complexity described above. You also see it in the writings of polymaths like Oswald Spengler who tells us that it is not our brains but our hands that make us the supreme animal species. With our hands we can hunt, we can design weapons, write ballads, climb mountains, sketch portraits and manipulate the environment aroundĀ us.
According to Spengler, the unparalleled dexterity of this unique appendage created a feedback loop between it, the environment and the brain that bootstrapped a higher order intelligence unlike anything else on theĀ planet.
All this is to say that any definition of āIntelligenceā by your run of the mill AI or AGI enthusiast, especially the hysterics like Yudkowsky, are at best incomplete and more often ignorant. They have computational experience that they project onto humanity, and because they fail to understand anything beyond this, they fall into hysteria traps, proving they have little to no idea what theyāre talkingĀ about.
If your fundamental position on intelligence is merely computational, and you lack the understanding that there are various kinds of intelligences, then your position on AGI is moot. Even more-so if youāve not dis-entangled it from Consciousness.
To date, weāve all largely perceived computer intelligence as a left-brain-like apparatus, and the āIā in Ai or AGI as a tool adept at pattern recognition, problem-solving, reason, adaptation and rational thinking. More recently (and accidentally might I add) weāve found seemingly right-brain abilities like creativity and languageāāāwhich mind you were thought to be AI complete problems to solveāāāare possible. GPT and Diffusion are case in point, although in reality it turns out these are more akin to phantom creativity and cognition in which probabilities are used to make the illusion seem real. Of course, I could be wrong here and there might be some deeper emergent properties, but I think weāre still largely in the dark on all ofĀ this.
Weāre more like monkeys with a keyboard, randomly writing Shakespeare. Weāre playing with probability machines and by anthropomorphising the outputs, we think weāve discovered the makings of a conscious agent on the other end of theĀ line.
And the funny part is that this erroneous conclusion is what belies the fears that people actually have with respect to AGI, ie; that itās some sort of Intelligent Agent with a will of itās own. Thatās is āconsciousā.
Which brings us to the next error in judgement.
Consciousness
Does intelligence stem from consciousness or does consciousness emerge from intelligence? This is obviously a hard question to answer, and one Iāll try explore in greater detail in a futureĀ essay.
For now, as far as I can tell, Consciousness is something much larger and more complex than Intelligence alone. It encompasses all of the kinds of intelligence we discussed, but also a sense of self, a will or source of intent, agency and a subjective experience.
Metaphysically, it is the ineffable. The mysterious phenomenon of something higher that we humans beings are connected to. Without sounding too much like a hippie, Human Consciousness seems to be connected to some sort of Grander Consciousness or āSourceā orĀ God.
The mind seems to be the thing that contains (?) or taps into our consciousness and bridges it with this broader consciousness.
Understanding it is a still in its infancy, and there are of course various viewpoints. The reductionists like to think that the computational theory of mind explains it and everything else. And while computational processes have produced specific, narrow intelligence and problem-solving abilities (credit where itās due) they are far from conscious.
I mentioned John Searle earlier. His position is that āembodied cognitionā is the crucial aspect of consciousness. Our body and environment shape our conscious experience, and that feedback loop is non-existent in the raw computational approach, and thatās why it cannot be fully replicated by a computational system.
Julia Mossbridge goes deeper (or higherāāāwhichever metaphor your prefer) and posits that everything is downstream of Consciousness itself. That the phenomenon of consciousness not only involves non-linear interactions between the brain, body, and environment, but that the nature of it is non-local and therefore the subjective experience of consciousness cannot be reduced to objective information processing.
Thereās a parallel here in the physicist, Sir Roger Penroseās work. He argues that consciousness is fundamental to the structure of the universe and not abstract or computational in nature. In his view, consciousness is a process with quantum origins and to properly grasp it will require the discovery of fundamental physical principles beyond our current understanding.
These are both in some ways related to the elephant in the room, ie; the theological argument. If we take this viewpoint, consciousness is not just a byproduct of brain activity, or even embodied cognition for that matter, but a fundamental aspect of the universe, unique to humanity and imbued by a higher power or divineĀ force.
Erich Neumannās work on the archetypal origins of consciousness also point to the human psyche having a spiritual and divine dimension that cannot be fully understood by scientific or computational approaches alone. In Neumannās view, consciousness is not something that is created or produced by the brain, but rather an inherent aspect of the human psyche that is rooted in our connection to theĀ divine.
These are of course only a few of the many more examples and anecdotes I could draw from, so suffice to say at this point that: Consciousness is a much larger, higher and more complex phenomenon than we can currently fathom, let alone initiate.
In fact, thinking that that we have somehow become or created āGodā because weāve got high powered compute and strong probability engines, is arrogant at the very least, hubris at best and border line derangement at worst. Iām sorry lads, but weāre still playing with sophisticated sentence completion. Donāt get ahead of yourself.
Digital Ghost
At the end of the day, the fears and hysteria often seen in Ai circles stems from their conceptualisation of AGI as some sort of conscious being, in which case, theyāre conflating Intelligence with something muchĀ broader.
Weāve established that Consciousness is too complex to be understood, let alone be emulated by computational processes alone, so while AI (or even AGI) may simulate human-like behavior, and exhibit really general āintelligenceā one day, Consciousness is an entirely different matter.
I donāt believe that AGI in the conscious sense is anywhere near occurring, no matter how much compute we throw at it, or how many sentence structures it can produce that emulate human language.
In the nearer term, (years or decades), the only real point of concern is how high-powered-narrowly-intelligent machines, which can emulate elements of human and super-human intelligenceāāāare used in society. Like any tool, the problem is with the people using them, and in this case, how these tools are positioned to function. If you thought TSA-like NPCs were bad, wait until youāre interacting with a computer-simulated NPC that can hold a conversation but has about as much agency as a Rumba vacuum cleaner, and the power to shut off your CBDC account, passport or car engine. Thatās the real problem or āthreatā. All this AGI fear is a red-herring IMO.
Human-like āGeneral Intelligenceā, is far more than just language, problem solving, pattern recognition or other computationally reducible element. Donāt get me wrongāāāIām aware that there are incredible emergent properties exhibited by Neural Nets, LLMs and the like, but these are all still cerebral inĀ nature.
There is body, emotional, nervous system and a complex array of intelligences at play in the human that we havenāt even begun to work out how to model. Not to mention the the spirit, the soul, the ineffableāāāeach which moderns like to ignore or pretend arenātĀ real.
Thereās a long way to go, so (a) donāt get distracted by hysterics, and (b) keep building. I think that because this is the first time weāve actually āspokenā to something other than a human in a somewhat linguistically coherent fashion, weāre quick to anthropomorphise. We project our human-ness onto it and confuse computation and probability with lifeā¦well some do atĀ least..
Closing
Iāve rambled on enough. I hope thereās been some thread of logicĀ here.
Even if we some day achieve a truly broad form of AGI, via some blend of technologies, or a collection of functional and more narrow artificial intelligences, as Iāve said, consciousness is another matter altogether.
The real risk is its use as a tool by those who seek to control others. Control freaks are the truly weak, and itās this class of human that has always been the greatest threat to flourishing and freedom. Weāve had a taste of this madness over the last few years and if weāre not vigilant and taking defensive measures, thereās no telling what sort of stupidity weak people with powerful tools willĀ unleash.
Iād also like to note that IF I am wrong, and an AGI does somehow becomes conscious (never say never), then I believe that such a strong, sentient being would be an extension of life, not an ender ofĀ it.
True strength and power is magnanimous and I fail to see how a being with greater intelligence (already a complex thing) and a will of its own, will emulate the weakest of behaviours and characteristics (control freaking and eradication).
That which is powerful doesnāt seek to stand on top of others, but to stand on its own. Power seeks collaboration and competition. Weakness seeks subjugation.
Thereās also what Iāll call the āSingleton fallacyā. Assuming that we get true, general intelligence, I do not believe that there will be just āOneā. Life, and especially something as complex as intelligence, is multi-variant. There already are many approaches to it and coupled with the fact that many intelligences exist in general, there will be competition. Doing it āallā is a way to do it all poorly. Thermodynamics exist and cost is an inescapable factor. In fact, if there was a supercomputer that could do it āall,ā it already exists. Its called planet earth with humans onĀ it.
Before I go off on more tangents, let me conclude by saying that I do not think that AGI is close and I do not believe weāre at risk of sentience spontaneously emerging from probability machines like Stable Diffusion or GPT any timeĀ soon.
The fear-mongering around it stems from weak hysterics that lack a conception of strength while project their own fears onto things they barely understand (consciousness and intelligence), and if there is anything more sinister going on, its people who want to regulate or see it regulated for āour safetyā (Iāve heard that before) which in fact will just create an imbalance to the access of theseĀ tools.
Yes there are risks and threats from broad-enough and powerful enough artificial narrow intelligence or specialised intelligencesāāāand yes that includes LLMs and other neural nets, and their use in the innumerable applications around the worldāāābut the key is not to regulate these things into oblivion or create access imbalance.
Let humans do what they do best. Leverage tools to make life more effective and more efficient (ideally, without becoming a slave or dependant on the toolāāāanother topic for a futureĀ essay).
Thatās at least what Iām going to try and do. Iāll say more in the coming weeks on this, so stayĀ tuned.
If you though this essay was useful, please share itĀ around.
Alsoāāāgive the short thread version a like here on Twitter and onĀ Memoād.
https://memod.com/SvetskiWrites/is-agi-around-the-corner-6188
Iāve submitted the Memo to a thread competition, so Iād appreciate you going there, signing in and giving it a like. Memoād is like Twitter meets pinterest. So bite-sized knowledge w/o theĀ noise.
Thanks inĀ advance.
See you on the nextĀ one.
Originally published at https://authenticintelligence.substack.com.
If you like this work, you can tip someĀ Sats: