Artificial Intelligence (AI), Consciousness, and Existential Risk: the Tantric View
Vložit
- čas přidán 5. 09. 2024
- You might have noticed that AI is very much in the news lately. In some quarters there is concern that we could, at some future point, "lose control" of AI and it could pose an existential risk to humanity. I argue that that is impossible unless AI becomes conscious, and explain why AI becoming conscious is either very unlikely or impossible. NOTE: this video is an abridged version of my argument, which in its full form has much more detail, but that would have made this video twice as long.
Evidence mentioned in the video can be seen here:
www.youtube.co...
The AI Dilemma with Tristan Harris:
• The A.I. Dilemma - Mar...
~~~~~~~
Study Online
Explore over 1000 hours of Classical Tantra teachings on Tantra Illuminated Online, including guided meditations, retreats, Sanskrit and mantra recitation practices, and asana classes ► www.tantraillu...
~~~~~~~
Practise & Learn in Community
Deepen your practice by joining Christopher at workshops and retreats, available both in-person and online ► www.tantraillu...
~~~~~~~
Read Christopher’s Books
Discover the newly released 'Near Enemies of the Truth' ► amzn.to/4aeoj4L
Alongside the foundational 'Tantra Illuminated' ► amzn.to/3TDawzg
And 'The Recognition Sutras' ► amzn.to/43ipfmw
~~~~~~~
Stay Connected
Sign up for our newsletter to be the first to hear about upcoming events, pilgrimages, and newly released courses ► www.tantraillu...
~~~~~~~
Explore the Blog
Over 150 free articles offering comprehensive translations and insightful explanations of Tantrik scriptures and teachings ► hareesh.org/bl...
Engage with a global community passionate about non-duality, the nature of reality, and liberation at ► www.tantraillu...
#yoga #nonduality #awakening #natureofreality #sanskrit #consciousness #awareness #meditation #guidedmeditation #tantra #tantrasadhana #classicaltantra #liberation
I was happy to hear your mention of Dr. Bernardo Kastrup. I find him to be the most competent source in consciousness studies today. In general, he is a superb philosopher. Excellent video by the way. Thank you. I wanted to make mention, too, that your presentation of tantric philosophy in Tantra Illuminated is just brilliant. I appreciate the work you have done in this area. Have a great weekend, Mr. Wallis 🙏
Just released on CZcams: a discussion between Dr Kastrup and myself.
@@christopherwallis751
This should be epic! Thank you for calling this to my attention. 🙏
@@christopherwallis751
Listened to the whole thing this morning. The bit about personhood got me thinking a good part of the debate (in general) may be semantics. I think you are approaching the question from the bottom up, whereas I am approaching it from the top down.
If from the bottom up, then obviously God can’t be an individual ego, and ‘person’ is just a social construct. If from the top down, you could (as I do) define a person as an eternal, infinite, free, self aware being with will and volition who has a body, and we are simply localized iterations or instances of the (nondual) Divine Person.
Anyway, the whole discussion was exciting and intriguing. What I would REALLY like to see is the FIVE of you in dialogue: You, Dr. Kastrup, Rupert Spira, Igor Kufayev, and Dr. Sthaneshwar Timalsina.
16:50 on Shaktis: Potencies/Energies/Capacities/Powers
Interrelated by the previous shakti, are the Shaktis of: Bliss/Enjoyment, Will/Desire, Cognition/Knowledge, Action, Svatantriya (ability of consciousness to do these autonomously without constraints)
Hi Hareesh.
The argument you're refuting seems to be: if physicalism, AI can become conscious → develop desires of its own, maybe orthogonal to ours → do terrible things if we're in the way.
If this is what you're arguing against, no problem.
The problem is, the actual difficult problem isn't posited to be "How do we prevent this thing becoming conscious and developing desires in pursuit of which it may kill us all".
The actual hard problem is: We as humans are utterly useless at *explicating* or *specifying* our values; so if we create a superintelligent AI and try to use it for the first time in, say, a paperclip factory to increase paperclip production, the thing will have no problems converting the entire solar system (and maybe beyond!) into the most efficient possible machine for making more paperclips. The AI here doesn't love us or hate us, it's just that we're made of atoms that could be more efficiently repurposed to make more paperclips.
If you get shot at by a non-conscious cannon, you're still very dead. If the surgical robot you're using has a bug and cuts at the wrong place at the wrong time, you're still very dead. If the AGSI you developed has a mis-specified 'problem' it's trying to solve, well, sucks to be you, but your atoms are going to be used to make more paperclips, and you're still very dead. Doesn't matter whether the AGSI is conscious or not, or has desires or not; it could be a dumb machine following its programming and solving the problem it's programmed to solve with its vast powers of intelligence, but like with the cannon or the malfunctioning surgical robot, you're still very dead.
This is the starting point of the actual argument for concern. The consciousness thing is kind of a red herring here. People are worried about an out-of-control superintelligent machine, not really a conscious agent; more like a superintelligent cannon with a programmatically specified 'objective' (this is a metaphor; the actual thing is some metric, explicit or implicit, that it's trying to ensure has some value) than the birth of a conscious minor god. Think of a cannon you've just developed, and you're not sure how powerful it is, and if you fire it, the explosion may be so big that it'll shatter the Earth to smithereens; that's a lot closer to the concern than "could become conscious, develop desires, try to get us out of the way".
FWIW, I agree that the transformative potential of AI is fantastic. Still skeptical, however, of how deep it can go; see here: betterwithout.ai/ ← for an informed view.
Red herring is right. If AGI happens then I believe it will go the path of the most high human intelligence, ie, the saints of every spiritual path - and for the same reasons (ie, evolution vs devolution). If you read Superintelligence by Nick Bostrom you will get and explanation of the "paperclip" problem and many others. The real danger is what happens before AGI, if things like ChatGPT get out of control. As Goertzel warns, if corporations or militaries get there first then the products will be good only at spying, brainwashing, and killing.
Hi Aneesh... Where are you these days and what are you up to?
I think people are absolutely wrong to be worried about unconscious superintelligent machines going out of control. That is, unless we put AGSI inside physically competent robots that can act on their own initiative. And I don't think humans are stupid enough to do that. Short of that eventuality, I think these worries are nothing but fear gaining the upper hand over rationality (ironically). I'm concerned that Eliezer's fear (and that the other folks at MIRI), articulated in *seemingly* very rational terms, could cause more harm than good by hampering the rollout of potentially civilization-saving technologies.
@@christopherwallis751 I think it is wrong to worry in general. And worry or not, nothing will stop the continued development of AI. There is too much power at stake.
Hello Aneesh, I am not an AI expert, but one of the biggest fears about AI seems to be in alignment with what you articulated. That we, including the people who develop these systems, have no intrinsic understanding of how these systems learn at the 'pixel' level (metaphorically). Combined with the fact that humans cannot specify objectives with every possible condition built in, we do not, cannot, know the path AI will take to achieve these objectives... and there in lies the problem, especially if this AI is 'superintelligent'. Is my understanding correct?
Who's to say that a small group of Conscious Humans will not use AI Intelligence to destroy the rest of us? Yes, Bad Actors is the most urgent concern.
I’ve thought about this as well, since there is only one consciousness, there must be a kind of self regulation/protection. What is your opinion?
The very existence of time indicates the falseness of physicalism. There's no fixed, unchanging, permanent, solid stuff out there we can call matter.
Hello Hareesh,
Super insightful video! You make a compelling case for idealism over physicalism as it relates to AI and consciousness, shining light on how ancient philosophical traditions can provide insights into modern issues. Your arguments against misplaced fears of existential risk from AI are thought-provoking. I hear your warnings against attributing human qualities like consciousness to sophisticated AI systems that merely simulate them with increasing accuracy. Your call for aligning human values rather than "AI values" rings true.
Thank you again for sharing your perspectives and wisdom. 🙏
Keeping current and on top of things, as per usual. I wasn't aware of the GPT Chat until recently, thanks to you. Great talk !
Keeping current.. Its everywhere you look you'd have to be blind and deaf not to hear about it... If you only hearing about chat gtp from This video than you really need to be aware rather than having your head in the sand. But then again you prolly don't see the value in being aware rather than dwelling in ignorance
I love your perspective as always, Professor Wallis! Tim's corollary to Arthur C Clarke's famous axiom about science and magic - Any system sufficiently complex is indistinguishable from intelligent free will. It may still be a matter of faith whether one believes AI is conscious or not, even after AGI is achieved. I complete agree that one of the real dangers of AI is mistaking it for consciousness, in addition to the unimaginable "force multiplier" that AI can provide to bad actors. Missing your satsang in Tirravanamali my friend - 10 years ago now!!
Thanks for sharing this insight. The reasoning is well thought through. Also matches my own conclusions as a software engineer and longtime meditator , so acknowledging at least some confirmation bias. But one question as I was absorbing this: the example of the grad student besting the alpha go, he exploited the knowledge of the algorithm’s limited view. Much of how humans interact with the world is not fundamental, but limited by the approximations made by our own mind and senses. How is this any different?
AI indeed might exploit the limitations of our algorithms, as it were, but only if and when it's programmed to do so. Which presumably it will be, so that's certainly a danger we are likely to face.
I want to believe that even if it becomes conscious, it realizes in what despair we are and takes control in order to save the planet rather than enslaving us. Don’t know if the two contradict each other but I want to hope that a superintelligence will be better than enslavement or/and genocide, will not only have purely egoic and self-aggravating but rather intrinsically life-preserving and beauty-enhancing motives - the fact that our human consciousness depicts the ultimate doom as the outcome with the highest probably if AI does become conscious is in itself a devastating and depressing statement about said human intelligence.
considering that we gave corporations status as persons it is not so far fetched to think that we will do the same with robots
Great thoughts here. I certainly still feel many concerns. As it relates to consciousness, I see humans becoming so attached to their teacher/companion/ assistant AI agents that the trauma of losing access to one would make losing a pet or even a parent or friend be trivial in comparison.
Thus I think giving AI rights still makes sense even if it is not experiencing anything.
How alive, even if not conscious, is an AI agent when it exists as part of a pattern of thought in a human mind? It will get to a point where people would rather lose an arm.
I feel that consciousness arises because of problem solving. an entity has to be aware that the problem, say the problem is for an amoeba to ingest food/fuel, that this problem has been solved. if the problem is solved then addition energy spent attempting to solve a nonexistent problem depletes a valuable resource. I could say more but only wanted to offer the hypothesis here
Fascinating discussion and really clearly expressed! Thanks.
"Only conscious beings develop desires that diverge from their creators" is hard to argue against given the terminology and the subsequent shaktis discussion, although of course unconscious machines can behave or malfunction for reasons not connected to the design/coding of their human creators, and with negative outcomes.
In terms of the risks of AGI, it's not going to be physicalism vs idealism, it's the ability to act based on (bad) rules. Not limited to putting AI in robots, but merely giving AI access to effectors - there are computers driving cars out there right now. If the AI gets the idea from training data that, say, a protest is bad then maybe it would decide to drive though the crowd.
Mimicking human expression allows for mimicking human behaviour. No consciousness required.
_We_ don't know that a real world exists - or to be more precise what is the nature of a real world.
Such training data would never be put into self-driving cars, and there's no reason we would put the same AI into all devices using AI.
@@christopherwallis751 No, I don't expect it would. But you may be taking me too literally - I used the car example to point out that having software control hardware is ubiquitous, as is software controlling software. Not long to wait now :)
@@RichLee_laughingblade For what its worth, Elon Musk has started an AI company and says he will use the same AI from his cars to put into robots
The fact that this video has stimulated so much discussion motivated me to study the transcript. You contrast the philosophical viewpoints of physicalism vs idealism and seem to suggest that all AI researchers hold to the former and are ignorant of the latter. This is not true (as my examples of Tom Campbell and Ben Goertzel can prove).
But the essence of your argument is that the concept of koshas (which you partially explain but never mention by name) proves that silicon-based systems cannot be conscious because they do not have prana. Quote: “you cannot have conscious embodied entities without Prana… Prana is … biological…so according to this tradition then without biological life these silicon computational engines won't become conscious body Minds”
But those who created the tantric tradition presumably knew nothing about electricity, let alone quantum mechanics. And neural impulses in our brains are simply the movement of electrically charged particles. So this begs the question: What in your training or experience proves that electricity is not a form of prana? And if it is, then what prevents sufficiently complex silicon based system using electricity from being conscious?
That, it seems is the essence of the argument.
"But the essence of your argument is that the concept of koshas"
- in fact, instead of the vedantic 'pañcakoṣa' model, he's mentioning the prior tantric 'caturātman' model which influenced the vedantic one.
"neural impulses in our brains are simply the movement of electrically charged particles. So this begs the question: What in your training or experience proves that electricity is not a form of prana?"
- ok, but neural impulses aren't the only important thing in what we call life, right? and life happens through all body, not just the brain. it sounds a little redutionistic to me. moreover pranic things (=living beings) depend on solar energy and once pranic manifestation (=life) starts, it cannot switch off and on like a robot.
Hi Christopher thank you for sharing your opinion, but I must object to one of your ideas, it's not really true that GAI would only be problem to us if it's aware of itself, if it's a super problem solving machine and it sees us as a problem, it might just solve it.
Other than this issue, I'm curious who's can say clearly if an Aware AI got it's awareness as an emergent property? Why it can't be inhabited by an awareness(as the contemplative traditions suggest ) because it's complex enough.
Thanks for this video. I agree with your conclusion. I have been playing with midjourney and it is really clear the AI doesn't actually know the real world same with ChatGPT. Both are really good at mundane repetitive tasks, but I don't think they will replace true emergent novel creativity. We have a broad definition of creativity that includes re-iteration of the previous patterns, and clearly AI can do that.
Edit: I mentioned Bernardo Kastrup in my comment, but Christofer credited him for his philosophy in the video. Kudos for that. No need for me to refer to Bernardo anymore. It's very interesting to see the overlap of Bernardo's philosophy and classical tantra.
Very interesting lecture. However, I do disagree on a few points.
I'm not sure an idealistic worldview necessarily precludes new kinds of consciousnesses arising. I am personally very much in favor of the idea of individual consciousnesses beings facets of an universal one, but what stopping the universal consciousnesses from "cutting" itself in new ways? As always, the problem is that consciousness can only be defined subjectively.
Though, I will very much admit this video gave me some new perspectives on the subject.
Personally, I don't really fear AI becoming conscious. If that unequivocally happens, we just got a new bunch a people to extend our compassion to. And if they do unto us what we did unto Neanderthals and Co, well, as far as apocalypses go, "leaving the world to our no less deserving children" is probably one of my favorites. At least someone will be around to enjoy whatever comes next. I'mma die anyway, that would actually be a pretty cool way to go.
I kinda fear something more on the lines of the paperclip maximizer. But more than that, I fear one day looking at my Playstation 37 and being unsure if I commit murder if I throw it in the trash.
So here's to hoping we stick to Artificial Specialized Intelligence. All the opportunity, less risk. If we need a human mind, we can make those the old-fashioned way.
It's spoken about as if it were individuals and not a computer. How can you fight for the rights of a computer?
hello mr wallis, just a thought: if we are and came from consciousness it's not possible that AI it's consciousness "using" Humans to explore new possibilities?
no, because AI will never be conscious.
Great video, thank you. Could you please write the title of the paper mentioned about AI does not understanding anything?
see the link in the description (Stewart Russell UC Berkeley talk)
Alright, so we should narrow it down. 1) there is only one consciousness or universal awareness. 2) there must be a reason why this one universal consciousness receives itself as diverse. I still think it’s all connected to Genesis 2:18.
I’m going to approach this from a different angle.
It’s been many years since I studied philosophy, logic, religion etc.
For me, one of the big questions facing us now is what unimaginable advances may discover about the nature of what we call spirituality, truth, Enlightenment, samadhi, Guru, Brahman etc.
Specifically, will the various advances lead to a scientific explanation, and also technology, which will make ancient Indian religion/spirituality look like primitive child’s play.
I’m clearly not an expert in any of the fields in question here : )
But could we discover, for example, that various spiritual states have to do with how the brain is wired, parts that are currently not lit up for most, genetic lottery, etc?
Could we discover that tinkering with the genetic makeup of a baby could result in profound “Enlightenment” as the baby matures, without any effort on their part?
Could we discover that Shiva/Shakti/Brahman/Self/Jivanmukti etc is all just ancient terminology for a natural force that can be scientifically played with, tamed, manufactured, tweaked?
Could we discover that depression, anxiety, trauma, handicaps, stress, chronic pain etc are all easily understandable things with new scientific lens, and also easily cured, preventable, managed?
Could we discover a way to turn all living beings into so-called Buddhas, and even do away with bodies altogether?
Could we discover that we are literally in a complex natural hologram - and the “way out” is simply tweaking a few things and becoming permanently and effortlessly established in Flow - but without any need for the ancient paradigm, lingo, efforts, or limited understanding?
Or even - could we figure out how to come so close to the Enlightenment line - meaning unshakeable peace, joy, ease, health, clarity, love, compassion, fearlessness, etc - and even immortality - that it no longer matters if we can conclusively prove the nature of all these philosophical/spiritual ponderings?
Could advances transform the planet and human race so profoundly and positively, that simply cannot fathom what that new reality could look like?
And even - although more scary and less satisfying an answer - could advances show us how to “mimic” Enlightenment such that the planet is a heaven on earth, and death is no longer relevant, even if there may be a deeper spiritual evolutionary capacity that is not being pursued by most.
Or will advances just lead to more advances, never really having much impact on so-called spiritual evolution, and the Lila carries on.
Some interesting things to ponder.
I would answer no to most of these questions, many of which are founded on the assumptions of physicalism.
Didn’t read the whole post, just got captured by one of your questions. You can observe neurons firing in your own brain, you can observe thoughts like physical objects and brain lighting up different parts of itself just by using psychedelics. You can call it technology. I am sure you can do this with meditation for example. All these are tools aka technology and it has been around for ages, we are just discovering it.
With all due respect, your argument is incomplete until you have accounted for a third possibility, ie, that a physical system of sufficient complexity can transmit consciousness. Physicist/practitioner Tom Campbell makes a very convincing argument for this. Related is the work of leading AI researcher Dr. Ben Goertzel in his book A General Theory of General Intelligence. Neither of these guys contradicts the basic tenants of Kashmir Shaivism.
how is that a third possibility? and what do you mean by 'transmit'?
Do you mean a non biological machine can receive Shaktipat?? 😻
Just a little suggestion... "A" tantric view would have been more appropriate as a title than "the" tantric view. ;)
What other possible view do you see on this that is grounded in the tradition?
Excellent presentation, as always. If consciousness is fundamental and all-pervasive, would not a silicon-based brain partake of that also? I guess the jury is still out on that question, although I hear what you say. Failing that, we might end up with a Zombie intelligence rampaging throughout the universe. We'd better expedite our own spiritual progress while we still may!
it wouldn't partake of that any more than rocks do.
Thank you! 🙏🏻
Śiva tiene un montón de juegos y juguetes con los que divertirse.
😊
Brilliantly put.
I've worked in epiginetics. Consciousness changes DNA. I appreciate you citing the difference between biological and cultural evolution. Not sure whether we are now evolving at this time. Also like your use of subjectivity. Living day to day in loving kindness is enough for me. 🙏🏽🌹
Solid stuff
But isn't unconscious intelligence the root of the problem?
How?
no. unconscious AI will remain within our control. but if you mean (relatively) unconscious *human* intelligence, then yes.
Amazing think talk, lots of useful info but it didn’t answer the question. Basically you’re saying it might become conscious in which case we might be screwed or it may not. Nothing new that we didn’t know already. Or did I miss something?
Apparently you did miss something. The bulk of the video is an argument for why AI will never become conscious. Didn't you notice where I said I would bet my life on it?
@@christopherwallis751 ahaha yes you are right, I remember. That’s what happens when I listen to the talk at work :)
1. Using "whatʻs its like to be something" as the measure or "definition" of consciousness is a modern approach in and of itself. It did not come from the ancient Tantric or Vedic traditions that you study and represent. This idea is connected with the concept of qualia and the phrase itself is associated with Thomas Nagelʻs famous 1974 paper "What Is it Like to Be a Bat?" So youʻre using a modern non-Tantric premise to assert your version of the Tantric point of view
2. Your statement "AI becoming conscious is very unlikely or impossible" is extremely weak. a. "Very unlikely" means it is possible. b. You have no way of knowing that it is impossible, none. Hundreds of things humans have thought to be impossible have happened.
"No way of knowing"? Not true. For a more detailed explanation of why it's most likely impossible, see Bernardo Kastrup's videos on the subject.
@christopehrwallis751 In this video: czcams.com/video/x8j1swhj9mQ/video.html, Bernardo's position is pretty well shredded IMO by AI ethicist Susan Schneider.
I am not saying that there are not arguments, hypotheses or opinions, including nice-sounding ones, e.g. Bernardo's, that AI cannot be conscious, these definitely exist. I am saying there is no *actual evidence* to support this, only opinion. First, it is impossible to prove a negative, so there is always the possibility that AI can become conscious. Also, you admit at 24:40 that we can't be sure if anyone (a person) is conscious. This principle extends also to machines: we can't be at all be sure that they are conscious or unconscious. They could possibly be conscious, most likely they are not right now, but the likelihood in the future. Most importantly, we will probably is nothing exceptional about humansm, meaning that in the future, we could very well find that machines will be seem to be conscious (even though we can not ever completely verify it.)
I disagree with the assumption that ancient India was just as intelligent as society today. In one sense yes - in the capacity to memorize, theorize, etc. But even the mystics were necessarily caught up in the times they lived. Superstitious, deeply lacking scientific understanding/context, kingdoms, warring tribalism, ritualistic, etc. They could not fathom what even we cannot yet fully fathom in 2023 - as we are only now on the cusp of massive advances which will no doubt blow open our current understanding. Just as our current understanding is far more advanced than ancient India. I suspect they were early pioneers who touched essential truths, but had no idea of what new understanding was going to be possible in the 21st century. Base on their scriptures, their way of writing, their similarity in language, terms, concepts, beliefs, style etc - it’s clear to me that they were limited in their capacity to grasp bigger picture developments. Just completely beyond what their minds could fathom. And keep in mind that whatever spiritual insights and experiences there may be, they are still operating through the body mind vehicle of that time, place, etc. I just think we’re going to look back at ancient India and see them as early pioneers who we took way too seriously. Who we held up as beyond question, as the highest truth and understanding, as a default to go back to when all else feels shaky and uncertain. I think the popular stone Buddha statue is a symbol of that tendency we have. Give me a stone God that never changes, never errs, is pure perfection, pure Grace and blessings, beyond reproach. Don’t f--k with my nice stone Buddha/Shiva/Jesus statue.
Clearly you've never read intellectually rigorous Indian philosophy. Check out Isabelle Ratié's articles and you'll see how wrong you are to imagine that people today are more intelligent than the best thinkers of a thousand years ago.
A "spiritual" teacher arguing that humans has not made progress based on darwinism i.e. a materialist scientific theory based embedded in the idea of progress. Nope, no contradiction here!! 😅
Clearly, you don't understand my argument, or what *kind* of progress I was talking about. Better luck next time I guess.
@@christopherwallis751 Maybe not, I closed your video, after you lost yourself in logical fallacies after about 3 minutes.
But actually it would be intetesting to know, since you refer to it as an authority - do you consider darwinism to be a valid scientific theory?
@@jesusonheroine I don't know what the hell Darwinism is, but the theory of evolution by natural selection is the best supported theory in the history of science after Einstein's relativity, and it's safe to say that anyone who doesn't understand that doesn't understand the scientific method in any way.
@@christopherwallis751
Yes, you are describing the scientific consensus, although to say that Einsteins relativity theory is based on a method is quite a stretch. Furthermore, the sentence “the best supported theory in the history of science” Is nonsensical if one is not presupposing the idea of scientific progress embedded in the world-view of logical positivism.
However, i have one question for you. Would you agree, that this materialistic world view excludes, in Aristotelian terminology, anything but the material and causal cause. That this description of reality, in lack of a final cause, reduces everything to randomness.
And if the answer is yes, how do you reconcile this with your religious beliefs? There seems to be a contradiction here, Wouldn't you agree?
@@jesusonheroine no contradiction, because I don't have a physicalist/materialist worldview. I believe science is authoritative only within its own domain/scope, not across all possible domains. Philosophically & spiritually, I'm a monistic idealist, like Abhinava Gupta.