ben and joscha on conscious ai

Joscha Bach Λ Ben Goertzel: Conscious Ai, LLMs, AGI – oct 2023

ben goertzel.. ai.. consciousness.. ben on agi ness.. et al

notes/quotes [livestream so no timing when i started.. see bottom of page for timed themes.. started adding times at 27 min.. of 2 hour video]:

host: j as everything is computational and b as other categories..

j: i think everything we think and perception happen in some kind of language.. a language cannot refer to anything outside of itself.. in order to be semantically meaningful it cannot have contractions.. if a language is self contradictory its terms don’t mean anything.. languages that work turn out to be reps .. computation is something that can be implemented.. i think for something to be existent it needs to be implemented in some form and means we can describe it in some kind of constructive language.. when i think about reality is about i need to do this in a language in which my words mean things.. otherwise what am i talking about .. need a language that is constructive.. i am a strong computationalist.. because i believe that if we try to use non computational terms to describe reality.. and it’s not just because we haven’t gotten around to formalizing them yet but because we believe we found something that is more than this.. we are fundamentally confused and our words don‘t mean things

whalespeak.. just because we have no idea what legit free people are like.. we assume ie: language (which is language as control/enclosure); invited vs invented; in a space ness; ..

need 1st/most: means to undo our hierarchical listening to self/others/nature as global detox/re\set.. so we can org around legit needs

idiosyncratic jargon ness via self-talk as data

so that if/when we do gather in a space.. it’s unlike any dynamic we’ve yet experienced.. it’s not about ie: trying to communicate.. it just is/already communication..

b: i think i start from a diff perspective on all this philosophically.. one minor point i need to quibble with.. if a logic contains contradictions it’s meaningless.. of course that’s not true.. contradictions are not necessarily fatal to having meaningful semantics.. i want to shift back to foundation of universe and everything here.. i tend to be pheonomonolgical in my approach rather than starting from a model of reality.. these things have a tendency to be hard to put into language.. because once you put them into language.. then yeah.. you have a language.. cause you’re talking in language.. but talking isn’t all there is to life/experience.. i don’t have any disagreement that this is the focal point if engineering an agi language.. but otherwise.. i don’t feel a need to reduce the whole metaphysical.. just because the program code lives there .. this is not so much about ai or computation formalism.. these are just diff philosophical perspectives which becomes arduous (difficult/tiring) to talk about because natural language terms are imprecise/ambiguous/slippery. you could spend your entire career trying to articulate what is really meant by relationship

lanier beyond words law.. rumi words law.. et al

j: comes down to the way our thinking works and what we think thinking is.. 1\ build things from first principles.. ie: computer programming .. to draw line need particular language.. bites as commands to computer.. have to learn these constructs to do things

oi.. nobody is arguing that.. but that has nothing to do with alive/free human being ness.

j: 2\ other way that can’t be done in computers.. strong intuition.. how did get into brain.. how able to make inferences.. that go beyond inferences i can make.. other end of perspective.. just make noises.. and if get away with it.. you’re fine.. not appealing to me because people can be very wrong in groups

oooof..

b: saying the only thing there is is language.. is that only thing we can do is languge

j: not saying that.. saying language is a rep.. a way to talk/think/model about things.. obviously not everything is a model.. just everything i can refer to is a model..

once you think you have to rep.. to me.. you kill.. representation ness et al

b: you can’t know that.. you can hypothesize that.. you can’t know that

graeber can’t know law

j: i cannot know anything that i cannot express..

but that not knowing doesn’t take it out of the realm of reality..

b: i can know many things i can’t express in language.. but i guess that’s just a diff flavor of subjective experience.. take what martin buber called the i-thou experience.. if you’re staring into someone’s eyes and you have a deep experience that you’re seeing that person.. sharing a shared space of experience/being.. in that moment .. that is something you both know you’re not going to be able to communicate fully in language.. and it’s experientially there.. now buber wrote a bunch of words about it.. and those words communicate something special to me and to some other people.. but of course someone else reads the words he wrote and says you were merely summarizing some collection of firings of neurons in your brain and in some strange way deluding yourself that it’s something else.. i think from w/in the domain of computation and sci.. you can neither prove/disprove that there exists something beyond the range of computation and sci.. if you look at sci data.. the whole set gathered by human race is one large finite bit set.. large set of data points w/finite precision to each .. then sci in a way is trying to come up w concise/reasonable explanations that can be used to predict outcomes of other experiments.. in a way that’s accepted by a certain community.. that’s a certain process.. a thing to do.. for producing computation models for producing finite sets of bits.. and that’s great.. nothing w/in that process is going to tell you that that’s all there is or isn’t to the universe.. it’s a valuable important thing.. now to me.. as an experiencing mind.. i feel like there’s a lot of steps i have to get to the point where i even know what the finite bit set even is.. or what community of people validating what that bit set really is.. or what a programming language is.. so i keep coming back to my phenom experience.. there’s this field of nothingness or contradictory nothingness just floating there.. then some indescribable forms flicker and emerge out of this void and then you get some complex pattern of forms there which constitutes the notion of bit set or experiment or computation.. and from this phenom view by the time you get to this business of computing and languages you’re already dealing w/a very complex body of self organizing forms/distinctions that popped out of the void.. and then this conglomeration of forms emerge out of void enough so we know ‘i am everything’ ‘the only thing that exists in a fundamental sense is what is inside me’.. if you’re inside that thing.. you can’t refute or really demonstrate that.. but again from an agi view it’s all fine.. because when we talk about building in agi what we’re talking about is precisely engineering a set of computational processes..

host: appears you’re using diff defn’s of knowledge

[ok.. now video has time lapse so.. ]

27 min – j: if you use the traditional philosophical notion of justified true belief.. it means i have to use knowledge in a context where i can hope i have the notion of what’s true.. ie: when i look at your face and experience a deep connection w you and i report i know we have this deep connection.. i’m not using the word know in the same sense.. what i am describing is an observation.. i’m observing that i seem to be looking at a face and that i’m having the experience of a deep connection.. and i think i can hope to report on this truthfully

why report?

j: but i don’t know whether it’s true that we have that deep connection.. i cannot actually know this.. i can make some experiments to show how aligned/connected we are.. to say that this perception/imagination has some veracity .. but i’m referring to a set of patterns.. dynamic patterns that i perceive.. and they’re stuff i can reflect on and convey/model.. this is a distinct category.. not necessarily in contradiction to what you are saying.. just using the word knowing in diff ways is implied here.. because i can relate the pattern to you that i’m observing or that i think i’m observing.. but this is a statement about my mental state.. it’s not a statement about something in reality.. about the world.. to make statements about the world.. i probably need to go beyond perception..

j: the 2nd aspect that we are now getting to is when you say reality and minds might have properties that are not computational.. yet are agi.. is entirely computational doesn’t need any kind of first principles wonder machine built into it.. that goes beyond what we can construct of automata.. are you establishing that agi is artificial general intelligence as potentially super human capabilities are going to be lagging behind what your mind is capable of

what? yeah.. they are lagging behind..1\ intellect as cd.. 2\ what computers can’t do et al.. oi..

29 min – b: no.. not at all.. i just think the other aspects are there anyway.. and you do’t need to build them

oof?

j: so you’re going to make the non computational parts of reality using computation

b: no.. you don’t have to make them.. they’re already there.. just take a more simple pov where you’re thinking about .. if i build agi like human mind.. most likely same form of first ness will be same in agi but don’t have to construct it.. already there.. it’s there in time.. didn’t need a theory of time.. awareness is ambient.. do’t need to build it.. it’s there.. of course diff constructions might have diff awarenesses.. so how to treat diff kinds of firsts if not relationship yet articulating diff already brings you into 3rd ness

32 min – j: i think it’s conceivable that objects have conscious intelligence.. but would require they have more complicated structure than computer used to communicate it to you.. seems there would be simpler ways for objects to be structured.. don’t need to be conscious to do anything like that.. to.. redefining consciousness itself.. to me consciousness is that thing that seems to be suspended.. that stops you from learning/interacting..

b: clinton quote: ‘all depends what the meaning of is is’ .. i may have lost something in the buffering process.. first of all.. about causality and raw experience (first ness).. first ness doesn’t cause anything.. if only thing you’re willing to attribute existence to is that it’s causal.. perse thought only 3d was of that nature.. the word consciousness is highly ambiguous.. i don’t care if people want to preserve consciousness for .. whatever.. lengthy debates on .. is particle conscious or proto conscious.. or raw vs reflexive consciousness.. in the end.. i’m like.. this is a game that overly intellectual people are playing to entertain themselves.. and it doesn’t really matter.. arguing about which words to associate w diff flavors/levels of experience.. running around in circles..

38 min – j: to me conceptually it’s an important question.. these are questions i think i can answer.. if i don’t my thinking will become so mushy that my thoughts are meaningless and i will not be able to construct anything

b: if the only kind of answer you’re interested in are rigorous scientific answers.. then you have your answer by assumption.. and answering questions by assumption is fine it’s practical it saves our time..

j: i think that’s what you’re doing.. i don’t see how you’re not just trying to answer by assumption.. then gives ie’s.. i want to take what you’re saying seriously.. but if you don’t take words.. we can’t communicate.. this is more a philosophical question.. we ahve to negotiate among the two or three of us

40 min – b: rightly or wrongly as a human being.. i’ve gotten bored w that question.. i couldn’t say it’s worthless.. but on the other hand.. i personally had to get bored w repeated debates that seem to go over the same points over and over again.. one of the things i get worn out with is people claiming my defn of this english word is the right one .. yours is the wrong one.. you weren’t doing that .. but just agave me a traumatic memory..

42 min – j: i’m not fighting about words.. i don’t care which words you’re using.. for me question of whether i project this property into parts i consider as reality is important..

43 min – b: let me talk about anesthesia.. always the nagging fear.. since we don’t really know how it works.. always possible you’re feeling the knife cutting thru.. ie: does the jaw hurt.. brain may be disabled by anesthetic.. i suspect under anesthesia.. your reflective theatre of consciousness is probably disabled.. so probably not a version of ben saying it hurts.. then forget it.. i don’t think that’s what’s happening.. on other hand.. i think the jaw is having its own.. i think there’s some raw feeling jaw itself is having.. even if not connected to reflective theatre of awareness in brain.. the jaw is bio cells.. some think those cells have experience.. but a brick when smash w axe doesn’t.. but i suspect brick also has some elementary feeling.. but not something can reflect upon.. but don’t know how to make that science.. cause can’t ask jaw.. because jaw doesn’t speak language.. not sure how you pin that down in experiment

47 min – j: there have been experiments on anesthesia.. there is .. one .. when applied.. can’t move.. other forget after..

54 min – j: on jaw.. vs tree.. vs whatever on consciousness.. i’m not assigning some kind of undo privilege to it.. i’m just observing behavior and look for means/motive behind that behavior.. and then i try to construct causal structure.. i might be wrong.. but at least it’s not because i have some kind of speciesism that i assign higher level to self because it’s me

55 min – b: alright.. yeah.. now i know what your motivations are.. kurt.. i have a higher level comment.. we’re like 1 hr thru convo.. i feel like the hard problem of consciousness is an endless rabbit hole.. not uninteresting.. but not the topic on which j and i have most original things to say..

j: i might interject.. what our most interesting disagreements is in b being a panpsychist (everything has consciousness) and me not knowing how to formalize panpsychism in a way that makes it diff from box standard functionalism.. so i do value this discussion and don’t think it’s useless.. but i basically feel that on almost everything else you mostly agree except for crypto

56 min – b: ok.. yeah.. to me that’s almost a zen thing

56 min – b: i think most issues we agree on quite well (agis and llms) obviously large language models.. are an amazing tech from ai application pov.. i don’t think it’s the only amazing thing to happen.. but it’s an amazing things.. if deployed properly could do man of things people are now doing.. could have huge social implications.. i think way algos are rep-ing internally.. are not what need to make a full on human level agi system.. doesn’t rep abstractions in a sufficiently/flexible/manipulable way to do the interesting things the human mind does.. just not the hierarchy of abstractions the way human mind is.. the subtler question.. what function get by glomming llm with other.. etc.. then have something turing complete.. but if hub is llm w limitations.. won’t do kinds of interesting things we think it can .. ie: invent new branch of math.. invent new genre of music.. figure out new variety of business strategy.. all these things involve a leap into the unknown beyond the training data.. to an extent that i think you’re not going to get w way llm are rep ing knowledge..

oh my.. to me.. not new.. rather.. diff degrees/versions of same song

intellectness as cancerous distraction

there’s a legit use of tech (nonjudgmental expo labeling).. to facil a legit global detox leap.. for (blank)’s sake.. and we’re missing it

1:01 – b: i do think llms are powerful as tools to create agi.. ie: one subproject in my agi project.. is llms to match english sentences into computer programs.. now can use logic engine to connect everything.. can use as guide to coming up w theorem.. i do thing there are ways to serve as component agi.. but if going to make hybrid agi w full human intell.. something beside llm has to play key role.. this sides in w llms not being motivated agents.. so could wrap a motivated agi around an llm if wanted to.. w/o that abstraction in our consciousness.. not going to get fundamental creativity.. in humans we do this amazing novel stuff.. by abstraction tied in w motivation..

yeah.. i don’t think we have any idea what legit free people are like.. to me.. nothing about motivation.. all this would be irrelevant/spinning our wheels.. talking et

1:05 – kurt: also there are emergent properties of llms.. ie: reflection

b: there are.. but they’re limited.. that does make it subtle.. you can’t say they don’t emerge knowledge.. there is emergent knowledge rep.. but very simplistic and limited.. doesn’t pop up form in context learning..

kurt asking j if he thinks new math could be created..

j: i am too stupid to answer this question.. but i can give perspectives.. personally i’m agnostic w respect to its abilities.. if look at what llm is doing it’s not a model of brain.. it’s a model of what people say on internet.. translating to rep.. then to tokens..

aka: whalespeak

1:07 – j: if count all training data.. difficult to make new dimension.. but can make diff combos

oi.. same song.. need something legit diff..

1:18 – b: to prove limitations of llms.. asking what tasks can it probably not do .. hard to rigorously prove/disprove what these systems are going to do because don’t have theoretical basis for it.. but still have to make a choice of what to pursue.. so are going into this field w/o rigorous proof..

mufleh humanity law: we have seen advances in every aspect of our lives except our humanity– Luma Mufleh

humanity needs a leap.. to get back/to simultaneous spontaneity .. simultaneous fittingness.. everyone in sync..

1:20 – b: want to go back to your comment that creativity is combinatory.. i think this ties in w the nature of the representation.. i moslty buy the notion that almost all human creativity is blending together esxisting concepts in some forms..

yeah.. to date.. whalespeak.. we have no idea what legit free people would/could do.. so keep replaying same song and thinking it’s diff..

b: i just think the *most interesting cases of human creativity are blending things together at a higher level of abstraction than the level llms generally rep things.. and most human creativity has to do w blending in motivational nature.. llm combining lower levels.. we do that also.. but most creative combine abstract patterns.. so i do agree** if had llm based on sufficiently amount of data/memory et al.. then yeah could produce jazz ness.. et al.. but so could others.. not that interesting.. question is.. could it do with less/more resources.. being able to invent diff calc.. jazz.. is a high bar.. a level of invention that individual ***humans don’t commonly manifest in own lives.. we know in history of ai no on ehas been good at predicting..

*yeah.. because all we have to date is whale ‘creativity’

**all irrelevant s.. because to me.. all data to date is nonlegit.. it’s like from whales in sea world

***again.. because all are whales in sea world

if want to replace humans w llms trained on humanity.. a lot would get done.. but stuck culturally.. no fundamentally radically new stuff

already not anything new.. since forever.. which.. not really that new ness matters in ie: making/creating things.. but more like .. new every day ness.. and graeber unpredictability/surprise law et al

_________

__________

_________

________

TIMESTAMPS: – 00:00:00 Introduction – 00:02:23 Computation vs Awareness – 00:06:11 The paradox of language and self-contradiction – 00:10:05 The metaphysical categories of Charles Peirce – 00:13:00 Zen Buddhism’s category of zero – 00:14:18 Carl Jung’s interpretation of four – 00:21:22 Language as “representation” – 00:28:48 Computational reality vs AGI – 00:33:06 Consciousness in particles – 00:44:18 Anesthesia and consciousness: Joscha’s personal perspective – 00:54:36 Levels of consciousness levels (panpsychism vs functionalism) – 00:56:23 Deep neural nets & LLMs as steps backward from AGI? – 01:05:04 Emergent properties of LLMs – 01:12:26 Turing-completeness and its implications – 01:15:08 OpenAI’s bold claims challenged – 01:24:24 Future of AGI – 01:31:58 Intelligent species after human extinction – 01:36:33 Emergence of a cosmic mind – 01:43:56 The timeline to AGI development – 01:52:16 The physics of immortality – 01:54:00 Critique of Integrated Information Theory (pseudoscience?)

_________

__________

__________

_________

_________