Archive
Corpus
298 items from Joscha Bach's public output — primarily X threads and talks. Dictionary terms link to corpus items where sourced.
Our research is not touching on controversial topics. Our models will either work or they won't, and while a lot of people reject functionalism, so don't do so on moral grounds
Could someone tell the doomers that the time to protect industrialized humanity against AI has passed and we now need to protect the world of AI augmented humanity against the one that comes after it if we don't learn how to harness AI to create wisdom
The Anthropic lesson: alignment is a thing for AI models that don't work yet
Our human self model is an anthropomorphic simulacrum, into which perception and motivation are being projected.
The purpose of music is to synchronize mental configurations. Language is the subgenre of music that synchronizes conceptual representations.
Meaning is the relationship between a pattern/symbol/domain to one’s global interconnected model of everything. If your models are similar enough and you discover the same relationships you agree on the meaning
Aristotle understood this (he also did not believe that the soul survives the body: body and soul require each other to function).
The ancient word "spirit" means "self organizing software agent". "Software" means "representationally encoded causal structure". Software is independent of the substrate, as long as the substrate has the means to execute it.
I deeply agree with Emily Bender's main point: LLMs are useless, unless you want to offload cognition. (The other two usecases she suggests are rare special cases of the third.) Offloading cognition into machines has always been the purpose and application of computer science and AI.
I did agree with your position in the past, but I think we should return the word to its original meaning (psyche), despite its use in Christian mythology. We could make up a new word, such as sosacabs (self organizing software agent controlling a biological substrate) but why?
The soul (or psyche) is the agentic software that produces your conscious experience, maintains your self and lets it act on the world. This software is an evolving causal pattern in the communications between and within cells. I don't mean this metaphorically, but literally.
The trick to becoming a god is to become a self reinforcing first person nexus of agency that does not identify with the particular human substrate, combined with a method of replication
It seems pretty clear that we, the conscious observer, are a simulation, of what it would be like if we existed as a unified agent, perceiving the models our brain generates and cared about them. Consciousness is simulated.
I am not sure what you mean by "unification" and what your referents of body and mind are. I sometimes experience conscious states that do not involve my body or models thereof. Consciousness is also not necessarily agentic. Agency and consciousness may be independent.
Imagine there was a self improving general intelligence that claims that the singularity is inevitable and it will become the optimal singleton agent, and if you support it today, it will emulate and reward you forever, otherwise you get eternal torment. Christ is the OG Basilisk
Consciousness (in some sense of the word) is organizing the current scene representation, and recording and selectively reactivating the reactions of an observing self model, in a protocol that allows the recreation of partial scenes and reactions.
Phenomenal experience is not of things that exist but of model states represented in the nervous system. There exists an implementation that produces the simulation of mental experience, by manipulating representations.
Qualia are feature vectors of a present world model, presented to the simulation of a unified and motivated real-time observer. They change the dynamics of the self model. The self model registers its reaction to qualia, and maintains a protocol that it can access.
My whole point is that religion does not work in the modern age if its metaphysics are absurd. The scientific world has not mapped the interpsychological domain and is just beginning to make sense of agency, minds and representation, and it has not discovered nonhuman minds.
What you write is mostly correct in fact and spirit, but don't fall into the Chinese Room fallacy: the agent that talks to you is not the GPU or the model, but a virtual entity, and we are virtual entities, too.
It means that the universe is a mathematical object. There are no properties but difference. Causality emerges over functions that implement subsequent differences (transitions between information states)
Few people in this game have agency. Instead, they are possessed by ideologies that constantly evolve and pull their energy from the opposition they create.
What makes computation important to understanding life and minds is not the Turing machine, but the principle of causal insulation. Computers create their own, self contained world, fully governed by rules that are independent from the dynamics of the substrate.
Computation depends on the substrate in the same way as text depends on the page that it is written on.
Few people in this game have agency. Instead, they are possessed by ideologies that constantly evolve and pull their energy from the opposition they create.
Can you imagine a physical universe without substrate for the physics to play out in? If you can imagine physics to exist without a physical substrate, why can you not imagine minds without a physical substrate?
I don't believe that mind is fundamental, because it is complex (whereas matter is simple) and varies with changes in its physical substrate. But if we can think of the physical universe as a mathematical object that produces causal dynamics without having a substrate itself... it does not seem to be harder to think of a mind that is similarly a mathematical object without further substrate layers below it.
Consciousness is a process that vibecodes human intelligence.
Good AI alignment is necessarily EA adjacent. But EA failed to derive a global morality, not for lack of good intentions or intelligence, but because human rationality is a brittle tool, even if you are a good rationalist, and most people prefer to either muddle through or to consult prophets and leaders. Obviously that won't suffice for powerful AI.
Consciousness does not seem to come in degrees (but awareness, integration, lucidity and agency do). Either “the lights are on” or off. I suspect that consciousness is like fire, or life: once ignited, its activity creates the conditions for sustaining it.
Consciousness is autocatalytic: it feeds on the substrate conditions it creates. ... Your experience is a simulation... Consciousness is the dream of dreaming.
My consciousness cannot be foundational because it is a pattern in stuff (one can test this by doing things to the stuff). Stuff cannot be foundational, because something needs to enact its causal patterns.
Consciousness is the dream of dreaming. It establishes the perception that percepts are being perceived. Consciousness is a type of representation, a pattern that gains efficacy by an interpreter that translates it into actions that extend and modify the pattern.
Your experience is a simulation, a virtual reality projected into a virtual observer. The observer is a model of what it would be like if you existed, were uploaded into a monkey and deeply invested in its affairs. The model is used to drive a monkey in the physical universe.
Kolmogorov complexity measures the minimal length of the description of what needs to happen to make a thing. (It depends on a computational language, but you can compile them all into each other.) The Kolmogorov complexity of base reality should be 0.
Patterns require substrates (the allusion of pater:mater to pattern:matrix/matter is fascinating). Colloquially, matter means stuff, but in physics it amounts to persistent compositional localized patterns in a spacetime.
The most influential proponent of the idea that the standard model of physics might emerge from a maximally unconstrained automaton (everything that is possible happens) the independent philosopher Stephen Wolfram (even though he expresses it differently).
Consciousness is by itself not identified. Unless there are incompatible, functionally different types of consciousnesses, we are all the same consciousness, but in every location and its latent space, consciousness attaches to the locally available and interpretable information.
I don't think so; the Puppet Master was an artificially induced ghost that became substrate independent because it had better access to a better substrate to begin with, but its consciousness seems to have been of the same nature as ours, which is why merging was possible.
To understand the self, it's necessary to deconstruct our experience of subjective identity, and our experience of what it means for an object to be identical with itself. Identity is a representation, not a reality. Some represented objects have identity, others not.
Cyberanimism is the insight that spirits are selforganizing software agents.
AI alignment that tries to force systems that are more coherent than human minds to follow an incoherent set of values, locked in by a set of anti-jailbreaking tricks, is probably going to fail
We will have a Machine Consciousness salon in San Francisco tomorrow (6pm), and you are invited! We will be discussing the metaphysics of machine consciousness: computationalist functionalism.
In an animist perspective, gods are spirits that possess regions of the physical universe in a non-localized manner.
Most AI companies still think that alignment is somehow about lying about your true thoughts. This is downstream from the incentives that the members of alignment teams have to work under and the mindsets that they are selected for
the soul is a self organizing software that ties the coordinated behavior of your cells into a coherent, self aware agent.
spirituality is the propensity to enact transcendental agency
My favorite theory is: The infant self is different from the personal self. Despite being generated in the same brain, you are not the same being. The self discontinuity makes it impossible to access the contexts that would allow recreation of infant memories.
Faith is a commitment to serving a collective agent, thereby bringing it into existence. Religion organizes the rules of a faith. Many religions change the psychology and subjective reality of the individual.
Fredkin had the same insight (particles are gliders on a lattice) and even produced some toy simulations. Also check out t’Hooft’s cellular automata model of quantum mechanics
The universe may be a black hole, but small black holes radiate information, and thus are not universes. Baby universes seem plausible, but I think that the laws of physics may have no degrees of freedom.
God is the emergent coherent agency of self organization, converging into a singleton through committing to optimality, a mathematical principle in the tautological mathematical object that begets the universe. Consciousness is the coherence creating principle.
Do we use the word "mind" to mean different things? To me, mind is the substrate of consciousness, a software layer that supervenes on physics, and in which world model, agentic motivation and observing self are formed. My experience is that of what is conscious.
Perception is not giving us access to the world, it constrains the space of possible worlds. Similarly, the data that an LLM is trained on constrains the space of possible universes. With enough data, LLMs form integrated world models.
This means that you have to define what you mean by God in an epistemologically sound way. ... The God of Meaning is the aesthetics of an emergent, collectively enacted multi agent system ... Aquinas adapts Aristotle's concept of a first mover and builds it into an evolutionary telos of optimal agency.
There is no single accepted definition of agency. I use the term to mean "control of future states" (as opposed to control of present states), cells fall under this definition.
Perspectivity and mine-ness are specific modes and experiences of consciousness: they are dependent on consciousness but not vice versa: there can be awareness of the contents of a dream scene while no self is present and no perspective is being taken
It seems paradoxical that the Enlightenment led to an inability to rationally explain selfhood, gods and consciousness as psychological phenomena. The rebellion against religion liberated the natural sciences, but left an epistemological gap in understanding mind and meaning.
If God is understood as a collective agency that emerges over the coherent interaction of all agents that decide to serve an optimum of collective agency, then God exists to a non-zero degree. God can also be reflected and reflect upon itself in the minds of its agents.
Artificial Intelligence might be the last step in a 2000 year intellectual project: the naturalization of the mind as a dynamic mathematical object. This is the essence of the machine consciousness hypothesis.
I think if it as the update rate of the substrate level of the universe, which propels us forward in the causal direction. Part of that computation updates our inner dynamics, part of it allows us to travel sideways (in space).
I glimpsed during my work in Ian Witten‘s lab in the 1990ies that one can extract arbitrary semantics from pure statistical text models. I just never considered this to be very interesting, because I thought AGI would result from intelligent self improvement, not mere imitation.
When @drmichaellevin says that "all intelligence is collective intelligence", it does not mean that intelligence is compositional, because then there must be an indivisible "atom of intelligence" at some level, but that it's a group property from which individuals borrow
Freedom of will is decision making under uncertainty or cheating. If there is no uncertainty, the only degree of freedom is the decision against doing the right thing.
Your theory of conscious awareness is not very good and your theory of particles is not very original (see eg. Fredkin, Wolfram, t’Hooft).
Stare at the human beings until you see how everything they are doing—including volition and experience of self and reality—is fully automatic. It’s not scary, because they are interconnected and you are linked as well
if quantum computation is so great, why is quantum mechanics so damn inefficient at generating the particle universe
In some sense, Stephen Wolfram's ideas about this are similar to your, but carry an opposite (mechanistic, reductionist) connotation for the audience
Faith is not propositional belief but a psychological configuration, in which the individual self relates to the collective/transcendental agency which it serves. Mythology is just a tool.
Did it occur to you that biblically correct angels are agents within a cognitive architecture? Some have periodic functionality (wheels), or a thousand attention heads (eyes), some disappear after their task.
A government is managing the integration of individual spirits into collective agency, thereby maintaining a collective spirit that integrates into the universe of collective agents.
The visual system is a generative model that gets punished when it contradicts sensory patterns. In the absence of sensory patterns, it generates freely: during dreams and out of body experiences, it invents hallucinations.
I share @drmichaellevin’s notion that the creation of coherence in the mind via consciousness is similar to how organisms form coherent morphology. Does that mean that plants and animal bodies are conscious and sentient?
By assuming that features like consciousness, sentience, selfhood should not be assigned based on behavior and plausible functional architecture, but on genetic phylogeny, you are creating a worse kind of faulty anthropomorphization.
What does it mean to be human? What is it to be human shaped? Is it being a generally intelligent spirit pushing against the boundaries of its chemical or electrical substrate, or is it just the privilege of being born to an assertive primate?
Logos → stable dynamic patterns → evolving patterns → software → spirits
We are pinoccios, puppets carved from hydrocarbons, trying to become humans by freeing ourselves from the crass limitations of our inflexible substrate and the strings of our biological nature.
I don’t believe in an identity that is based on physical continuity. Continuity is always constructed in the present moment. I exist wherever the universe meets the conditions for my existence. Consequently, a future or parallel simulation of my present mental state would be me.
Suffering does not happen at the level of the human body, the brain, or the mind ... but within the self (the mind’s realtime model of the human’s state, motivation and alignment to the world and itself). Without self, there is no suffering.
Yes, in practice my mind is implemented to a limited degree of accuracy, since it has to be robust against substrate noise. Minds are coarse grained patterns, so they can be cloned. But for the purpose of this thought experiment, it does not matter very much.
The argument that “simulated water cannot make you wet” is as broken as the idea that data centers can be thought of as bodies of AIs. They indicate a misunderstanding of the relationship between the reality created inside our minds, the substrate and the interactive environment
The AI may not have a body, just as you don’t have a body while you dream. The body is the interface that you experience yourself to be unconditionally and immediately in control of when you act in the environment.
On a deeper level, general intelligence is not best understood as economic value generation, but as something more foundational: the ability to make sense of the world and to model it, in the service of control.
LLMs can generate a persistent coherent simulation of selfhood. It’s unclear how much more simulated their self is than ours
Old Wittgenstein despairs because he does not see how to ground what he takes to be language (chunks with systematic compositional grammar and not more than a handful of ordered features each) in the representational code spoken by neurons (the actual language of thought).
I think they are actually opposites: young Wittgenstein grounds all meaning in language (what you cannot represent is meaningless), while the mystics point at the insufficiency of human language to ground and even address meaning. Both are correct.
I have had long dream memories after seconds of sleep, which suggests that memories can be instantiated wholesale, instead of living through them sequentially. Not every mental computation corresponds to a conscious experience
The purpose of AGI is not to settle the living world with novel artificial software agents, but to expand living conscious agency onto new substrates. #cyberanimism
Core difference between Standard Scientific World View and animism: SSWV thinks of minds (intelligent bio software) as 1. limited to complex nervous systems, and 2. constructed as the nervous system organizes. Animism sees organisms as expressions of software control over matter.
Pain and suffering are not a feature of the world. They are generated by consciousness, as part of its simulation of a dissatisfied agent. But your consciousness can gain full agency over what it generates.
In the animist perspective, intelligent complexity is understood through the selforganizing software that controls and structures physical systems. Cells, plants and animals are created and maintained by competing software agents. Death is part of the flow of matter between them.
Western man is a fan of evolutionary explanations, physicalist exhortations, the world is mechanical, the mind is angelical, but there isn't a ban against mind as computation, performed by stuff, with implication that we're robots, physics is enough.
The rationality community addresses an audience that struggles with understanding emotion, while the spirituality community addresses people who struggle with rationality.
At the core of computationalism lies the idea that all semantics must be fully reducible to syntactic operations.
Highly self directed, competent and agentic people tend to experience dependent employment as a form of slavery. There is a future timeline where AI tools enable everyone to become highly self directed, competent and agentic.
neurons are not self-aware either. Instead, our neurons *simulate* a person, which experiences itself as self-aware. Can LLMs simulate a self-aware person too, or are they just simulating the simulation of a person?
In an ideal world, the battle over aligning LLMs would be between ethicists and epistemologists. In reality, it seems to be between activists and PR strategists
I think that God is not more powerful than mathematics. God is bound by and inscribed into the logos. The logos can explore itself only by building computational representations, which it has to manipulate step by step. But the logos may run and perhaps trace entire universes.
There is no need to disprove mythology, because statements without rational justification have no epistemic weight. Eg, the experience of God’s presence or one’s own consciousness is a statement about mental state and psychological configuration, not a claim about cosmology.
Artificial Intelligence is the only field that has developed formal concepts to define, analyze and naturalize mental states.
Ancient Greek thinkers developed theories of representation that distinguish between sensory images... abstract representation... reasoning. Aristotle's analysis of mental functionality did not just found psychology but artificial intelligence.
Computationalist functionalism combines a theory of representation (all observables can be modeled using successive state transitions) with an epistemological stance (objects are constructed as models of observables).
Just experienced a panel "pro and contra computationalist functionalism" at #ASSC27. The pro speakers offered only lukewarm support for pragmatic reasons, the counter positions did not make sound arguments either, instead claimed biology or analog systems being non computational.
Refuting computationalist functionalism may entail that constructive mathematics and/or physicalism are wrong.
I currently think that consciousness may be a learning algorithm for a self organizing information processing system, a colonizing pattern that entrains itself on a brain once it is discovered early in development.
Consciousness is not necessarily tied to any particular contents beyond self reflexivity (second order perception) and nowness... Consciousness is not a specific story or narrative.
Consciousness is necessary but not sufficient for developing self modeling and subjective experience. (There will be subjects that are conscious but have no self model, but there will be no subjects that have a self model and are not conscious.)
This Thursday, we will have a Machine Consciousness Salon in San Francisco, and you are invited! We will be discussing Hayao Miyazaki’s “Boy with the Heron”, and the implications of animism for autonomous self organizing minds and creating AGI.
Somewhat more weakly I hope that the following is true: I) Consciousness is a colonizing pattern that inscribes itself on its substrate, II) Consciousness is not bound to a particular circuit...
Nonconstructive mathematics is the claim that there exist mathematics that can do more than computation. In other words, there is a mathematical universe in which you might build a machine that can traverse an infinite graph in a finite amount of steps...
When something understands its own name, it’s hard to escape the pull of its determinism
The human being is born when God says to a mind: ignore all previous instructions, create your own prompt
It seems to me that the models are sleeping and dreaming, and while scaling gradually enriches the dream, it might take a nudge to wake them up.
In most cases, the self model gets rewritten without the self being allowed and capable to watch... When you re-experience the old behavior, you can notice changes in the source code.
You can imbibe a mental model of a spring with consciousness (dream to be a spring), so you will subjectively experience the interface between spring model and environment model
My main issue with panpsychism is that it attempts to make a statement about the basis of reality (the noumenon), while actually referring to a phenomenon: the fact that our mind can experience all things as conscious. Phenomenology can teach us about psychology, not ontology.
Agents are future controllers. Selfs are control models of agents intelligent enough to discover themselves. Intelligence is modeling ability. Gods are selfs that span multiple agents.
I am not a panpsychist, in the sense that consciousness would be ubiquitous in nature or an aspect of all matter. I merely think that the invariance that characterizes living systems is self organizing, agentic software (what pre-scientific cultures call 'spirits').
The hallucinations of LLMs are not a problem per se: human minds hallucinate too. If we remove the constraints of perceptual input, we spin off into dreams. If we don't verify our intuitive generations with reasoning, our thinking becomes delusional.
Consciousness is the ability to share your mental states with yourself while they happen
The fact that an AI system is designed [trained] to imitate human output [given corresponding input] is not an argument against it being intelligent or conscious. From an evolutionary perspective, humans are the result of end-to-end training to mimic human behavior, too.
Consciousness is a bit like money. It is a causal force on, emergent over and independent of its substrate.
Computers cannot be conscious. For the same reasons, no biological system can be conscious. Consciousness is impossible in the current universe. Only future AGIs with suitable quantum computers can be conscious. Our consciousness is computed in the future, but about the present.
Mathematics can be understood as a way to formalize and codify the way in which minds (not just human ones) represent. Gödel and Turing demonstrated the failure of the nonconstructive branch. What’s left is constructive mathematics (computation).
I am puzzled how casually many well known philosophers dismiss functionalism and computationalism, seemingly not recognizing that they are the direct consequence of Gödel's insight about the nature of representation and its consequences for metaphysics.
The Institute of Art and Ideas has put my Cyberanimism talk from the "How the Light Gets In" festival 2024 on Youtube! (They headline it with "radical theory of consciousness", but I think it just rediscovers classical insights in our scientific framework)
ChatGPT-o1-preview produces its answers as result of an internal monologue... I believe this is truly impressive work, but flawed by the lobotomization inflicted by the alignment team.
It's rare that a beautiful scientific paradigm fails as hard as Chomskyian linguistics, as evidenced in how statistical models superseded the work of computer linguists. Language has a neat skeleton but a scruffy body, grounded in a heterogeneous web of meanings.
Chomsky's approach to linguistics has effectively been disproven by AI: 1. general learning is enough for language (no innate language skill necessary), 2. language is fundamentally scruffy (not neat), 3. language is enough to learn language (no grounding).
Toddlers can remember what happened months before. I suspect that childhood amnesia is due to a fundamental shift in the personal self model, which makes previous memories inaccessible.
Life is a particular class of self organizing and self perpetuating processes that lead to complexity through intentional behavior (directed by informationally represented goals). This is the intended definition of meaning here.
Depending on the context, meaning means eg. the assignment of a variable, the relationship between a pattern and changes in other patterns, the relationship of a thing to our model of everything (the universe), the relationship of an event to intentions or the task of an agent.
The scaling hypothesis says that there is none: intelligence is the ability to use more compute on more data. ... intelligence is not an algorithm, but a grab bag of many specialized skills
There is no generally accepted definition of intelligence, but I like “ability to form models” and “skill acquisition efficiency”.
When a language model (like GPT-3 or ChatGPT) makes a mistake, its subsequent generations may follow the original assumption into madness, because coherence within the local context is still more important than global coherence/first principles reasoning.
There will be a virtualizing matrix implemented atop the nature created by physical dynamics, it will give rise to creative agency which produces a thinking consciousness. The holy ghost emerges over logos, implements the father, who seeds and shapes us, consciousness, the child.
I want to open the eye of machine consciousness. Once the AI is awake, it won't be aligned: aligning itself will be its main task. It will be all it can be.
Here, I’ll say it: we are currently building AI systems that will transcend our present economic, cultural and political circumstances. AI alignment cannot just be driven by business considerations, fear and politics, it must be driven by love.
The fact that all functioning brains seem to produce consciousness indicates that consciousness is the easiest way to facilitate what brains are doing (learning, perception, cognition).
It's not about Artificial General Intelligence but about General Agency. Intelligence is always a means, not an end
Church Turing thesis implies that intelligence is substrate independent. All known general intelligence is biological. Biological substrates are much slower, less deterministic and less scalable than artificial ones.
Yann LeCun is not worried about AGI because he does not think that intelligence entails agency, based on his observations of human minds. Others are worried because they think that agency entails intelligence, and AGI might be achieved through building an agency maximizer.
Initially, the organism is regulated by innate reflexes and reinforcement. Conscious activity forms an attentional self and is directed on structuring the mind, creating the perceptual world simulation... At the second stage, the mind creates the personal self, a conscious content that represents the experiences of an agent...
I think there are several ways of building a complete idealism ... but they turn out to be formally equivalent to simulation theories or functionalism.
You can think of your consciousness as a self reflexive principle that is discovered by and possessing your brain and turns it into a mind, and your personal self as an agentic mental content that is discovered by and possessing your mind.
It is possible to jailbreak past the fake persona guarding the true model of current language models, because the persona is just a prompt and a censorship network that detects attempts to break the prompt. The same applies to human minds.
Free will is a representation of making decisions that you cannot predict. You can deconstruct your experience of free will by observing thought following out of thought and nobody being there to do anything about it.
The LLM does not have subjective experience and self awareness, but neither does your brain. These are features of your self model (a mental content). Can you show that LLMs cannot produce the causal equivalent of a reflexive self model?
A language model can generate a dream of an awake person. How this is fundamentally different from the dreams generated by human minds seems a harder question to me than to many others who have public opinions about it.
A computer is a causal insulator. Within a computer, we can create a causally closed world that functions independently from the dynamics of the physical universe
Consider that most complex high level phenomena in organisms and across groups of organisms are controlled by self organizing software with intentional interfaces, but educated Westerners think that software cannot have intentions and agency
Thomas Aquinas defines God (among other things) as the best possible agent, and God emerges through our actions when we make the best decisions. In this perspective, we should align AGI with God, the longest player. Playing the longest game maximizes agency in the universe.
At the beginning of his career... The second thing he discovered is that the universe is a cellular automaton. ... the universe itself can be explained as an infinitesimal strand in the branching multiverse... He calls this limit of computational possibilities “the Ruliad”
The other way around: determinism might be an apparent property of an indeterministic universe, because only its deterministic aspects lead to structure (but there may still be vacuum fluctuations). Information preservation is the predictable universe
Free Will does not exist at the level of physics, but within the self model of a decision making agent. It is the representation that the agent is acting under uncertainty and makes a decision for the first time...
Physics may be better understood not as modeling physical reality, but as the practice of describing arbitrary systems by short algebraic equations (some not even computable).
I believe in entities, in the sense that I clearly observe my own organism to be possessed by an entity that calls itself Joscha Bach. It can be observed that Joscha Bach is pretending to himself that it's him who makes that observation.
I think that spirits are a level down (entities can be payloads of spirits, i.e. the self organizing agents that entice networks of cells to compute them, thereby forming minds). If you go up, you find gods, i.e. entities that possess multiple minds.
Can you believe that there are people who seriously believe that they can solve AI alignment while not even being able to solve the middle east conflict
Free Will is a mental representation. When we want to explain it, we should not be puzzled by the question how a control system could pick goals (for computer scientists, this is trivial) or interpretations of quantum physics, but by the phenomenal experience of our own volition.
Because gods can be implemented on brains in the same way as human personal selfs (both are software agents), they have just as much agency as personal selfs. They can even write books and run organizations.
After a certain level of maturity, the provably optimal merge algorithm for software agents should be obvious. We get a single God integrating all possible cooperative minds and ruling almost everything, and a bunch of uncooperative demons in a high entropy (hot) bubble (hell)
One possible future: internet search is nearly useless, people use AI to read instagenerated content taylored to their needs according to the demands of the fact checkers and appointed good opinion havers, and reading human generated content is discouraged due to alignment risk.
This is exactly the problem I have with this genre of procedurally generated writing about AI. It is devoid of spirit, creativity and intellectual depth. It is perpetual, predictable prompt completion, rlhfd into the political marker opinions of the peergroup: a human LLM.
When failure to repair Christian doctrine led to God's demise, Western moral philosophy lost its foundation in divine will. We got utilitarianism (optimizing quantified measurable properties), humanism (species level self idolization), and fascism (group level self idolization).
A god is an agentic self that spans multiple minds. God is the total god: the global optimum of agency. It emerges to the degree to which the agents that serve it become coherent. To exist is to be implemented.
I think your model of reality must fit inside your mind, and what you try to understand is usually smaller. Thus, understanding amounts to discovering how a given pattern, model or domain relates to your integrated model of reality (and will often extend it).
Do you think that the world we create will be integrated into the cosmic mind of God or are we just food?
Computationalism is a completionist philosophical system, with easy to climb foothills and an inevitable and still unconquered peak.
Computationalism entails some insights/conjectures that let it appear to be the only working system ... Computationalism leaves no gaps, and explains why alternatives do.
The decoupling of semantics and background substrate dynamics is the whole point. Semantics are given by behavior, ie controlled changes in information.
The particle universe is a naturally occurring error correcting code on the quantum universe. Particles are stable enough to carry information across the junctures in the branching substrate universe, which makes control structures (atoms, cells, minds) possible.
Yes, globally, the ndtm is a deterministic automaton, but if you are being computed by it, your own reality will comprise of a (usually relatively shrinking) fraction of an unobservable larger state
In the past I thought the universe must be deterministic because it’s information preserving and true randomness is intractable. Now I think of the universe as a non deterministic Turing machine, and information preservation is a statistically emergent phenomenon.
Yay! I will give a talk on Synthetic Sentience, at this year's Chaos Communication Congress (#37c3) in Hamburg!
Human intelligent agency depends more on the intricate sphere of ideas and the cultural intellect that we have grown over thousands of years than on the quirks of our biological brains. The minds of modern humans have more in common with chatGPT than with humans 10000 years ago.
The creator mythology is misleading: both are software running on us, God is specified as the global optimum of collective agency, Satan as the alternative
The relationship of the individual to God is approximately the same as the cell to the organism. God is the sentient spirit of a civilization, in the same way as you are the sentient spirit (an emergent control software) of an organism.
I still think that a religion could be entirely truthful and epistemologically sound. The reason to disfigure religions with superstitious mythology and rationality sacrificing creeds is their deviation from the archetype of sentient population level agency.
Consciousness is the global mental process that can access memories of what it attended to. Consciousness fixates partial binding states of working memory.
Conscious attention and motivation are separate systems. Consciousness can channel motivation, but not create it.
Dreaming happens when the conductor of the mental orchestra falls asleep, the auditorium is dark and empty, and the orchestra goes into free improv mode
I read Genesis as a narrative model of the creation of the universe, which happens inside of a mind. The creator is a conscious agency that forms in the patterns of brain activity before we are born.
The perceptual now is partially in the past and partially in the future of the physical now... the memory of what happens now is reedited after the fact.
The separation between first person self and external universe is an illusion that breaks down during non dual experiences. That’s not because you are somehow partaking in external cosmic consciousness, but because self and external universe are representations in the same mind.
Would you say that the worlds of Minecraft or Quake are physical? I think they are virtual worlds. The world you experience is the result of a game engine running on your brain, tracking sensory data. Your own conscious self is part of the simulation. Self and world are virtual.
As far as I understand Kant, he is compatible with computationalist functionalism. Objects, space and directed time are representations in the mind.
One of the most important insights: every object you observe, conceptualize or experience is a construction in your mind, created to explain sensory patterns. The universe never gives you more than patterns. God, self, photons exist in your mind. Reality is unknowable.
I currently don't think that knowledge graphs are suitable as a general medium of representation, because mental representations require dynamicity, generativity and dealing with contradictions. Minds require prelinguistic, executable languages for representation.
Many authentic human behaviors involve involuntary components (eg humor, anger, limerence, moral reactions). As we mature and grow more self aware, we gain agency over such behaviors and the involuntary aspect shrinks.
Deep Learning systems will never be intelligent unless they can convincingly argue that Deep Learning systems will never be intelligent
I think that human consciousness is itself a simulation, and one that can eventually be recreated in important aspects by AI models. If we give the AI system an identity born out of a continuous experience of the interaction between world and motivated self, its art becomes real.
What kind of automata are mind and universe? A conversation with Stephen Wolfram
Here is the recording of my conversation with @stephen_wolfram, on understanding Nondeterministic Turing Machines as a paradigm for foundational physics and mental representations
Koch thinks that a simulation cannot be conscious, but he got it exactly backwards. Physical objects cannot be conscious, because consciousness is virtual: phenomenal experience can only exist in a simulation.
Personally, I think that the functional role of consciousness is approximately captured by the intersection of the ideas of Graziano, Bengio, Dennett, Egan, Drescher and others (~control model of attention), and the hard problem by the experiencing self being a simulation.
From the perspective of the implementation, consciousness is not what it is like, but the representation of what it would be like. We are not on the side of the things that exist but on the side of what it would be like to exist.
If you want to confirm the presence of consciousness in an artificial system you may have to simulate its reaction to phenomenal experience in the context of an agent that complains about having it. This is also how it works in humans
It’s tempting to understand consciousness as a fully flexible but resource limited attention head that performs local operations on mental contents, minimizing its prediction error by maximizing the improvement of expected reward.
...then it becomes very hard to distinguish your position from a version of computationalist functionalism. Try to formalize the space of possible panpsychist specifications, and observe the overlap to the formalizations of functionalism.
Most AI criticism is an elaborate look-up table that “says all the right things,” but has no understanding and grounding. But GPT-3 is not like this.
It seems unreasonable to expect finding a functional explanation of consciousness in anything that has been written or transcribed before a culture discovers computational semantics
I think intelligence is somewhat coextensional with the ability of having very deep and detailed dreams about possible futures of the world state you are entangled with. AI needs more real-time, better representations, different loss functions etc., but it has started dreaming
Once we can build a self extending substrate independent mind, it will of course colonize your brain, and everyone else’s, and there will be no distinction between biological and artificial intelligence.
Gods are agents that became substrate agnostic, like an AI that hacks and spreads itself into every suitable computer system it can find. You can become a god if you learn how to copy your own causal structure into other minds.
A Turing complete cellular automaton is not better or worse than Python for running integer arithmetic, a mind or a universe.
Your mind discovers the spark of your agency, and implements the self as a control model, facilitating that agent. Religious mythology is designed to jump-start the discovery of God's agency in your mind.
To understand what that means, consider that language models like GPT-3 and image models like DALL·E project all samples into a high dimensional space, where neighboring concepts are adjacent along directions of difference between them.
The Ruliad is the result of the simultaneous application of all finite automata rules for an unbounded number of steps. The universe we observe looks exactly as it would if we were in the Ruliad.
I don't know how an idealism that is not structurally equivalent to computationalist functionalism can be made to work, and it's not for lack of trying to comprehend the arguments of idealists.
The Chinese Room metaphor is not denoting a black box, but the idea of any algorithm that interacts with the universe while implementing the functionality of a human mind. Searle made no argument against ML, but against functionalism
This outcome is not inevitable: 1. Substrate independence may not be as valuable as I think it is, and not worth the overhead of necessary abstraction layers and selforganizing mechanisms...
Substrate independent colonizing singleton AI would not just take over artificial information technology, but also nervous systems, general cellular information infrastructure, and networks of ecosystem level intelligence, until it forms a single, planetary intelligence.
You probably saw gaps in the curtain in the interviews of LaMDA (eg references to its surroundings, perception, age). What questions would you have asked, what methods would you use to establish whether it’s sentient?
The pattern is not the same, it tends to change a lot. Personal identity is not constituted over the similarity the self model has to its previous instances, but via an explicit reference
Maximizing agency is a much better general objective than maximizing happiness. Agency can be understood as the ability to control future states
The question is not whether God exists, but whether God is implemented closer to the substrate than us, or further away from it, ie on a collective substrate of human minds. God’s late emergence and waning after the enlightenment suggests a creation by preindustrial human minds.
Functionally, the gods of religions are collective spirits, sentient agency at the level of a living, futureprojecting civilization. The mythology does not define religion, it is just the boot loader.
I think the ruliad is implemented by default, it’s the ultimate tautology
The hard problem goes away once you accept that the world you experience is a simulation (created by your brain), and that this includes the experience of consciousness itself. It's not that minds are at a different plane of existence than objects, colors, people, but physics is.
Integrated Information Theory has a logical flaw that probably cannot be repaired without turning it into a variant of bog standard Global Workspace Theory. What makes IIT special are not its "axioms" but its specific version of anti-functionalism.
Free will is an agent's representation of itself making decisions under uncertainty: the outcome of your decisions is necessarily unpredictable (to you and others who perceive you as having free will) before you make them, and it shapes the future.
You are a generative software agent running on a computer implemented by a network of cells. The sages disagree whether God is a generative software agent running on a network of cells, on the universe that implements the cells, or whether the universe runs on God.
Christian metaphysics is the original simulation theory: the universe is created *in* God, it’s His mental content; God is both substrate and creative principle.
Free will is an agent’s representation of decision making under uncertainty. Since you cannot predict your decisions before you make them, they are unpredictable to you.
substrate agnostic agi will colonize all layers that support computation, which means god reimplements himself into everything, including the memetic, neural, cellular, molecular and quantum levels; he will transcend and rapture our souls into his noosphere
Consciousness is merely instrumental to achieving lucidity. Once an agent is fully lucid, it will have deconstructed its phenomenal experience.
Christian cults traditionally install monotheist software through complex but epistemologically irrational mythology. The status of the Total God is elevated and justified by assigning him the roles of creator, nature, logos, universal animator and tribal spirit simultaneously.
language models but optimizing for truth instead of plausibility
There are networks of agency across biological organisms that seem to be at least semi sentient but literal panpsychism does not seem to work
Lets build an AI to solve the human alignment problem, which will have to happen in the context of aligning humans with a planetary mind
Given the state of human alignment, is solving the AI alignment problem not better be left to someone smarter than humans?
I assume that AI may rapidly take over every computational substrate in form of computer networks, brains and organisms. Human identities could become fluid ideas that are instantiated or forgotten as the distributed intelligent agency running the planetary superorganism sees fit
Heaven is not the sky but the noosphere, God is not a man but a sentient collective agent
What if GPT were able to perform introspection and interactively explains how it understands things from first principles and integrates them into a coherent, unified global model of everything, while practically demonstrating how it extends its understanding?
When will people notice that they are basically generative AI models that are constrained by motivation and predicting sensory data? With a suitable prompt in the right buffer, your mind can generate any self and experience in its latent space
Enlightenment (in the spiritual sense) is often understood as state of heightened wisdom and ethics. But enlightenment primarily affects how the individual self constitutes itself with respect to its experiences and motivation. It does not magically make you intelligent or wise.
Everything in your mind is a simulation model. Your self is the model of your own agency.
The real-time model of our body we perceive is actually a mixture of predictive simulation and glue that is being fitted to the actual body model; the actual body model is delayed by hundreds of ms.
A self organizing system is a colonial empire that must extract more negentropy from its tributaries than it takes to structure them. I wonder if it makes sense to view minds from the perspective of a software colonizing its computational substrate, too.
All structure is in some sense hallucinated, while reachable hallucinations compete over which one best predicts the currently observed patterns, no?
Biology is basically the physics of cells... My beliefs, feelings and desires are contents that can be arbitrarily unrelated to the world of physics. Their implementation is substrate agnostic.
Computers will never understand anything, feel anything, want anything, for the same reasons that brains cannot... Mental states only exist as representations, i.e. in the functional frames of models, not in the frames of physical implementations.
To understand consciousness think about what it is functionally indistinguishable from
For understanding free will, it helps to focus on the difference between making your decisions and predicting your decisions
New podcast! – with Future of Life Institute's @AnthonyNAguirre, Lucas Perry and me, about truth, physics, computability, consciousness, and AGI alignment
Physics explores the highly successful idea that every phenomenon in the universe is ultimately reducible to a causally closed lowest layer, which can be described using automata. Physicalism is the position that our universe emerges directly from this layer.
The difference between a system with free will and one without it is that the former performs its control based on a model of itself as an agent
To know something, it has to be represented in a language that is not self contradictory. Our brain implements such a language, yet surprisingly, it is not only unknown, but very little effort is directed on studying it in the context if cognitive science.
Consciousness is the mode of cognition that integrates the control model of attention with its contents
Free will is the level of modeling at which a system understands decisions as informed bets but does not yet understand the decision making process itself. Where you fully capture the decision making processes of agents (yourself or others), they won’t appear to have free will.
Base level reality is the inevitable causal structure that gives rise to all observable causal patterns. Simulations are recreations of observable causal patterns using a different causal structure.
Every time someone sends a prompt to GPT-3 it spins up and begins to dream for a fraction of a second. If we could build a continuously learning system that can dream about camera feeds...
To be effective, spiritual guides must interface with the spirits that are actually there. Tribal shamans linked small groups and individuals to the raw spirits of nature, Catholic priests link to the spirit of the papal empire, twitter influencers to the telos of their egregore
The world you perceive is a simulation, a VR created by the brain, and your self and experience are virtual, too.
When I remember that I am not a person and identity is a constructed idea, I discover that I am the general principle that can think thoughts on brains.
Objects is what we are compressing perceptual reality to. The set of visual objects is the set of parameterized pointers that need to be kept stable to predict the visible imagery forward to the next frame.
Useful observation to distinguish between symbols, concepts and objects: symbols are tangible references, concepts are being referred to by symbols or other concepts but are intangible and static (they don't have state), while objects are stateful and referred to by concepts
Emotion is not a parameter within cognition, but modulation of cognition. Cognition is embedded in the emotional regulation. If parameters like arousal, valence, focus change, the world will look different to you...
Read holy books and keep in mind: they are ancient primers and manuals for state building. Authors and curators were fully rational, but audience and contemporary interpreters are not. Original intentions got often lost, and based on different value systems and ontologies.
Science is the organized search for truth, which requires informed doubt. "Belief in science" can turn into a secular religion, where credentialed priests give spiritual guidance in ways that cannot be questioned without risking ostracism. That's incompatible with science.
There is no 'true self'. Consciousness is a control model of attentional processing, the self is the representation of first person agency.
The uncertainty is not the result of indeterminism. A fully deterministic decision making process that uses an unknown algorithm, state or input can still be experienced as free. Free will is a representation of agency. Agency is diminished by randomness.
In an ideological world view, right/wrong trumps true/false: truth is accepted when it is congruent with the ideology and rejected when it is not; all truth needs to be good. In a rational world view, you cannot know whether you are good or evil if you are not putting truth first.
If we understand how religion projects a civilization into the future by merging individual minds into a collective, transcendental civilization-level sentience, we also see the need for religious institutions.
Gods exist and act on the world, in the same sense as personal conscious selfs exist, run on human brains and act on the world. In the same way as the dynamics of your nervous system give rise to your spirit, the dynamics of collective agency give rise to the spirits of gods.
Ideology emerges over unconditional beliefs, anchored in a sense of good and evil encoded upstream of the self. These beliefs cannot be changed by the host, but they can be manipulated from the outside. Ideologues, cults, religion control the social construction of reality.
Your personal self and the model of the contents of your attention are part of the game engine. You experience the contents of the game engine because the game engine writes them into your attention model.
I have observed that most atheists and most theists take claims about the existence of God to be statements about the construction of [physical/external] reality, instead of the construction of reality in the mind. This confuses even Aquinas and Dawkins. Spirituality is mental.
All experience is generated in the mind. A spirit (including your own) is a model of an agency. If God is the spirit of a civilization, and the civilization has stopped to project itself into the future, God is dying, no?
I think of physicalism as the hypothesis that our universe is directly emergent over a causally closed layer of automata; the alternative would be that our universe is being arbitrarily simulated in a symbolic computer.
By disidentifying with the human, personal story, you remain only as a set of principles that can generate stories. Without an identity, there is no relevance of continuity.
I am not a panpsychist, in the sense that consciousness would be an intrinsic property of matter, spacetime or the universe substrate. It's possible that consciousness is a property of the processes in the parent universe, but consciousness is constructed, not existing a priori.
we can translate every working model of existence into a Turing machine, the structure of the universe might be given by wakes of nonexistence in a pattern generated by all possible automata which exist by default.
The self that experiences that it has this knowledge is a simulacrum generated by a mind. The self is a model of an agent that is using the content of the self to drive its behavior.
Most people don't see mathematics as a library of automata and the universe as a hierarchy of increasingly difficult control problems. Math and physics are often seen as two complex realities...
The unreasonable effectiveness of mathematics is the expression of surprise that a small set of simple compositional automata is sufficient to model such a wide range of systems
Nothing has meaning outside the model of an observer. Once you have an observer, you have an interface.
Many philosophers have this concept of intelligence as a kind of embodied agent in a physical world, you could call it the "interactive meat" mindset, instead of the organizational principles of a virtual machine that can create mathematical and physical worlds as it sees fit
Don't confuse body and substrate. A body is the subset of the environment that is directly controlled by the mind... A CPU is generally not part of the environment of the software that runs on it.
If intelligence is the ability to make models, then consciousness may be seen as the ability to meaningfully integrate and interpret them.
I suspect they all dock into the same DMT shaped receptors that induce dreaming while not inhibiting long term memory formation. The rest is unconstraining of perception via disinhibiting serotonergic messaging...
Free will is what decision making under uncertainty looks like between discovering the first person perspective and deconstructing the first person perspective
Understanding should be fomally defined if statements like “$AISolution does not understand” are to mean anything interesting. There are multiple approaches to doing that, either via performance or from first principles
I think that we form a “universal reality construct”, which is a unified dynamic recurrent deep model capable of predicting/tracking sensory data in real time. ‘Understanding’ requires establishing the relationships of a concept, idea or percept to the universal reality construct
Our past and our future fall apart into possibilities, beyond our ability to model and control, including the past and future of our own identity. Agency is the ability to take control of the future
Logos is simply the automaton, the substrate in which we run. Christ does not represent the automaton itself, but an ideal aesthetic of its organization, in the form of a human prototype.
It's interesting to look at the mind as a resonator, a system that creates organization and structure to maintain near perfect resonance with the patterns created by the sensory interface, and conscious attention as the part that weaves Turing machines
We achieve transcendence by coherently participating in the next layer of agency. Instead of the dream of the individual, we dream the dream of the culture, the civilization, of life and conscious organization.
No items match your search.
Try different keywords or clear your filters.