Ben Goertzel

My and Gabriel Axel's new book "The Consciousness Explosion: A Mindful Human's Guide to the Coming Technological and Experiential Singularity" -- now available via Bookshop amazon.com/dp/B0D8CYC2NM?… -- Amazon/Kindle bookshop.org/p/books/the-co… -- Or free PDF theconsciousnessexplosion.ai

Beyond Human Comparison: Extending DeepMind’s AGI-Test Framework with Imagination, Beneficial Agency, and Pressure Robustness ... bengoertzel.substack.com/p/beyond-human…

Quantumity is in the relation to the observer!! -- It sounds weird but the math bears it out -- sometimes a smaller system should view a larger "classical" system as a "quantum" system... Furthermore, a sufficiently complex classical system should often model ITSELF as a quantum system (roughly: because its deliberative core cannot fully perceive or model its own overall state). I lay out the arguments in detail here: open.substack.com/pub/bengoertze…

I've been playing a bit with some new ways of thinking about (classical & quantum) brain dynamics: bengoertzel.substack.com/p/rethinking-c… ... Deep neural nets are great as far as they go but they leave a lot of what happens in brains on the table. Via mappings of optimal-control equations into fluid-dynamics equations on the one hand, and quantum-mechanics equations on the other... one can model the brain as a synergetic, resonating combination of a "fluidic neural network" and a macrosopic-quantum-network ... where the quantum aspect is generally weak but can potentially be consequential when the fluidic network reaches bifurcation points.... Speculative stuff but I believe consistent with known neuroscience data -- and also giving us a reminder that we could really use a lot more neuroscience data, our understanding of brain dynamics remains quite limited... While my current preferred approach to AGI (Hyperon) is very much not brain modeling based, there are still some lessons for Hyperon AGI from this sort of investigation -- but I'll lay some of these out in a follow-on post...

Yes. The only people who will not be able to find anything to do post Singularity will be those who are depressed and unimaginative, and so stubborn they refuse to let AGI doctors give them a brain tune-up ;p .... Whether the money economy in its current form survives the Singularity is not the most important point -- and if it doesn't, the only ones who will really miss it will be the handful who were super-rich in the Before-Time and regret the loss of their (often tho not always ill-gotten) "special status" ... "Doing rewarding activities alone and/or together" will continue to exist, "networks of exchange among humans" will continue to exist... and the form these things take in the Era of Abundance will be splendid-- --- Unless that is we manage to botch the Singularity launch via creating greedy or violent AGIs serving narrow ends first, instead of compassionate and loving and generally helpful AGIs ... if we botch this aspect of the launch (which some folks are trying hard to do!!) then the outcome feels less certain to me...

Many people believe that there is a fixed amount of work in the world, and if we give these jobs to machines, humans will not have jobs and starve. This intuitive model of economics is fundamentally wrong. Our wealth depends on the amount and quality of goods and services we can produce and distribute among each other. Automation allows us to make more of everything for everyone. There is always more to do for us, things that we could not afford to do before automation allowed us to get away from the important drudgery of agriculture, manufacturing, and now documenting, calculating, evaluating, memorizing and so on.

I'm not usually one for metaphors, but working on tech these days with increasingly impressive AI assistance reminded me of something: I once had an ebike where you had to pedal to get started, but then after. you reached a certain speed, the motor would take over. Working on AI R&D these days feels like that to me. We've been pedaling really hard and getting some good exercise and exhausting ourselves, but now we can feel the motor starting to kick in... which means we need to pay a higher percentage of our attention to steering! But where the analogy breaks is, in this case, once the motor fully kicks in, so will the self-driving function that steers better than us -- so that once the thing is fully revved up, the only point in it ever letting us steer OR pedal would be just for the lulz...

Evidence is to Logic what Energy is to Physics ... from a simple analogy/mathematical-mapping to a derivation of Quantum Logic Networks for general scalable quantum reasoning on quantum computers... open.substack.com/pub/bengoertze…

LeCun's "Superhuman Adaptable Intelligence" is a special case of AGI , not an alternative... in-depth blog post by me, bengoertzel.substack.com/p/lecuns-sai-i… Taking familiar ideas he has long rejected and re-presenting and rebranding them as his own, is sorta par for the course for the glorious @ylecun of course... @GaryMarcus knows that story ;) ... In this case, his SAI is straightforwardly shown to be a special case of decades-old definitions of AGI under some special assumptions... and these assumptions are reasonable but not really quite adequate: they leave out key stuff like safety and open-ended evolution...

Where Energy and AI Wars Collide: Why Decentralized AGI Matters for World Peace ... new blog post by me open.substack.com/pub/bengoertze…

AGI-26 research conference, back in Silicon Valley -- SFSU July 27-30 ... agi-conference.org 18 yrs of AGI conferences, 20 yrs since the original AGI workshop ... and at this point we feel soooo close to human-level AGI (which will then bring us ASI) ... LLMs are a powerful tool and component but they don't get us all the way there, and adding in some of the creative approaches from the AGI conference research community may well be what gets us over the finish line...

Hyperseed v2: Toward a Semantic-Primitive-Based Ontology that Actually Does Something ... new blog post by me: open.substack.com/pub/bengoertze… Hyperseed v2 is both a new and pretty interesting philosophy of life, the universe and everything -- with a postmodern, relativist, Buddhistic flavor expressed with an analytical-philosophy++ level of mathematical precision ... and a tool for exponentially speeding up creative reasoning in AI systems, especially for cases that require a combination of rigor, imagination and cross-domain thinking.

and the situation when we launch a big proto-AGI breakthrough from Hyperon/SingularityNET will be much messier but quite different Our core AGi will be open so the military can use it if they wish However, we will of course not offer consulting, hosting, any proprietary Atomspaces we might have . etc.. for uses like autonomous AI killing or mass surveillance And it will be noticed that our core AGi being open allows perceived "enemies" to use it also... One can imagine a "race" to get more compute nodes in the global beneficial AGI network than in private forks created by groups focused on their own narrow aims...

Watching the dialogue btw Pete Hegseth and Anthropic... the shit is getting real.... And one of the realities we can see is: Even if in the US (unlike say China) it ends up OK for a major tech company to say no to helping with autonomous killer robots and mass surveillance .. in the end other tech companies with roughly equivalent technologies will say yes... commoncause.org/resources/pete… One can imagine different flavors of the story: If a company had a truly unique proto-AGI or AGI technology and it was going to take competitors years to catch up, then might the US gov't/military exert more force to ensure this tech could be leveraged according to their own taste? But right now we're not precisely in this situation -- Claude is awesome but not really qualitatively soooo much better than other available LLMs the military can use (though it does have its particular strengths).... Anthropic can presumably make a robust revenue stream from customers who don't insist on the right to use their tech inside autonomous killer robots and unrestricted spy networks... but the most notable thing is that we are not (currently??) (yet??) seeing the whole US AI industry give this sort of pushback....

@IanTimotheos sure happy to be on your podcast send me a DM...

@bengoertzel @bengoertzel We met briefly at RareEvo'23 when I was working with Charles. Now I'm leading an AI Agents community in Waterloo. Would love to reconnect and get you on my podcast and talk decentralized Agents. Interested to squeeze in something like this in March?

Hmm, so I tried to coax a fully human-readable proof of the Four-Color Map Theorem out of a few LLMs, based on the Kauffman/Spencer-Brown direction plus some goofball ideas of my own (e.g. replacing Laws of Form with Constructible Duality Logic) ... (here it is, just for fun: drive.google.com/file/d/1KXX2hX… ). I was hoping I could then formally verify (or refute) the proof, but unfortunately to work around some tricky issues I ended up introducing some quantum topology ideas in the proof, and it turns out we don't yet have nice Lean4 (or similar) formalizations of Penrose-Kauffman colorings and related quantum topology stuff .... So we need to get a little further with accelerated auto-formalization first... Josef Urban and @QonoVuor and others are pushing this forward fast though... We are so so close to automating math at a superhuman level, on a historical time-scale, but there is still a bunch of plumbing to push through... And what I see with using LLMs + autoformalizers to do math is, there is still a significant missing link of creativity .... For instance if this 4-color proof is actually correct, this will be largely due to my own weird idea about how to introduce quantum topology into the proof, which is an idea the LLMs did not come close to introducing themselves, even though they "know" way more quantum topology than I do... However we can likely supply that missing link of creativity with Hyperon... and then...

Turns out Chomsky's notion of Universal Grammar can be formulated mathematically in terms of quantale weakness and logical semantics, and this lets you connect it cleanly with vector embedding models and gives a new view of transformer NN architecture... drive.google.com/file/d/1Gh0m3P…

we have Distributed Atomspace 1.0 and a fully MORK-integrated version is coming soon... we can already run our own open LLMs locally but will soon be hosting them on SNet platform via ASI:create ... ASI:chain which is the real decentralized infrastructure is in Devnet.... so all these things are coming along quite nicely... And as for the control of combinatorial explosion while scaling memory and inference -- YES you are right, that is critical, but we have decades of theory and math and prototyping telling us what path to follow here...

@bengoertzel We are early... Scaling requires: large-scale inference control, optimized distributed Atomspace, independence from centralized LLMs, and real decentralized infrastructure. The bottleneck isn’t agents — it’s controlling combinatorial explosion while scaling memory and inference.

Incipit MeTTaClaw !! -- Simple, elegant OpenClaw-type system coded in MeTTa by the inestimable Patrick Hammer from SingularityNET/Hyperon AI team... 200 lines of MeTTa ... github.com/patham9/mettac… .... It is being improved daily, this version uses proprietary LLM but in a couple days it should be using locally hosted open LLMs. Once ASI:Create or ASI:chain is further along it will reference LLMs hosted on our decentralized infrastructure. There is already some Hyperon Atomspace based long-term memory built in. Right now it doesn't do anything amazing yet, but the point is the architecture is designed for integration of all sorts of advanced cognition algorithms (conveniently coded in MeTTa to work against the system's memory) and also for self-modification.... (Relevantly -- The PeTTa compiler for our MeTTA AGI language is quite fast now, and the MORK Atomspace infrastructure is fast too and can fit a lot of nodes and links in RAM. Distributed Atomspace works but is not fully MORK-ified yet, that will come in the next 1-2 months though. PLN probabilistic logic inference on MORK Atomspaces, including those obtained from natural language interpretation via LLMs, is working OK but we are figuring out which. methods will work best for large-scale inference control.. which will be critical for leveraging MettaClaw's Metta-ness into super-smarts...)

Sooooo --- there has been some commentary online in recent days about my emails with Jeffrey Epstein as released in the “Epstein files,” and I felt compelled to post something to clarify .... I've done so in a 10 min video youtube.com/watch?v=WAJE7y… and also textually on my substack bengoertzel.substack.com/p/goertzel-vs-… Just to warn you though, my interaction with Epstein was basically fairly boring; I had nothing to do with his personal life, I never went to his island or flew on his plane ... I did meet him F2F a number of times, and over the years I did obtain a modest amount of AGI research funding from some of his foundations for some organizations helping with my AGI R&D. I was not in the know about his horrible crimes or I wouldn't have been dealing with him. Epstein's focus on centralizing power in himself was the exact opposite of what I’m after with my current work on decentralizing AGI and building an open and transparent Singularity. However, Epstein did recognize the value and importance of AGI well before the scientific or business mainstream, which is what led me into the funder-fundee relationship with him that feels so super unfortunate in hindsight.