[erikpukinskis]

It's amazing how many people speak about AGI as if it's even a theoretically plausible thing.

It's about as plausible and well substantiated as the Christian God. Which isn't to say it's impossible, but just that it's almost entirely an article of faith. There are a few anecdotes which maybe hint at its possibility. But nothing even approaching a rational explanation of how it might be built.

Also, the model of cognition you are describing (information processing or I/O) is very old, from the 1960s. It was inspired by the discovery of the computer. There are other models, like Ecological Psychology, Embodied Cognition, Distributed Cognition.

It is tempting to draw a box around the brain and posit that it's the most important interface and that all of the important information passes through it. But whenever you break down so-called "cognitive" phenomena, you find that often very little information passes through that barrier. The lions share of encodings remain outside the brain, and for any given animal task, quite a lot of information processing happens outside the brain.

In a weird way, the information processing model is a vestige of the notion of a soul. If you really accept "physical fundamentalism" as OP describes it, then the interface between the brain and the rest of the world is nothing at all. Just a ribbon of atoms with a name. No more interesting than the interface between your stomach lining and the bacteria inside, or between the vibrations in front of your mouth and all nearby ears.

The only reason to center the brain/environment interface is to try to separate what you consider to be the essential identity of a person from their physical grounding. I.e. to maintain a model which includes a soul.

[eli_gottlieb]

>Also, the model of cognition you are describing (information processing or I/O) is very old, from the 1960s. It was inspired by the discovery of the computer. There are other models, like Ecological Psychology, Embodied Cognition, Distributed Cognition.

False dichotomy. As far as I'm remotely up-to-date on theoretical neuroscience both information processing and embodied cognition are correct. The brain processes the interoceptive and exteroceptive information received from the body, computes first- and second-order statistics, and uses those to emit actions that explore-and-exploit the body and environment (including the environment's capacity to process information or convenient structures in the environment that don't bear memorizing).

>The only reason to center the brain/environment interface is to try to separate what you consider to be the essential identity of a person from their physical grounding. I.e. to maintain a model which includes a soul.

Sure. If you want a complete description of a person, at minimum you need their whole body, including but not limited to their brain.

[tim333]

If you figure humans have general intelligence and are built out of regular matter it would seem theoretically plausible for us to build something else that has general intelligence out of regular matter. I think you may be overcomplicating things.

[erikpukinskis]

So you're saying that because general intelligences exist and are made of matter, we can conclude that it is possible to build a general intelligence out of metals and salts?

I don't see how that follows. Seems totally speculative to me.

If you mean possible in the "I can imagine it" sense then of course it's possible. So is a zombie apocalypse. Is it possible in the "there are models which predict it as a possibility"? Not to my knowledge.

The only such theoretical model I've seen is "processing power is increasing eventually, therefore computers will be able to do anything soon, including AGI" but that's a totally flawed model.

[harshreality]

I fail to see what bearing paragraphs 3-6 have on the possibility or probability of AGI.

AGI beliefs don't compare to religious beliefs just because you say so. You simply don't have the same knowledge background as people who think AGI is realistic or even likely. Maybe you have more or better knowledge, or maybe they do... but just saying they're wrong is pointless.

The boundary for systems-level consideration of intelligence is often placed around the brain because that's most convenient, not because nothing interesting happens at larger (sociology, ecology) or smaller (Drosophila, Portia) scales.

[erikpukinskis]

> You simply don't have the same knowledge background as people who think AGI is realistic or even likely.

That's ad hominem. But if you need to know my credentials to be able to take me seriously, I have a computer science degree and studied Cognitive Science at one of the top cogsci PhD programs in the U.S. (didn't finish). I have been a professional programmer for 20 years, in between school.

My point is that, like God, unless you can explain how it might work, you don't have a scientifically workable theoretical model.

[monadai]

At this point, it seems there is no manner of detailing that will convince people that it's plausible. I'm honestly unaware as to why people can't fathom it. It's a systems issue. If you want to resolve a systems issue, you need people who have worked on complex systems.

Having done the research myself, having the computational models in front of me, and currently developing iterative capability levels, I can assure you its plausible and I've developed some pretty complex infrastructures and systems during my time in industry.

I encountered the same issue in industry when I was developing the network infrastructure equipment that ensures your packets traverse the net. "You can't do that". "That isn't possible". "You can't shave off 300ms from that process. No one has touched that code in 10 years"

Why yes you can and I personally already have a track record of doing what is said can't be done. Everything is implausible until its made plausible. So, You'll never truly know until you try.

The biggest hurdle blocking people's way is that they're choosing to run down the same path. Why should you expect different results when everyone is approaching the problem the same way? That and, instead of the most knowing people getting in the trenches and attempting to write code, they remain in the philosophical camp and their works are tossed across the wall to the applied engineering camp. Rarely do you find someone who wears multiple hats or straddles the fence. I straddled the fence, saw what I saw, and now I'm developing it.

There are few who want to start from scratch and build up models. Many are ripping models from work done in the 60's,70's, and 80's without a second thought as to what was the thinking behind them. I chose an alternative path. It's paying off.

The model of consciousness that is being used is actually not detailed in any way (purposely). So, it is not very old. I have a stack of annotated white papers on the pioneers from the 60's/70's/80's centered on this inquiry and present day papers on : Global workspace theory (GWT), integrated information theory (IIT), etc. I fail to see any deep connection between my approach and their approach.

I intentionally haven't given any detail about the computational model of consciousness that is at the center of the architecture nor even the slightest detail on how to implement it. Given the climate in this space, I hope you can appreciate why.

You see a box. I see a relationship. There are no broad boxes over anything I am developing. The diagram was made in simple reduced form to help one conceive of the ties and flows to and from the world. People sing high praise of OpenAI and OpenAI gym. I experimented with similar open source packages when attempting to create a virtual environment for testing my code. I resolved on different packages and developed my own gym. I needed more access to the core/gut functions. From what I can tell, there are several other groups/companies/individuals that have done the same. No mention of them ever. No praise. Which is fine but it just goes to show you how there are likely numerous groups making headwinds in this area that no one has ever heard of.

>In a weird way, the information processing model is a vestige of the notion of a soul. If you really accept "physical fundamentalism" as OP describes it, then the interface between the brain and the rest of the world is nothing at all. Just some atoms of many. No more interesting than the interface between your stomach and your brainstem, or between the vibrations in front of your mouth and all nearby ears.

Interface/Relationship. There are no 'boxes' until you create one.

> The only reason to center the brain/environment interface is to try to separate what you consider to be the essential identity of a person from their physical grounding. I.e. to maintain a model which includes a soul.

Objective reality (governed by strict laws like physics).. Subjective experience. Pay close to attention to the wording I use as I don't give many details.

Seeing will eventually be believing. Once made manifest, you won't be able to deny its plausibility. Seems one can save a lot of time skipping attempts to try to convince people and just get to the development.

But yeah, consciousness isn't that serious. You just have to think outside the box to begin making progress on it. Whether or not were in a simulation is immaterial. The word 'simulation' really loses its meaning once you peer deep into the constructs that underly the universe. What does that even mean and how, even if you discovered it was a simulation, would you alter it in any meaningful way. Don't you think the person who created it, given how amazing it is, had the wherewithal to implement safe-guards/alerts? Or even made universal laws that forever restricts you from certain things? It's better that you focus on how it works than trying to define it. It makes for good story telling but I'd rather just dig in, understand it, and make use of that understanding instead. Again, do you want to sit around philosophizing and dreaming about it all day long or do you want to start converting that understanding into something ground breaking?

P.S - A component of the research that was conducted centered heavily on physics/quantum physics. It is quite important to understand the 'environment' and its laws when working on AGI.

[erikpukinskis]

I just want to say: I wholeheartedly support you and any and all research in this direction. I think AI is and will be hugely important.

You think I will be surprised how "general" the AIs are. I think you will be surprised how similar to human bodies you have to make them for them to approach a human definition of "general".

This is the only sense in which I think AGIs are impossible... not that they couldn't be fabricated, but that they would be functionally indistinguishable from a human with access to some good subroutines.

> Seeing will eventually be believing. Once made manifest, you won't be able to deny its plausibility.

I look forward to it. I say the same thing about my own work all the time, of which many people are incredulous. But still, you must admit that doomsday prophets say the same kind of thing. Again, please don't take that as disbelief, just not-yet-belief.

Regarding the specifics of the models, I would humbly submit this (as old as I am) paper on Ecological Psychology as a good, usable alternative to the information processing model of cognition: http://www.trincoll.edu/~wmace/publications/realism.pdf. If you don't want to wade through the high philosophy stuff, pages 194-209 get into more concrete specifics of the model they are proposing.

To me it seems fundamentally different than the I/O based picture my OP proposed. I would be interested to hear your perspective. And hit me up if you ever feel like coming out to Oakland to chat with a fellow crazy person with a passing knowledge of cognitive science and quantum mechanics. I would love to buy you a coffee.

[monadai]

Thank you very much. I support others and any research in this direction as well. Many aspects of my architecture are deeply biologically inspired. I find it to be of no surprise. I may grin from ear to ear from time to time when I see the function of a biological construct but I'm definitely not surprised :)

If you're capable and that's what you see, why not integrate it into a conceptual model and then a computational model and then write the code for it? When I say its possible I mean to say that, once you have the right model, you can write any code you want to make it a reality. The capability and hardware are already here today and more is on the horizon. One of the stacks I have lying about are spec sheets and white papers regarding Nvidia's GPU microarchitectures, HSA architectures, latency white papers, etc. I was literally waiting on their new microarchitecture to experiment with some new concepts. Future chips on the horizon make me :) deeply. So, unlike the past, the compute power is there and more is coming quite soon.

I scanned through 194-209. Excellent analysis and construction of a framework. I have hundreds of pages of this kind of analysis that I performed in the initial stages of my inquiry. This is excellent work. The thing is, you need to begin constructing models and theories from it. Start prototyping code and see how it fits together with other subsystems and aspects of your architecture and theory. You really just have to get in there and start stringing together code and start dissecting and trashing things that don't fit and locking in things that do fit.

My perspective is that the paper highlights a fundamental first step of how you approach AGI : By reasoning through your thought process, cognitive functions, and interaction in the world.

I would love to take you up on that offer and will when I return to the bay area. I always enjoy a good conversation on these matters. I'll take tea :)

For now, I'm in a bit of an undisclosed location doing dev. I Decided it would be best to get away from the high noise level in the valley while I focus on my R&D effort.

[psyc]

I read 194-209. I think their approach is fantastic, and it's rare to see such careful thought on this subject. However... nothing struck me as being incompatible with an information IO model?

[erikpukinskis]

Their core hypothesis is that you can't separate any kind of agent from their physical capabilities in the environment and still have cognition. That perception and action are not two phases but are an atomic unit of analysis. Perception is action. And what is perceived is not signals but relationships between an organism's capavilities and it's environment.

[psyc]

I just wanted to back you up, because I have a funny feeling you'll be criticized (common for frontierspeople) and I was thinking a lot about this very thing just yesterday. I've been a computer scientist for 25 years, and I've always tried to approach everything on the most basic terms of information, transformation, and IO. I revisit this fundamental way of conceptualizing daily, no matter what problem I'm working on.

Like you, I can't fathom what the objection is to AGI. A human body is complex, but human behavior is simply not that complex in the grand scheme. I have never encountered or thought of any theoretical difficulty with an algorithm mimicking human behavior, or the behavior of a vastly more intelligent sci-fi AI for that matter.

I don't care whether AGI has internal subjective experience. At least, I don't care beyond circular philosophical musing. I certainly don't care as regards the construction of AI, because it has no bearing on the construction of AI for the forseeable future.

AGI is going to sneak up on everybody incrementally, and intelligent people will still be arguing about whether it's possible, even while it happens.

[monadai]

>I've been a computer scientist for 25 years, and I've always tried to approach everything on the most basic terms of information, transformation, and IO. I revisit this fundamental way of conceptualizing daily, no matter what problem I'm working on.

Indeed. Breaking things down to their most raw and simplistic forms allow you to build them back up as you see fit or translate something from one domain to another. Through the deconstruction effort, you maintain the formula and methods to construct it back with confidence :)

> Like you, I can't fathom what the objection is to AGI. A human body is complex, but human behavior is simply not that complex in the grand scheme. I have never encountered or thought of any theoretical difficulty with an algorithm mimicking human behavior, or the behavior of a vastly more intelligent sci-fi AI for that matter.

Well, it depends how deep underneath it all you want to go. You can create a mimic machine or you can really get underneath the hidden layers and concepts and get at the root and create a truly artificial version of something. You have to be willing to dig deep and pursue answers wherever they are to be found even when they contradict your fundamental world view.

> I don't care whether AGI has internal subjective experience. At least, I don't care beyond circular philosophical musing. I certainly don't care as regards the construction of AI, because it has no bearing on the construction of AI for the foreseeable future.

Good, you have to be willing to put all your preconceived notions out to pasture. If you come to have a solid model that includes x,y,z or excludes it, it shouldn't matter as long as the model functions and is true to a stand-alone computational engine.

> AGI is going to sneak up on everybody incrementally, and intelligent people will still be arguing about whether it's possible, even while it happens.

Yeah, I kept asking myself why hasn't someone just gotten in the pits and started stringing together and forming a system. I got tired of asking questions and yelling out on an airhorn "this way everybody". So, I spent ~3 years researching the matter, created conceptual and computational models, and am now developing them.

>AGI is going to sneak up on everybody incrementally, and intelligent people will still be arguing about whether it's possible, even while it happens.

*nods head

[eli_gottlieb]

This is one of the longest posts I've ever seen that managed to say almost nothing aside from dropping a few popular buzzwords.

>There are few who want to start from scratch and build up models. Many are ripping models from work done in the 60's,70's, and 80's without a second thought as to what was the thinking behind them. I chose an alternative path. It's paying off.

What does your "alternative path" have to say about visual categorization/object recognition? Do you have ImageNet results?

>You see a box. I see a relationship. There are no broad boxes over anything I am developing. The diagram was made in simple reduced form to help one conceive of the ties and flows to and from the world. People sing high praise of OpenAI and OpenAI gym. I experimented with similar open source packages when attempting to create a virtual environment for testing my code. I resolved on different packages and developed my own gym. I needed more access to the core/gut functions. From what I can tell, there are several other groups/companies/individuals that have done the same. No mention of them ever. No praise. Which is fine but it just goes to show you how there are likely numerous groups making headwinds in this area that no one has ever heard of.

Can your "alternative path" play Atari games? How about Nintendo games? Can it complete Link to the Past for the Super Nintendo, as I could at age seven?