I have a hunch that we will discover (not invent) the strong AI not in Neural Networks, but in Swarm logic. Infinitely complex systems can emerge from simple units multiplied exponentially. I still can't stop being amazed by crazy thing they have discovered in Conway's Game of Life.
My guess, which reads more like sci-fi than anything is, the strong AI will emerge as a side-effect of IoT communicating with each other. Once the number of connected things hits a critical mass (I'm talking about billions), it will happen.
We're still a few iterations away from what I call the primordial network that could lead to that. We need some sort of automatic peer-to-peer communication between nodes, that is not in place yet, but I'm guessing will become norm in the future to solve a different problem.
I think "Edge computing" is taking us to that direction.
This is an interesting direction. But what is the critical attribute of a swarm that would give rise to strong AI?
The argentine ant supercolony is an enormous swarm, but we don't really expect to be able to have a conversation with it, or see it deal with abstract concepts, but they can solve lots of local approximate optimization problems.
Slime molds are single cells but you might think of them as a swarm of nuclei and organelles that can solve hard optimization problems when in the right form, but again they don't seem like a general intelligence.
Is the missing piece that swarms must be of agents with certain properties? Or that the interactions between the members of a swarm need to have some specific properties?
You raised some very valid questions. I'm speculating here, so take my answer as sort of sci-fi premise. I think as the internet and IoT grows, it becomes more decentralized, it will be come unfeasible for all these nodes to communicate to some central nodes (i.e. servers), so we will move toward automated peer-to-peer communication, with some simple routing and optimization rules.
These of these nodes automatically detect, connect, and communicate with the nearest node. Think of it as bluetooth vs wi-fi hybrid of the future. The aim is to make a more resilient internet by eliminating single points of failure (i.e. AWS servers down), but the end result will be some sort of rudimentary network, that will lead to "emergence" of some sort.
I just realized I've been describing "Edge computing".
I cannot imagine this happening in the world we live in.
I would not my IOT lightbulb (or worse, a camera) to communicate with my neighbor's house -- there are significant availability, privacy and security concerns.
The same way, I remember seeing lots of open WiFi networks, twenty years ago... and now they are virtually gone.
Instead, it would be still centralized servers, but with more reliability -- think fiber + 5G backup at home, and multi-region/cloudfare workers approach in the cloud.
There are precedents for this type peer-to-peer operation, where inherently limited protocols achieve a subset of functionality with acceptable privacy implications. For example, videogame distribution clients which download using a torrent-style process.
Sure, but while the data is peer-to-peer, the control is fully central. Those clients are not going to be doing anything without command from main server.
I really don't think we get to decide if we allow this or not. A company like Cloudflare or any other edge computing provider is easily in the position to make this happen, for completely innocent and valid reasons.
What I'd like to emphasis is what I'm suggesting will be the unwanted side-effect of this type of peer-to-peer communication.
It'd still be pretty centralized though -- it will be like bittorrent, not like gnutella. Organizations want statistics and usage data. People want to be able to connect from anywhere, even when far from home.
Saving bandwidth is useful, but I see no future for de-centralized control. The control connection is cheap and allows a lot of things that would be hard with decentralized control -- health monitoring, staged rollouts, instance security updates. You can see this trend even today -- there are all service meshes like Tailscale, BeyondCorp-like distributed systems, hey, even Google Wifi -- which have central control and distributed data.
So even if something strange emerges, once it starts taking up non-trivial resources, it will be logged by central server, detected (by humans) as a bug, and killed with a routine firmware update.
One more thing, I think the viability of communicating with an intelligence of a different kind is still an open question. It's speculated that intelligences that are too different in form or quantity might not be able to communicate with each other at all.
Likely the most probable scenario of our contact with extraterrestrial intelligence is they will completely ignore us, and will not attempt to communicate, or even if they attempt to, we will not perceive it as communication.
Communication needs to have purpose. Negotiating some kind of trade seems the way to go. We can provide ants with unimaginable "wealth" and they have unique skills that could be useful to us. Just not useful enough for us to make the effort?
What I wanted to suggest is even if this communication is possible. This question has been the main theme of two of Stanisław Lem's books, "His Master's Voice" and "Solaris".
Essentially, an intelligence can be so alien that the concept of communicating with it would be impossible.
You could get to that point if we inflate the value of our communication. This chat here might have (imaginary?) value to its participants but its absolute value is in application, if we are not going to do anything with it and the result is similar enough to not having the communication then we might as well [say] dance around the camp fire?
You could stretch the definition in the other direction too: if you crush an ant with your finger or put a spoon of honey next to their city you've communicated something. If you crush enough of them and they attack you we got ourselves an exchange in a language both parties understand.
I wouldn't attribute to much value to any definition that desires to define communication as an impossibility. Sure it could be done but such sour mind is best avoided.
The interesting property is embedded in the parent's use of the word "emerge" I think. It is the property of emergence that takes swarms of ants and forms an organized colony, or a network of nerves that creates intelligence.
I think it's something along the lines of.... randomized interactions of entities, constrained only by the attributes of the entities.
One thing that makes me think that is possible is Conway's Game of Life. The rules are simple, and on the face value, doesn't create anything exciting, but then we have this: https://www.youtube.com/watch?v=C2vgICfQawE
But this is true of many things.
The most interesting is computer itself.
The 6502 CPU only has 56 instructions. They are very simple, and on the face value, do not create anything exciting. But then, we have.. every program and game on Atari, Commodore 64, NES, Apple II.
Is there a quantitative difference between glider factory emerging from simple game of life rules, and incredible complexity of Elite game emerging from simple instructions of 6502?
Between the nodes and in their configuration [by lack of vocabulary] something like an n dimensional ink blob arrises. If this just happens to map reasonably to some n dimensional problem and the parameters of the swarm can be tweaked (evolve) to better map to it you can have logic with few moving parts. A simple example would be to first place the sugar pot in random locations until it ends up next to the coffee machine from where it migrates to the coffee table. The bags of sugar (to refill the pot) move around similarly.
I was on the same opinion as well. But I would like to borrow form Conway's Game of Life. The initial rules are simple, and to the untrained person, it just creates random blobs that disappear after a while. But then we have things like this: https://www.youtube.com/watch?v=C2vgICfQawE
I think emergence and complexity are important topics here. It's important that we actually understand how the pieces come together though. Maybe that's a place for functional programming/logic and frameworks that make composition easy
This seems too complex, too prone to failure, and too difficult to keep powered. I imagine the first industrially viable modular robots will be solid-state, with a mains power supply which is able to connect ambiently or arbitrarily. Power distribution circuits will be dynamic, and the abundant power will be used to control multiple electromagnets per module.
Would holes in the steel reduce the magnetic properties very much?
Holes could enable sensors, charging, and optical communication. For inter-node communication, radio signals can conduct between the shells and would be far more robust than optical. But optical would be more resilient against interference, jamming, and eavesdropping.
I had a dream once that I went to a panel in my home and reconfigured the whole place to be a centred around an open swimming pool, the nanobots my flat was made of just reconfigured themselves around me.
Water is very heavy, so you'd need to reconfigure the whole block to make that happen. I wonder how that would play out - it might make a good premise for a science fiction short story.
48 comments
[ 8.3 ms ] story [ 109 ms ] threadIf they could make it so you could control them as a whole, once linked, then they would be getting somewhere.
Maybe have some other RC pieces that are bars and the spheres could be the joints. Have it all linked with some IK action.
I think "Edge computing" is taking us to that direction.
The argentine ant supercolony is an enormous swarm, but we don't really expect to be able to have a conversation with it, or see it deal with abstract concepts, but they can solve lots of local approximate optimization problems.
Slime molds are single cells but you might think of them as a swarm of nuclei and organelles that can solve hard optimization problems when in the right form, but again they don't seem like a general intelligence.
Is the missing piece that swarms must be of agents with certain properties? Or that the interactions between the members of a swarm need to have some specific properties?
These of these nodes automatically detect, connect, and communicate with the nearest node. Think of it as bluetooth vs wi-fi hybrid of the future. The aim is to make a more resilient internet by eliminating single points of failure (i.e. AWS servers down), but the end result will be some sort of rudimentary network, that will lead to "emergence" of some sort.
I just realized I've been describing "Edge computing".
I would not my IOT lightbulb (or worse, a camera) to communicate with my neighbor's house -- there are significant availability, privacy and security concerns.
The same way, I remember seeing lots of open WiFi networks, twenty years ago... and now they are virtually gone.
Instead, it would be still centralized servers, but with more reliability -- think fiber + 5G backup at home, and multi-region/cloudfare workers approach in the cloud.
What I'd like to emphasis is what I'm suggesting will be the unwanted side-effect of this type of peer-to-peer communication.
Saving bandwidth is useful, but I see no future for de-centralized control. The control connection is cheap and allows a lot of things that would be hard with decentralized control -- health monitoring, staged rollouts, instance security updates. You can see this trend even today -- there are all service meshes like Tailscale, BeyondCorp-like distributed systems, hey, even Google Wifi -- which have central control and distributed data.
So even if something strange emerges, once it starts taking up non-trivial resources, it will be logged by central server, detected (by humans) as a bug, and killed with a routine firmware update.
Likely the most probable scenario of our contact with extraterrestrial intelligence is they will completely ignore us, and will not attempt to communicate, or even if they attempt to, we will not perceive it as communication.
Essentially, an intelligence can be so alien that the concept of communicating with it would be impossible.
You could stretch the definition in the other direction too: if you crush an ant with your finger or put a spoon of honey next to their city you've communicated something. If you crush enough of them and they attack you we got ourselves an exchange in a language both parties understand.
I wouldn't attribute to much value to any definition that desires to define communication as an impossibility. Sure it could be done but such sour mind is best avoided.
I think it's something along the lines of.... randomized interactions of entities, constrained only by the attributes of the entities.
The 6502 CPU only has 56 instructions. They are very simple, and on the face value, do not create anything exciting. But then, we have.. every program and game on Atari, Commodore 64, NES, Apple II.
Is there a quantitative difference between glider factory emerging from simple game of life rules, and incredible complexity of Elite game emerging from simple instructions of 6502?
(Cf. Alexander Bell's kites & Bucky Fuller's geodesic mega-structures.)
Makes me wonder if Arnie could have defeated him/it by throwing lots of strong magnets into the mix.
Holes could enable sensors, charging, and optical communication. For inter-node communication, radio signals can conduct between the shells and would be far more robust than optical. But optical would be more resilient against interference, jamming, and eavesdropping.
[1] https://en.wikipedia.org/wiki/Orb_Swarm
Alternative experiment to do in computer.
Make each sphere with a neuron that learns about space, time (given n cordinates of movement, how many cpu cycles it took to move).
Next make each sphere made of optical material that can see outside from within.
Wait until the sphere learns a visual model of its surroundings.
Now the sphere can predict movement of other spheres in motion. It can focus on a single particular sphere.
Let it find a way to communicate and then human snoop in.
The sphere learns about itself from other sphere. (A group of spheres atached to them self and mimicing a sea wave)
Overall a good candidate to be in here https://github.com/imvetri/artificial-intelligence/blob/mast...