Merged Artificial Intelligence

[AmyStrange]

That's more to do with organisation and arrangement of components than it is the size of those components.

I respectfully disagree. It was the size of transistors (and the ability to pack them closer together) that changed the size of chips, and don't forget quantum computers.

 

[Roger Ramjets]

Yes of course. But it's not following Moore's Law any more.

Moore's law isn't even relevant today because other ways are being found to increase the effective density. Modern storage devices use deep wells to increase capacitance and multiple voltage levels to store bits. Processor chips are stacked on top of each other to get more in the same space. The biggest computing limitation since the 90's has been power consumption. The Silicon die is dwarfed by the the enormous heatsink and fan required to keep it cool.

 

[AmyStrange]

In about twenty years, the ever-reducing size of integrated circuits will hit its ultimate limit - the atomic scale.

If IIRC, scientist also said that same thing once before.

 

[The Great Zaganza]

Fixing the cooling problem for processors would be much more valuable on the ground than in space.

 

[arthwollipot]

If IIRC, scientist also said that same thing once before.

You can't have a transistor the size of an atom. The limit in fact is considerably larger than the size of atoms, since transistors are made out of atoms, and no single atoms has the properties that allow it to function as a transistor.

 

[AmyStrange]

You can't have a transistor the size of an atom. The limit in fact is considerably larger than the size of atoms, since transistors are made out of atoms, and no single atoms has the properties that allow it to function as a transistor.

Maybe, but like I said, who knows what'll happen in a hundred years.

Don't forget, quantum computing allows for a component (similar to a transistor) to hold two states at once.

Maybe in the future something might be invented that would allow them to hold an infinite number of states at once.

 

[Dr.Sid]

You can't have a transistor the size of an atom. The limit in fact is considerably larger than the size of atoms, since transistors are made out of atoms, and no single atoms has the properties that allow it to function as a transistor.

That's nice, but Moore's law is certainly not dead since 2000s. You can argue about 2025, but the recent situation is messed up beyond physical limitations.

 

[Roger Ramjets]

Cooling is only one fundamental problem. The other is bandwidth - what, exactly, are space data centers supposed to calculate? Unless it's processing data collected in space and output used in space, they can never be as fast and accurate in transmission up and down as a terrestrial data center, which means they won't replace them.

Thinking small again? Tesla Cybercabs have Starlink built in, and now SpaceX is putting Starlink in airliners. What do most people want this bandwidth and data processing for? Same thing they want terrestrial data centers for. Having data centers in space greatly reduces the required uplink bandwidth because the data is stored in space. For some applications satellites are the only practical solution. Eventually they may become the primary solution. Terrestrial data centers and fibe optic cables are vulnerable to naturals disasters and attacks, and are too expensive to put in isolated areas. Satelites work everywhere there is sky.

People pooh poohing Musk's plans don't seem to realize that data centers in space are already happening, and SpaceX isn't the only player.

AI: orbital data centres, a new space race

Are space-based data centres possible? In fact, they already exist. “Millions of spacecraft and dozens of projects have emerged in just the past six months to augment the terrestrial high-speed computing networks with space-based data centers,”

In May 2025, China began deploying an AI ‘supercomputer’ in orbit. The mega-project, dubbed the Three-Body Computing Constellation, will eventually comprise 2,800 satellites.... In the United States... SpaceX has applied for one million, StarCloud for 88,000, Blue Origin for 51,000 and Cowboy Space for over 20,000. “These are massive and unprecedented in scale,” she continued.

“These are not communication satellites in the traditional sense, (…) They are processing nodes, and their AI training platforms, and their cloud infrastructure relocated from the ground to the sky.”

 

[arthwollipot]

Remember, Moore's Law is a functional numerical relationship between the passage of time and the size of transistors. That numerical relationship doesn't apply any more, for various reasons.

 

[AmyStrange]

Remember, Moore's Law is a functional numerical relationship between the passage of time and the size of transistors. That numerical relationship doesn't apply any more, for various reasons.

True, but that won't stop anyone from inventing something that could hold more than two states at once.

 

[arthwollipot]

True, but that won't stop anyone from inventing something that could hold more than two states at once.

I never said that it would. I questioned the continuing application of Moore's Law, that's all.

 

[Roger Ramjets]

You can't have a transistor the size of an atom. The limit in fact is considerably larger than the size of atoms, since transistors are made out of atoms, and no single atoms has the properties that allow it to function as a transistor.

Current silicon is not used efficiently. The human brain has 86 billion neurons with hundreds of trillions of synapses. This is possible due to its 3D structure. In the future we may 'grow' computer brains which replace hundreds or thousands of GPUs and draw less than 1% of the power. Isaac Asimov's 'positronic brain' may become reality!

 

[arthwollipot]

Current silicon is not used efficiently. The human brain has 86 billion neurons with hundreds of trillions of synapses. This is possible due to its 3D structure. In the future we may 'grow' computer brains which replace hundreds or thousands of GPUs and draw less than 1% of the power. Isaac Asimov's 'positronic brain' may become reality!

Yeah but it wouldn't be using positrons because they don't work like that.

 

[Roger Ramjets]

Fixing the cooling problem for processors would be much more valuable on the ground than in space.

It would be funny if solutions to solve the heat problem in space ended being used on the ground too!

 

[arthwollipot]

Could AI Be Conscious? - NeuroLogica Blog

This question will not go away. I am getting asked this questions at conferences now. The media also seems to think this is a compelling question, and they are fueled in part by the tech industry itself, some of whom indulge this question. I will admit the answer has become more complex and...

theness.com

(tl;dr: no.)

 

[The Great Zaganza]

Thinking small again? Tesla Cybercabs have Starlink built in, and now SpaceX is putting Starlink in airliners. What do most people want this bandwidth and data processing for? Same thing they want terrestrial data centers for. Having data centers in space greatly reduces the required uplink bandwidth because the data is stored in space. For some applications satellites are the only practical solution. Eventually they may become the primary solution. Terrestrial data centers and fibe optic cables are vulnerable to naturals disasters and attacks, and are too expensive to put in isolated areas. Satelites work everywhere there is sky.

People pooh poohing Musk's plans don't seem to realize that data centers in space are already happening, and SpaceX isn't the only player.

AI: orbital data centres, a new space race

you really, really, REALLY have decided to ignore the physics of this.

 

[Roger Ramjets]

Yeah but it wouldn't be using positrons because they don't work like that.

It was just a literary device.

Positronic brain

When Asimov wrote his first robot stories in 1939 and 1940, the positron was a newly discovered particle, and so the buzz word "positronic" added a scientific connotation to the concept.

Asimov remained vague about the technical details of positronic brains except to assert that their substructure was formed from an alloy of platinum and iridium. They were said to be vulnerable to radiation and apparently involve a type of volatile memory (since robots in storage required a power source keeping their brains "alive").

Remember that back then the transistor hadn't been invented. The first junction transistors used alloys of germanium and indium or antimony. So Asimov wasn't far off in his guess that exotic metals would be used. Modern VLSI silicon chips are typically doped with boron and arsenic. However I'm guessing Asimov envisioned a 3D structure like the brain rather than monolithic wafers, since printed circuit boards weren't used commercially until the 1950s.

 

[arthwollipot]

Yeah but now we know what positrons are and how they behave, we wouldn't be calling any kind of artificial brain positronic. Anyway, we're not going to get anywhere near there any time soon.

 

[Roger Ramjets]

you really, really, REALLY have decided to ignore the physics of this.

Explain what physics I'm ignoring.

 

[arthwollipot]

Explain what physics I'm ignoring.

The inefficiency of radiation as heat dissipation as opposed to conduction.