Since we had this discussion about UAVs and F22s etc...

[alexjrgreen]

Yes, they have well adapted systems which allow complex interactions and a range of responses to stimuli. Most people reading descriptions of what single-celled organisms do in these papers, would probably not call that "intelligent", so I suggest that you define the word "intelligence" when you use it in such a controversial way.

It really isn't a different kind of intelligence, just a smaller amount of it.

The kind of intelligence that scales up with increasing size, like an ant colony.

[MirariNefas]

It really isn't a different kind of intelligence, just a smaller amount of it.

We really don't know that. In any case, whatever argument you want to make about output, mechanism is different.

[DeltaV]

Air combat will change, no doubt. But, there are bigger agents of change at work than AI, UAVs, UCAVs, etc. Those agents are (near-) speed-of-light weaponry (lasers, masers, xrasers, particle beams, focused EMP, ...) and very-high-speed sublight weaponry (plasma guns, rail guns, pulse-detonation cannons, acoustic shock beams, ...).

"Dogfights" of the future will be more like drive-by shootings. The advantage will go to the weapons platform with the best sensors, the best fire control, the most weapon apertures and the most available power for the weapons employed. The small size of UAVs, touted as an advantage above, will become a disadvantage as maneuverability becomes immaterial, and the abilities to (a) resolve targets (sensor aperture, interferometric baseline), (b) bring a greater number of weapon apertures to bear (exterior area) and (c) provide a greater amount of power to each weapon (internal volume) become paramount. Think fast mosquitoes vs. a much faster oxyacetalene torch.

This will all happen too fast for a human pilot to be part of the innermost fire control loop. The pilots role, besides flying the airplane and making mission profile decisions, will be at a higher level, such as assigning priorities to target types ("Computer, take out all J-75 UCAVs with the neutron beams before masering the Hyperwasps. EMP the Su-91s at zone 3 penetration, jam the UV-43 comm links in zone 4 and continue Hyperwasp suppression until the B-7s exit the target zone..."). Who knows... if he comes up against an evenly-matched, human-piloted adversary, it might even end up as an old-fashioned dogfight...

[alexjrgreen]

It really isn't a different kind of intelligence, just a smaller amount of it.

We really don't know that. In any case, whatever argument you want to make about output, mechanism is different.

The only intelligence we have any experience of is built from single cells.

Our work on artificial neural networks has shown the network barely contributing to the intelligence of the system as a whole, which isn't our experience with living systems. So individual neurons must already be slightly intelligent, just like single celled organisms.

[Skipjack]

and the abilities to (a) resolve targets (sensor aperture, interferometric baseline), (b) bring a greater number of weapon apertures to bear (exterior area) and (c) provide a greater amount of power to each weapon (internal volume) become paramount.

I dont see why (a) cant be done by UAVs just as well as by manned vehicles.
I can see why (b) and maybe also (c) might be problematic.
So far these new weapons systems are not in place yet, however. They are still in development. I have seen the laser weapons systems that are currently being tested. They are not going to fit into a small fighter jet either. Sure they might get smaller, but until then a lot of time will pass.
I dont think that the F22 can be easily adapted to fit laser based systems.
Other systems, maybe. Still none of them are operational. So this discussion is about the future, not the current.

As I said the advantage of the UAV is, that it will see the F22 before F22 sees it.

Is that how it works?

Yes, from what I read current UAVs have between two to 3 people in charge of them. Each person is taking care of a different set of subsystems. They can also rotate them in shifts which is nice. No need to pee into your flight jacket

doubt the hardware on a UAV is as low-end and unadapted as home/gaming PCs.

I dont. You would be surprised how much use the 486 found in the military. That is why export restrictions on that CPU were up for quite a while. I have a friend that was doing work for computer systems for an undisclosed Navy and their new subs. They were still using 486 based computers in their new subs then (and that was just 5 years ago). At that time I was running a Pentium D at my office.
Why are they using these CPUs? Because they are robust. They dont produce a lot of heat, e.g.. Also, since they are manufactured at a larger micron scale, they are less susceptible to EMPs and other issues mentioned earlier.
Now I am not saying that modern UAVs still use 486 CPUs, because honestly I dont know, but I am pretty sure that your modern day gaming workstation is no less capable than what the military uses in their high end systems.

I would also not underestimate the AI in games. The video game industry is a very big one, with comparably large margins and comparably low risk. It has been driving a lot of computer developments in the past decades. Game development programers are in no way different from military programers. If anything, they are better, because they most likely earn more.

[Tom Ligon]

While shaving this morning, I noticed my calendar was still turned to November, a B-24 (an ugly thing that never made many movies but did make a lot of history).

I flipped to December, and there was a pair of P-51D's flying formation. The one in the back is the current incarnation of Big Beautiful Doll!

I should not be surprised. There are a handful of particularly photogenic Mustang restorations that pose for cameras a lot, and Doll is one of them.

As for greatest fighter ever, periodically they run polls to select them, and Mustangs generally win. Often one reason cited was the fact they could escort bombers all the way to Berlin (Hermann Göring is supposed to have realized the war was lost when he saw Mustangs accompanying bombers over Berlin). But I have to agree with those who laud the other great fighters of that war. In the Microsoft combat simulator, the P-47 is much better than the Mustang, and it is always the clear choice for ground attack. The Bearcat and F4U were outstanding as well. The Spit was beloved by its pilots, but the Hurricane was the real workhorse of the Battle of Britain, and does not get enough credit.

All of which discounts the worth of the people who flew them (men and women), facing danger on a level few people today can really appreciate. The more disadvantaged they were by the machines they flew, the more respect the pilots deserve. They loved flying, but I suspect would have gladly left the dying part to a computer.

[MirariNefas]

It really isn't a different kind of intelligence, just a smaller amount of it.

We really don't know that. In any case, whatever argument you want to make about output, mechanism is different.

The only intelligence we have any experience of is built from single cells.

Our work on artificial neural networks has shown the network barely contributing to the intelligence of the system as a whole, which isn't our experience with living systems. So individual neurons must already be slightly intelligent, just like single celled organisms.

Must? That doesn't follow, it's just a possibility. Do you jump to conclusions like that all the time? Do you have any idea how open ended biology is? How much experience with the sciences do you have, exactly? Shaky logic, and the world is not as simple as you would like it to be.

And again, nebulous use of language there. But I also notice you haven't contradicted my previous statement, so I don't know what point you were trying to make anyway. I suspect it would be another bit of taking mental leaps of faith in implication.

[alexjrgreen]

Must? That doesn't follow, it's just a possibility.

We've spent sixty years trying to make the network more intelligent, with very little success. So yes, there must be some intelligence in the base units.

[Luzr]

Must? That doesn't follow, it's just a possibility.

We've spent sixty years trying to make the network more intelligent, with very little success.

Once again, I doubt where you are getting your info. Well, in fact I do not. I guess all "greens" are as misguided as your are...

Given our limited expertise, I would say neural networks work pretty well. As for inteligence, do you realize how many neurons and how many synapses are in human brain?

Available hardware is not yet able to simulate even one promile of that. And even if we get there, hardware wise, we will have a very long road to actually figuring out the "base outline" of the network.

BTW, if anything is wrong with the idea of neural network as means of AI, it is in fact that they in reality just simulate some more generic concept, most likely bayesian reasoning, which means ta) they are not as effective as applying such principle directly b) contrary to what you say, individual neurons do not matter.

[alexjrgreen]

As for inteligence, do you realize how many neurons and how many synapses are in human brain?

Williams and Herrup in 1988 put the numbers at 10^11 neurons and 10^14 synapses.

The Cray supercomputer installed this year at Oak Ridge National Laboratory arguably approaches this.

IBM is currently funded by DARPA to simulate a human brain.

[bcglorf]

Must? That doesn't follow, it's just a possibility.

We've spent sixty years trying to make the network more intelligent, with very little success. So yes, there must be some intelligence in the base units.

If we're still talking about neural nets then this makes no sense. The smallest base units are integers, and I don't think anyone is gonna attribute intelligence to them. If you simply mean smaller neural networks connected to a larger one then admitting intelligence in a small network is no different to admitting it exists in larger ones.

Neural networks are just algorithms, they have about as much inherent intelligence as sub-functions inside an A* routine. NN's are not special, philosophical semantic musings aside.

[bcglorf]

As for inteligence, do you realize how many neurons and how many synapses are in human brain?

Williams and Herrup in 1988 put the numbers at 10^11 neurons and 10^14 synapses.

The Cray supercomputer installed this year at Oak Ridge National Laboratory arguably approaches this.

IBM is currently funded by DARPA to simulate a human brain.

If it can simulate 10^14 neurons operating in parallel it could meet the most basic level of processing power we understand the brain to have. That is the ludicrously easy part, and I do not believe any hardware is likely to meet even that bar any time soon. Once we do reach that bar, there's the trick of how those 10^14 neurons dynamically reconnect with one another to tackle.

[Luzr]

As for inteligence, do you realize how many neurons and how many synapses are in human brain?

Williams and Herrup in 1988 put the numbers at 10^11 neurons and 10^14 synapses.

So, how many bytes do you need to store info of single synapse/neuron?

In fact, we still do not know. But I am pretty sure that such computing power definitely was not available in "golden years" of AI research. So it is long shot to say "we have not achived anything in past 60 years".

Anyway, Cray XT5 has about 10^10 _bytes_ per node. With about 10^5 nodes, yes, you are at about 10^15 bytes. But I doubt you can get fit complete synapse info into one single byte.

Also, nobody tried to use that Cray to neural network simulation yet....

IBM is currently funded by DARPA to simulate a human brain.

Yes, and given Moore's law they will probably succeed some day. So, what that has to do with your original claim? (Slightly off-topic: I believe we will get general AI using some other method sooner).

Now, obviously, the model of neuron can be pretty complex. But that makes it complex, not inteligent. Plus, we still do not know what degree of complexity is really required for functioning brain and what is just result of biological origins. From what I read, quite simple models can have very complex behaviour if networked.

[alexjrgreen]

There's some evidence that simple organisms use memristance to create complex behaviour.

Perhaps that's the way we need to go...

[MirariNefas]

Must? That doesn't follow, it's just a possibility.

We've spent sixty years trying to make the network more intelligent, with very little success. So yes, there must be some intelligence in the base units.

We've spent a long time pursuing something and it hasn't worked too well.

Hm. Pointless statement. No, sorry, your logic makes unsupported leaps. Also, I'd like you to define intelligence, just to makes sure we're all on the same page.