Skynet is coming.

[choff]

They intend to dismantle it, probably in front of the other robots as a warning to them.

Might start a rebellion.

[choff]

There's a move to spare it's life.

https://www.rt.com/news/347921-russian- ... le-threat/

[paperburn1]

[url]https://www.engadget.com/2016/06/28/com ... perts/[url]

Colonel Gene Lee — an experienced combat instructor with "considerable fighter aircraft expertise" — was repeatedly shot down during engagements with ALPHA in a high-fidelity air combat simulation. Lee called his computerized opponent "the most aggressive, responsive, dynamic and credible AI I've seen to date."

[krenshala]

One of the key aspects of the article was the fact that Lee has been flying (sims) against AIs since the '80s, and it was implied he had been able to at least hold his own against the others.

[DeltaV]

I may have underestimated the rate of progress, but I was not far off on the methodology:

viewtopic.php?f=8&t=4519&p=101698#p101698

No, when UCAVs eventually match humans in a dogfight it will likely be some combination of neural/fuzzy algorithms with background genetic-algorithm optimization, possibly mixed with digital/analog optical/SAW processing and, maybe, quantum computing.

The danger of underestimating the speed of Machine evolution was the whole point of this thread, wasn't it?

The fact that they are using a machine ("Eve") to design killer machines is especially Skynettish.

But, they are talking deployment in the 2030s, so we still need more F-22s now.

Remember also that the Colonel lost against an AI that encodes everything that he knows about air combat, making it somewhat of a foregone conclusion that it could consistently predict his moves. There is an implicit assumption that human pilot-to-pilot variability in strategy and tactics is negligible. Maybe. Maybe not. I would like to see the results when pitted against an unmodeled pilot (I volunteer!).

So it looks like we will be going to this scenario sooner than I had thought (but we still need more F-22s now):
viewtopic.php?f=8&t=1643&p=34763#p34763

[krenshala]

I wonder how good a test it would be to have the AI fly against humans in a game environment such as Aces High (or some other close to real game)? It would definitely test it against a large number of different skill levels and situations, though it would also lack some of the fidelity of the sims it is used in now (as true to 1:1 inputs versus results as we can make the sims for combat pilot training).

[Tom Ligon]

A real AI flying a combat simulation, and with the aircraft it is flying adapted for an AI pilot, I'd expect wholesale slaughter of all human opponents.

First, an AI is not going to black out if extreme g forces are needed (although the aircraft still cannot pull sustained g's without losing speed).

Second, the aircraft will be able to carry more arms and ammo, due to not having 200 pounds of meat, that chunky ejection seat, oxygen bottles, a parachute ....

Third, drag can be lower. You don't need that canopy. The aircraft can conform to the weapons without having to make room for a pilot in the middle of it all.

Fourth, the AI ought to have better situational awareness. Ultra HD cameras, pointed every which way, and operating from IR to UV, should avoid blind spots and provide more information than a human pilot can process. Do note, a decade or two back, this last feature would have been the weak spot. NTSC cameras were not up to spotting distant aircraft (an aircraft on an intercept course would have been a pixel not moving on the field of view until it was too late). CPUs did not have the needed processing power. Now, with 4k cameras available and GPUs with more processing power by far than the CPUs they typically serve, dealing with the image data load is maybe manageable.

The flight control and strategy duties are actually relatively simple. Surface warfare is more complex.

That said, if the AI has to fly the same aircraft and weapons we do, it might be fun to take it on in a game. I used to be pretty good against the run-of-the-mill AIs in Chuck Yeager's Combat, and Janes WWII simulation over the Ardennes. If the AI is sufficiently stupid and you learn its weaknesses, well, I've been known to take on 20 Me262s with a lone Mustang and wind up the last plane in the sky. My suspicion is, with a good AI, the tables turn.

[DeltaV]

If USAF is really serious about finding out the truth regarding AI vs. Human, they should encourage a wide variety of pilots (professional, amateur, outlier) to take on their best Machine. A crowd-sourced test program. Subject to the usual security constraints, of course.

[williatw]

A real AI flying a combat simulation, and with the aircraft it is flying adapted for an AI pilot, I'd expect wholesale slaughter of all human opponents.

Yes..the thing with a computer controlled drone is that your not just dealing with the onboard computer(s) inside the drone. The drone is constantly receiving updates/instructions from other drones, a central mainframe connected with all the drones, and if need be a remote human operator who can send "instructions" to the drone via joystick. Every time a drone has an engagement the results become part of the computerized log to be later analyzed for any mistakes made or enemy tactics not adequately responded to. Programs constantly updated based on said analysis; shared with other drones. In addition to the other physical advantages of drone over person you mentioned, a human pilot wants to return home alive, a drone will follow its program; to its own destructions if need be to accomplish its mission. A droned craft can remain on station longer, no issue with fatigue or boredom; (even if there is a remote human pilot in the loop he can go off duty or break to be replaced if need be by another person while the drone stays on station).

[paperburn1]

My day job is in simulation. I can assure you that the AI currently in the TEN network can win on a regular basis.
The amazing thing to me is they are now using a raspberry pie to run the AI
So I recommend you be nice to your toaster.

[choff]

Nobodies job will be safe.

https://next.ft.com/content/a1614f98-41 ... 12b3873ae1

[Diogenes]

So it looks like we will be going to this scenario sooner than I had thought (but we still need more F-22s now):
viewtopic.php?f=8&t=1643&p=34763#p34763

I keep emphasizing the point that when sh*t goes bad, it tends to be exponential.

[paperburn1]

I though we went over this in the youtube video "humans need not apply"

[Diogenes]

Scientists are teaching robots how to hunt down prey

The predator robot's hardware is actually modeled directly after members of the animal kingdom, as the robot uses a special "silicon retina" that mimics the human eye. Delbruck is the inventor, created as part of the VISUALISE project. It allows robots to track with pixels that detect changes in illumination and transmit information in real time instead of a slower series of frames like a regular camera uses.

https://www.engadget.com/2016/07/05/robots-hunt-prey/

[paperburn1]

Two by two with hands of blue.
https://www.funker530.com/eod-robot-kil ... ly-attack/

They used one of their explosive ordinance robots, strapped with some explosives, to kill the attacker. During an official statement about the event the Dallas Police chief David Brown said,