An AI defeated human fighter pilots in an air combat simulator (2016)
binaryloom.com
binaryloom.com
That could make a big difference. It's easy to make a self-driving car for instance if you can feed it the position of every other car in the world, with their movement data, and all other obstacles.
The article also refers to ALPHA being able to defeat the human pilot with impeded sensors, which implies that it was being fed information from simulated sensor systems.
Coz a pointless and massively destructive war is OK as long as we end up winning in the end?
Paper link for those interested : https://www.omicsgroup.org/journals/genetic-fuzzy-based-arti...
(from a June 2016 press article : http://magazine.uc.edu/editors_picks/recent_features/alpha.h... )
It's still worthwhile to develop high-performance piloted airframes, since people are still actors in any conflict, and will need to operate within the theater of combat.
Even if primary operations are handled by unmanned systems, redundancy and fallback systems are needed. Not just troop transport from A to B. Well-armed vehicles capable of deploying weapons will still need to be piloted and maintained.
Police still use horses. The military will still find utility in piloted warbirds.
You're comparing airframes that are shipping today to a hypothetical drone that doesn't even exist. It's not even a remotely fair comparison.
https://en.wikipedia.org/wiki/Sunk_cost#Loss_aversion_and_th...
The sunk cost fallacy is in game theory sometimes known as the "Concorde Fallacy",[8] referring to the fact that the British and French governments continued to fund the joint development of Concorde even after it became apparent that there was no longer an economic case for the aircraft. The project was regarded privately by the British government as a "commercial disaster" which should never have been started and was almost cancelled, but political and legal issues had ultimately made it impossible for either government to pull out.
Stopping the program now would result in a trillion dollars having mostly gone to waste. Most of that money went to the initial development of the program, and a comparatively small amount is now spent on each new airframe. Scrapping it right as the upfront costs begin to pay off as airframes are delivered makes no fiscal sense whatsoever.
Operational F-35Bs have been delivered to the military since 2015. F-35As since 2016. Drones capable of fulfilling the roles the F-35 will do not yet exist. You'd be sinking a program where the overwhelming majority of upfront R&D has been paid off in exchange for starting a new program from scratch that may not deliver a drone for another ten, maybe fifteen years.
Yes, today we could design and build a new set of aircraft with more modern capabilities. But with your logic, after fifteen years of R&D — just as units begin rolling off the production line — we'd be looking to scrap the program to start anew with whatever advancements those fifteen years in technological advancement might have unlocked.
You're comparing a plane that's been in development for twenty years with a hypothetical drone with today's tech that wouldn't be ready for another fifteen to twenty years.
besides, remember when Iran (maybe the Chinese) just took our new drone by hacking it?
https://en.wikipedia.org/wiki/Iran%E2%80%93U.S._RQ-170_incid...
Amazing that the AI was running on a Raspberry PI... I'm amusing image processing happens on a "low resolution". Also, this seems to be interesting:
"They tackled the problem using language-based control (vs. numeric based) and using what’s called a “Genetic Fuzzy Tree” (GFT) system, a subtype of what’s known as fuzzy logic algorithms.
States UC’s Cohen, “Genetic fuzzy systems have been shown to have high performance, and a problem with four or five inputs can be solved handily. However, boost that to a hundred inputs, and no computing system on planet Earth could currently solve the processing challenge involved – unless that challenge and all those inputs are broken down into a cascade of sub decisions.”
That’s where the Genetic Fuzzy Tree system and Cohen and Ernest’s years’ worth of work come in."
"The training happend on a 500$ commercially available computer"
"It uses linguistic input instead of numeric input"
Can i conclude that ( after reading the article), that the training matches "situations"/"variables" to words. Eg. The enemy is far, close. The friend is behind me/in front of me/above me and then i go up/down/do a roll.
Multiple generations later, some linguistic variables are left away, so only the necessary ones remain.
From the comments on Reddit ( https://www.reddit.com/r/CredibleDefense/comments/4q8h8z/pap... )
"This is not another opinion piece vaguely arguing how AI will "one day" replace pilots. The authors seem very cognizant of all the challenges and complexities. They clearly limit and define the scope of their work."
"This "research paper" is a commercial. The expert fighter pilot is an employee of the company and his job is to promote it."
None the less, gonna read on some Fussy Logic :p - https://www.tutorialspoint.com/artificial_intelligence/artif...
Edit: Link to previous HN discussion -- https://news.ycombinator.com/item?id=11993366
> What sets ALPHA apart is the genetic fuzzy tree decision system that can calculate an opponent’s movements or strategies 250 times faster than you can blink.
That's just not fair.
One more comment, can this thing (ALPHA) land an Airbus A320 on the Hudson?
Assuming an AI like this was given 100% control over a commercial plane, it would have been programmed with such a checklist. It would have also been tested on this ability in a simulator (just like regular pilots). The AI would have easily worked out in an emergency that it couldn't reach the runway, or any alternative runway and not even bothered trying.
The only question is, would the AI have decided that the Hudson was a valid surface for a water landing attempt? And that really depends on the database programmed into the plane. It can't really see the river, so the database would need to have the river marked as water, at the correct elevation. It also can't really see the bridges (maybe if it had LIDAR), so the bridges would have to be included in the database too, so it could pick a good stretch to try and land it. Oh and boats too... either it would have to dynamically spot them, or get lucky.
But if the database correctly reflected reality, then it would see the opportunity, try to execute it and potentially succeed.
I'm not saying it wont eventually happen, but no: they unequivocally cannot fly our aircraft in takeoff/landings, emergency situations, etc.
An AI pilot can learn from every mistake made by every AI pilot, can absorb everything important from their collective experience.
It's not a fair fight. If the designer set about testing the AI's ability to land on water, eventually it'd find a way of doing it, and then that could become part of the corpus of experience.
AI pilots stand to theoretically be better than any pilot ever. If there's ever a better human pilot, their experience can be documented, examined, and imported into the core AI.
Humans will most likely be used in "decision-making" roles.
AI really should have no place in military but it's going to happen. Some spineless fucks are going to spread their legs for a 5 star general to sell their company.
The defence industry is in the business of killing humans and you play a direct role by selling your knowledge and skills to make it more efficient. Fuck that shit.
China and Russia will certainly build AI drones as soon as they possibly can. The only thing that happens if western countries don't build them is that they are at a significant disadvantage.
I disagree actually. Nuclear weapons render chemical and biological weapons moot, and no major superpower is willing to give up nukes. The United States has no need for chemical and biological weapons because it's keeping it's nukes. I think that AI will present a comparable advantage.
Also, AI does not (yet) create the sense of dread in the popular imagination that nukes did during the cold war. Incinerating an entire city in a few seconds is terrifying on an existential level to any human. Having computers control drones and tanks seems.. almost boringly inevitable, by comparison. I doubt there will be unified political will to prevent it.
Also, I think the international climate has radically changed. Agreements like TPP, which the powers that be supported, were not able to pass. In the era of Trump I highly doubt there is enough global trust for Russia or China to agree not to research militarized AI. Basically I don't see a single major power agreeing not to develop militarized AI in the near future.