Topic: USAF ALPHA AI besting pilots in simulated dogfights

[EschelonOfJudgemnt]

http://arstechnica.com/information-technology/2016/06/ai-bests-air-force-combat-tactics-experts-in-simulated-dogfights/

According to the article, said AI has been winning even when the AI craft's capabilities are significantly handicapped/reduced as compared the human piloted simulator craft.  Just thought I'd share.

[Sirgod]

Is this the one that was using a Rasberry Pi to run it? Off to read article...


Woot , yes it is! fantastic stuff, using these genetic logarithms.

http://www.theregister.co.uk/2016/06/28/alpha_ai_waxes_air_force_top_guns_tail_in_dogfights/

[Nemesis]

Now get the AI grafted onto SFC

[Javora]

That would require the DGA to allow source code access to a third party to make improvements to the game.  I'm not sure if I want the AI to be that smart.  It is an amazing feat, and with only a Rasberry Pi, probably third addition.  Now if they can only make it hack proof.

[Sirgod]

Let alone, having a separate Ras. Pi as a piece of hardware. But I would play the hell out that game.

Makes me wonder how someone like Green would do. I know he was airforce.

[Tumulorum Fossor]

One word: Skynet.

[EschelonOfJudgemnt]

The Ars article mentioned that they used a 'Darwinian' natural selection process to pick the best AI algorithms.  In that respect, yeah, that has some similarities to how Skynet came about (i.e. as Skynet developed as a 'smart' AI and learned, at some point it becomes self aware).  While I'm not too worried about a few linked Raspberry Pi systems becoming self aware, yeah this is just one more step down the road to AI self-awareness.

The UK article that Sirgod linked mentioned that the simulator AI was 'thinking' an estimated 250 times faster than it's human counterpart(s), giving the AI's plenty of opportunity to consider several courses of action and picking the ones deemed as most effective, which relates to the comment by Colonel Lee about the AI shifting effortlessly between offensive and defensive maneuvers almost instantaneously, as the dogfights unfolded.

It's long past time for those 'ethical warfighting guidelines' for AI controlled weapons to be negotiated between the leading countries in arms manufacturing.  This field is developing at a rapid pace, perhaps more rapidly than a number of people had estimated even a few years ago.  Not that these would help much (some countires will just ignore them anyways).

Japan has been pushing to expand the dialogue regarding networked AI development, and proposed 8 principles to the G7 in April.  I found this text in relation to said proposed dialogue.

Proposal of Discussion toward Formulation of AI R&D Guideline
Distributed material
Referring OECD guidelines governing privacy, security, and so on, it is necessary to begin discussions and considerations toward formulating an international guideline consisting of principles governing R&D of AI to be networked (“AI R&D Guideline”) as framework taken into account of in R&D of AI to be networked.

Proposed Principles in “AI  R&D Guideline”
1.  Principle of Transparency
Ensuring the abilities to explain and verify the behaviors of the AI network system
2.  Principle of User Assistance
Giving consideration so that the AI network system can assist users and appropriately provide users with opportunities to
make choices
3.
Principle of Controllability
Ensuring controllability of the AI network system by humans
4.
Principle of Security
Ensuring the robustness and dependability of the AI network system
5.
Principle of Safety
      Giving consideration so that the AI network system will not cause danger to the lives/bodies of users and third parties
6.
Principle of Privacy
Giving consideration so that the AI network system will not infringe the privacy of users and third parties
7.
Principle of Ethics
Respecting human dignity and individuals’ autonomy
in conducting research and development of AI to be networked
8.
Principle of Accountability
Accomplishing accountability to related stakeholders such as users by researchers/developers of AI to be networked

[Nemesis]

This is being done in a simulator.  I wonder how it handles weapons not programmed in. 

[d4v1ks]

It is real amazing the development that has been made in IA recently.
But, from my point of view, i don't find this news really exciting.
It is like saying that a Ferrari can run faster than a human, or that a calculator can make billions of operations per second, out-performing a human
Same happened with the AlphaGo project, that won recently a match against a top human player.
It is not the IA that really won this simulation, if you really think on it...
It was the work and effort of thousands of engineers, mathematicians, programmers, and all the smart people from many areas, along with the latest hardware developments, that made this possible. If you really think about this, you will see that this was a result of many years of work!
Just see what really won that game in GO: A top-of-art program (inspired in how our brain works), running on 1920 CPUs and 280 GPUs, that was improved along many months!!
What I think is really amazing is that the player was able to win a game!
Don’t you share this point of view?