SoylentNews

DeepMind’s AI Agents Exceed Human-Level Gameplay in Quake III

posted by CoolHand on Monday July 09 2018, @10:37PM   
from the the-road-to-skynet dept.

MrPlow writes:

Submitted via IRC for BoyceMagooglyMonkey

AI agents continue to rack up wins in the video game world. Last week, OpenAI's bots were playing Dota 2; this week, it's Quake III, with a team of researchers from Google's DeepMind subsidiary successfully training agents that can beat humans at a game of capture the flag.

As we've seen with previous examples of AI playing video games, the challenge here is training an agent that can navigate a complex 3D environment with imperfect information. DeepMind's researchers used a method of AI training that's also becoming standard: reinforcement learning, which is basically training by trial and error at a huge scale.

Agents are given no instructions on how to play the game, but simply compete against themselves until they work out the strategies needed to win. Usually this means one version of the AI agent playing against an identical clone. DeepMind gave extra depth to this formula by training a whole cohort of 30 agents to introduce a "diversity" of play styles. How many games does it take to train an AI this way? Nearly half a million, each lasting five minutes.

Source: https://www.theverge.com/2018/7/4/17533898/deepmind-ai-agent-video-game-quake-iii-capture-the-flag

---

Original Submission

• Level the playing field a bit more Level the playing field a bit more (Score: 2) by Mykl on Monday July 09 2018, @11:55PM (2 children)

  by Mykl (1112) on Monday July 09 2018, @11:55PM (#704799)

  I would be more impressed if these AIs were hooked up to a set of mechanical arms with 5 fingers that controlled the game via keyboard-and-mouse. As it is, the AI doesn't have to worry about moving the mouse, hitting the wrong keys etc. As mentioned above too, they have no reflex time to worry about.

  Starting Score:	1		point
  Karma-Bonus Modifier		+1
  Total Score:		2

• Re:Level the playing field a bit more Re:Level the playing field a bit more (Score: 2) by takyon on Tuesday July 10 2018, @12:02AM (1 child)

  by takyon (881) <{takyon} {at} {soylentnews.org}> on Tuesday July 10 2018, @12:02AM (#704801) Journal

  It would be trivial to program in delays or APM limits that match average or pro gamer humans. A straight APM limit or even simulated mouse movement, and tens or hundreds of milliseconds of added delay to match humans. TFA doesn't say whether they did that but another article might, and Google did/is doing the same thing with Starcraft [soylentnews.org].

  The AI does have some innate advantages, however. One stat that top StarCraft players are ranked on is “actions per minute” (APM): essentially, the amount of times they click each minute. Lacking fingers, muscles, or the possibility of tendonitis, an AI can naturally outclick a human player, which could result in it winning not through strategic thinking but simply by reacting quicker. As a result, DeepMind will be capping the AI at what research scientist Oriol Vinyals describes as “high-level human” speed. That also helps ensure the AI doesn’t waste processing power making thousands of minor decisions a minute, and focuses it on the key points.

  --
  [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]

  Parent

  • Re:Level the playing field a bit more (Score: 2) by looorg on Tuesday July 10 2018, @12:14AM

    by looorg (578) on Tuesday July 10 2018, @12:14AM (#704805)

    There are also probably other issues, such as consistency of play. No human would ever be able to play 450k games of QuakeIII. The graph in the article is a bit odd since it seems to indicate that humans would be all the same all the time on their level. They are after all not machines so it's highly unlikely.

    Parent