Brockman: Simply by simulation – you let m calculate a lot. Algorithms n extract DieRegeln, according to which AI has to act. The essential is not that it n responds to problem A with solution B. But that it learns to generalise: If a similar problem arises as a, but which is not exactly problem a, it also finds a solution. In case of robot hand, we have randomized some parameters: size of cube, coefficient of friction, some aspects of environment. The geschahalles in simulation. When we tried robot hand for first time in Derrealen world, it turned out that five can auchgeneralisieren when acting in world of REM. Five comes clear with changed parameters in a real environment.
Online Time: But how Trainiertman an AI for a computer game?
Brockman: Five Hatim time lapse hundreds years DotA 2 against or Kis played before it videocombi times against a man has entered. It turned out that Diekünstliche finds intelligence: people behave strangely. Very different from Kis. However, five was able to get into crowd. Five has something like depth. This is also necessary, comparison Kis wants to build, which will in future take considerable responsibility in real world delivery and autonomous.
if ( typeof AdController !== ‘undefined’ !window.Zeit.isMobileView()) { if ( !document.getElementById( ‘iqadtile4’ ) ) { var elem = document.createElement( ‘div’ ); elem.id = ‘iqadtile4’; elem.className = “ad ad-desktop ad-desktop–4 ad-desktop–4-on-article”; elem.setAttribute(‘data-banner-type’, ‘desktop’); document.getElementById(‘ad-desktop-4’).parentNode.appendChild(elem); AdController.render(‘iqadtile4’); if ( window.console typeof window.console.info === ‘function’ ) { window.console.info(‘AdController ‘ AdController.VERSION ‘ tile 4 desktop’) } } } if ( typeof AdController !== ‘undefined’ window.Zeit.isMobileView()) { if ( !document.getElementById( ‘iqadtile4’ ) ) { var elem = document.createElement( ‘div’ ); elem.id = ‘iqadtile4’; elem.className = “ad ad-mobile ad-mobile–4 ad-mobile–4-on-article”; elem.setAttribute(‘data-banner-type’, ‘mobile’); document.getElementById(‘ad-mobile-4’).parentNode.appendChild(elem); AdController.render(‘iqadtile4’); if ( window.console typeof window.console.info === ‘function’ ) { window.console.info(‘AdController ‘ AdController.VERSION ‘ tile 4 mobile’) } } }
Online Time: Why do people verhaltensich strange for artificial intelligence?
Brockman: This shows up in very different ways. One example is character selection in DotA 2, DieFiguren is called Heroes. We have built AI interface in such a way that we ourselves sehenkönnen for how good five assesses individual heroes. In competition, Diemenschlichen players chose ir recognizable after that, which are especially popular among gamers. The AI thought this was not a good idea, she was looking for her heroes according to her own Berechnungenaus, whom she considered powerful. Based on this, five divided before first match, Esbesitze only because of figure selection of both sides already a 95-percent chance of winning. What made human players somewhat perplexed: ir experience of various figures spoke clearly against it – how could KIderart victory be certain, even though she had chosen seemingly weak heroes?
