176:
camera. Furthermore, while there was a cap on the number of actions over a five-second window, AlphaStar was free to allocate its action quota unevenly across the window in order to launch superhuman bursts of activity at critical moments. DeepMind quickly retrained AlphaStar under more realistic constraints, and then lost a rematch with
Komincz. Starting in July 2019, the new, constrained version of AlphaStar anonymously competed against players who "opted in" on the public 1v1 European multiplayer ladder. By the end of August 2019, AlphaStar had attained "grandmaster level", ranking among the top 0.2 percent of human players.
1329:
1339:
244:
that "AlphaStar has become the first AI system to reach the top tier of human performance in any professionally played e-sport on the full unrestricted game under professionally approved conditions... Ever since computers cracked Go, chess and poker, the game of StarCraft has emerged, essentially by consensus from the community, as the next grand challenge for AI."
243:
DeepMind argues that insights from AlphaStar might benefit robots, self-driving cars and virtual assistants, which need to operate with "imperfectly observed information". Silver has indicated his lab "may rest at this point", rather than try to substantially improve AlphaStar. Silver himself argues
175:
and others argued that AlphaStar still had unfair advantages: "AlphaStar has the ability to make its clicks with surgical precision using an API, whereas human players are constrained by the mechanical limits of computer mice". AlphaStar also had a global view rather than being limited by the in-game
251:
argues, disapprovingly, that "military analysts will certainly be eyeing the successful AlphaStar real-time strategies as a clear example of the advantages of AI for battlefield planning". In contrast, Silver argues: "To say that this has any kind of military use is saying no more than to say an AI
188:
calls "the exploration problem": discovering new strategies would otherwise be like finding a "needle in a haystack". Agents then play each other and deploy reinforcement learning. These main agents also learn by playing against suboptimal "exploiter agents" whose purpose is to expose weaknesses in
216:
pro Raza "RazerBlader" Sekha stated AlphaStar was "impressive" but had its quirks, succumbing in one game to an unorthodox army composition made up of only air units. The UK's top player, Joshua "RiSky" Hayward, expressed some disappointment, saying AlphaStar "often didn't make the most efficient,
234:
expressed his continuing skepticism about deep learning, stating: "So far the field has struggled to take techniques like this out of the laboratory and game environments and into the real world, and I don't immediately see this result as progress in that direction". AI researcher Jon Dodge was
217:
strategic decisions". Pro Diego "Kelazhur" Schwimer called AlphaStar's play "unimaginably unusual; it really makes you question how much of StarCraft's diverse possibilities pro players have really explored". AlphaStar's opponents often did not realize they were playing a bot.
235:
surprised by AlphaStar, stating he did not expect such a "superhuman" performance for "another couple of years"; in contrast, Churchill states "StarCraft is nowhere near being 'solved', and AlphaStar is not yet even close to playing at a world champion level".
81:"real-time play, partial observability, no single dominant strategy, complex rules that make it hard to build a fast forward model, and a particularly large and varied action space." Though difficult,
154:
data into its software. In August 2017, DeepMind and
Blizzard released development tools to assist in bot development, as well as data from 65,000 past games. At the time, computer scientist and
171:
On 19 December 2018, DeepMind's bot defeated "a top professional player", Grzegorz "MaNa" Komincz, 5-0. DeepMind announced the bot, named "AlphaStar", on 24 January 2019. A journalist at
184:
Unlike AlphaZero, AlphaStar initially learns to imitate the moves of the best players in its database of human vs. human games; this step is necessary to solve what DeepMind's
168:, tech journalist Tom Simonite stated, "No one expects the robot to win anytime soon. But when it does, it will be a far greater achievement than DeepMind's conquest of Go."
226:
called AlphaStar a "landmark achievement" for the field of artificial intelligence. Churchill stated that he had previously seen bots that master one or two elements of
719:
Vinyals, Oriol; Babuschkin, Igor; Czarnecki, Wojciech M.; et al. (30 October 2019). "Grandmaster level in StarCraft II using multi-agent reinforcement learning".
381:
203:
assessed that AlphaStar's 2018 victory required "a fair amount of problem-specific effort" and that general-purpose methods were "not quite ready for StarCraft".
771:
449:
475:
659:
856:
604:
85:
may still be tractable with current technology because "its rules are known and the world is discrete with only a few types of objects".
630:
20:
1130:
501:
361:
1153:
917:
162:
guessed it would take five years for a bot to beat a human, but made the caveat that AlphaGo had beaten expectations. In
62:
1299:
295:
Arulkumaran, Kai; Antoine, Cully; Togelius, Julian (July 2019). "AlphaStar: An evolutionary computation perspective".
1368:
322:
185:
552:
577:
1231:
890:
66:
212:
judged AlphaStar's new constraints, adopted for the July 2019 matches, to be "fair" this time around.
147:
in
November 2016, alongside a plan to release an open development environment for bots in Q1 of 2017.
1342:
1288:
1378:
1332:
1024:
409:
Garisto, Dan (30 October 2019). "Google AI beats top human players at strategy game StarCraft II".
197:
After his 5-0 defeat in
December 2018, Komincz stated "I wasn't expecting the AI to be that good".
128:
119:
subsidiary in 2014, after demonstrating self-learning bots with superhuman ability at a variety of
822:
127:
predicted
Deepmind's research "leads to StarCraft in five or ten years". In March 2016, following
1243:
910:
46:
1363:
1163:
1018:
797:
140:
92:
1373:
1219:
1030:
143:
where (unlike Go) much of the "board" is invisible. A formal collaboration was announced at
728:
275:
8:
1294:
732:
526:
45:. It was unveiled to the public by name in January 2019. In a significant milestone for
903:
752:
430:
328:
300:
983:
756:
744:
434:
422:
357:
349:
318:
200:
58:
332:
1314:
1202:
1190:
736:
414:
310:
164:
32:
1136:
947:
827:
159:
660:"DeepMind's StarCraft 2 AI is now better than 99.8 percent of all human players"
1048:
694:
418:
104:
103:
DeepMind
Technologies was founded in the UK in 2010. As early as 2011, founder
884:
740:
450:"Computers That Crush Humans at Games Might Have Met Their Match: 'StarCraft'"
1357:
1196:
1096:
230:, but that AlphaStar was the first that can handle the game in its entirety.
124:
314:
297:
Proceedings of the
Genetic and Evolutionary Computation Conference Companion
1309:
989:
942:
748:
426:
248:
222:
172:
41:
1304:
1266:
266:
231:
527:"Google's Artificial Intelligence Project May Start Playing Video Games"
1158:
1056:
631:"DeepMind AI is secretly lurking on the public StarCraft II 1v1 ladder"
120:
772:"Mind meld: Artificial intelligence is improving the way humans think"
605:"An AI crushed two human pros at StarCraft—but it wasn't a fair fight"
382:"A.I. Mastered Backgammon, Chess and Go. Now It Takes On StarCraft II"
1249:
1078:
995:
926:
261:
132:
112:
553:"Google DeepMind's next gaming challenge: can AI beat StarCraft II?"
354:
Human
Compatible: Artificial Intelligence and the Problem of Control
352:(8 October 2019). "Chapter 2: Intelligence in Humans and Machines".
1090:
476:"Google buys UK artificial intelligence startup Deepmind for £400m"
305:
208:
144:
36:
1124:
977:
718:
502:"DeepMind founder Demis Hassabis on how AI will shape the future"
70:
1001:
937:
270:
116:
823:"DeepMind claims breakthrough as AI masters StarCraft II game"
1225:
857:"AI becomes grandmaster in 'fiendishly complex' StarCraft II"
895:
798:"DeepMind has finally thrashed humans at StarCraft for real"
344:
342:
1237:
252:
for chess could be used to lead to military applications".
294:
91:
is a popular fast-paced online real-time strategy game by
339:
73:(2016) were considered major milestones; some argue that
49:, AlphaStar attained Grandmaster status in August 2019.
850:
848:
846:
578:"Google DeepMind AI Declares Galactic War on StarCraft"
843:
653:
651:
135:
player, Hassabis publicly mulled building an AI for
695:"DeepMind claims landmark moment for AI in esports"
648:
150:By 2017, DeepMind was experimenting with feeding
1355:
57:Games created for humans are considered to have
911:
404:
402:
375:
373:
123:games. In February 2015, computer scientist
918:
904:
399:
370:
21:AlphaStar (satellite broadcasting service)
304:
77:would also be a major milestone, due to
408:
348:
1356:
854:
769:
692:
688:
686:
684:
682:
680:
628:
550:
499:
16:Software designed to play StarCraft II
899:
657:
473:
447:
379:
139:, citing it as a strategic game with
1338:
111:"the next step up" after games like
1154:Quantum Artificial Intelligence Lab
886:DeepMind StarCraft II Demonstration
795:
677:
603:Lee, Timothy B. (30 January 2019).
602:
63:progress in artificial intelligence
13:
1300:Generative pre-trained transformer
770:Heaven, Douglas (21 August 2019).
19:For the broadcasting service, see
14:
1390:
878:
474:Gibbs, Samuel (27 January 2014).
448:Cheng, Jonathan (22 April 2016).
1337:
1328:
1327:
380:Daley, Jason (30 October 2019).
129:AlphaGo's victory over Lee Sedol
855:Sample, Ian (30 October 2019).
815:
789:
763:
712:
693:Kelion, Leo (30 October 2019).
658:Statt, Nick (30 October 2019).
622:
596:
570:
551:Byford, Sam (4 November 2016).
544:
519:
493:
467:
441:
356:(Hardcover ed.). Viking.
288:
1:
925:
796:Lee, Alex (30 October 2019).
500:Byford, Sam (10 March 2016).
281:
179:
52:
629:Amadeo, Ron (11 July 2019).
192:
7:
255:
10:
1395:
419:10.1038/d41586-019-03298-6
98:
39:that plays the video game
18:
1323:
1289:Attention Is All You Need
1280:
1259:
1212:
1183:
1176:
1146:
1117:
1110:
1071:
1040:
1011:
970:
963:
956:
933:
741:10.1038/s41586-019-1724-z
238:
1369:Applied machine learning
315:10.1145/3319619.3321894
47:artificial intelligence
1164:Tensor Processing Unit
141:incomplete information
93:Blizzard Entertainment
69:(1997) and DeepMind's
65:. IBM's chess-playing
386:Smithsonian Magazine
276:Pluribus (poker bot)
269:, a similar bot for
115:. DeepMind became a
1295:Future of Go Summit
733:2019Natur.575..350V
454:Wall Street Journal
247:Computer scientist
158:tournament manager
131:, a world champion
1041:In popular culture
299:. pp. 314–5.
1351:
1350:
1276:
1275:
1172:
1171:
1106:
1105:
1067:
1066:
957:Computer programs
727:(7782): 350–354.
363:978-0-525-55861-3
189:the main agents.
61:as benchmarks of
59:external validity
1386:
1341:
1340:
1331:
1330:
1315:Google Workspace
1181:
1180:
1115:
1114:
1111:Machine learning
968:
967:
961:
960:
920:
913:
906:
897:
896:
887:
872:
871:
869:
867:
852:
841:
840:
838:
836:
819:
813:
812:
810:
808:
793:
787:
786:
784:
782:
767:
761:
760:
716:
710:
709:
707:
705:
690:
675:
674:
672:
670:
655:
646:
645:
643:
641:
626:
620:
619:
617:
615:
600:
594:
593:
591:
589:
574:
568:
567:
565:
563:
548:
542:
541:
539:
537:
523:
517:
516:
514:
512:
497:
491:
490:
488:
486:
471:
465:
464:
462:
460:
445:
439:
438:
406:
397:
396:
394:
392:
377:
368:
367:
346:
337:
336:
308:
292:
33:computer program
1394:
1393:
1389:
1388:
1387:
1385:
1384:
1383:
1379:Google DeepMind
1354:
1353:
1352:
1347:
1319:
1272:
1255:
1213:Language models
1208:
1168:
1142:
1118:Neural networks
1102:
1063:
1036:
1007:
952:
948:Google DeepMind
929:
924:
885:
881:
876:
875:
865:
863:
853:
844:
834:
832:
828:Financial Times
821:
820:
816:
806:
804:
794:
790:
780:
778:
768:
764:
717:
713:
703:
701:
691:
678:
668:
666:
656:
649:
639:
637:
627:
623:
613:
611:
601:
597:
587:
585:
584:. 9 August 2017
576:
575:
571:
561:
559:
549:
545:
535:
533:
525:
524:
520:
510:
508:
498:
494:
484:
482:
472:
468:
458:
456:
446:
442:
407:
400:
390:
388:
378:
371:
364:
350:Russell, Stuart
347:
340:
325:
293:
289:
284:
258:
241:
195:
182:
160:David Churchill
101:
55:
24:
17:
12:
11:
5:
1392:
1382:
1381:
1376:
1371:
1366:
1349:
1348:
1346:
1345:
1335:
1324:
1321:
1320:
1318:
1317:
1312:
1307:
1302:
1297:
1292:
1284:
1282:
1278:
1277:
1274:
1273:
1271:
1270:
1263:
1261:
1257:
1256:
1254:
1253:
1247:
1241:
1235:
1229:
1223:
1216:
1214:
1210:
1209:
1207:
1206:
1200:
1194:
1187:
1185:
1178:
1174:
1173:
1170:
1169:
1167:
1166:
1161:
1156:
1150:
1148:
1144:
1143:
1141:
1140:
1134:
1128:
1121:
1119:
1112:
1108:
1107:
1104:
1103:
1101:
1100:
1094:
1088:
1082:
1075:
1073:
1069:
1068:
1065:
1064:
1062:
1061:
1053:
1044:
1042:
1038:
1037:
1035:
1034:
1028:
1022:
1015:
1013:
1009:
1008:
1006:
1005:
999:
993:
987:
981:
974:
972:
965:
958:
954:
953:
951:
950:
945:
940:
934:
931:
930:
923:
922:
915:
908:
900:
894:
893:
880:
879:External links
877:
874:
873:
842:
814:
788:
762:
711:
676:
647:
621:
595:
569:
543:
518:
492:
466:
440:
398:
369:
362:
338:
323:
286:
285:
283:
280:
279:
278:
273:
264:
257:
254:
240:
237:
220:Ian Sample of
206:An article in
201:Stuart Russell
194:
191:
181:
178:
105:Demis Hassabis
100:
97:
54:
51:
15:
9:
6:
4:
3:
2:
1391:
1380:
1377:
1375:
1372:
1370:
1367:
1365:
1364:2018 software
1362:
1361:
1359:
1344:
1336:
1334:
1326:
1325:
1322:
1316:
1313:
1311:
1308:
1306:
1303:
1301:
1298:
1296:
1293:
1290:
1286:
1285:
1283:
1279:
1268:
1265:
1264:
1262:
1258:
1251:
1248:
1245:
1242:
1239:
1236:
1233:
1230:
1227:
1224:
1221:
1218:
1217:
1215:
1211:
1204:
1201:
1198:
1195:
1192:
1189:
1188:
1186:
1182:
1179:
1177:Generative AI
1175:
1165:
1162:
1160:
1157:
1155:
1152:
1151:
1149:
1145:
1138:
1135:
1132:
1129:
1126:
1123:
1122:
1120:
1116:
1113:
1109:
1098:
1097:AlphaGeometry
1095:
1092:
1089:
1086:
1083:
1080:
1077:
1076:
1074:
1070:
1059:
1058:
1054:
1051:
1050:
1046:
1045:
1043:
1039:
1032:
1029:
1026:
1023:
1020:
1017:
1016:
1014:
1010:
1003:
1000:
997:
994:
991:
988:
985:
982:
979:
976:
975:
973:
969:
966:
962:
959:
955:
949:
946:
944:
941:
939:
936:
935:
932:
928:
921:
916:
914:
909:
907:
902:
901:
898:
892:
888:
883:
882:
862:
858:
851:
849:
847:
830:
829:
824:
818:
803:
799:
792:
777:
776:New Scientist
773:
766:
758:
754:
750:
746:
742:
738:
734:
730:
726:
722:
715:
700:
696:
689:
687:
685:
683:
681:
665:
661:
654:
652:
636:
632:
625:
610:
606:
599:
583:
579:
573:
558:
554:
547:
532:
528:
522:
507:
503:
496:
481:
477:
470:
455:
451:
444:
436:
432:
428:
424:
420:
416:
412:
405:
403:
387:
383:
376:
374:
365:
359:
355:
351:
345:
343:
334:
330:
326:
324:9781450367486
320:
316:
312:
307:
302:
298:
291:
287:
277:
274:
272:
268:
265:
263:
260:
259:
253:
250:
245:
236:
233:
229:
225:
224:
218:
215:
211:
210:
204:
202:
198:
190:
187:
177:
174:
169:
167:
166:
161:
157:
153:
148:
146:
142:
138:
134:
130:
126:
125:Zachary Mason
122:
118:
114:
110:
106:
96:
94:
90:
86:
84:
80:
76:
72:
68:
64:
60:
50:
48:
44:
43:
38:
34:
30:
26:
22:
1374:StarCraft II
1310:Google Pixel
1084:
1055:
1047:
1012:Competitions
990:AlphaGo Zero
943:Google Brain
864:. Retrieved
861:The Guardian
860:
833:. Retrieved
826:
817:
805:. Retrieved
801:
791:
779:. Retrieved
775:
765:
724:
720:
714:
702:. Retrieved
698:
667:. Retrieved
663:
638:. Retrieved
635:Ars Technica
634:
624:
612:. Retrieved
609:Ars Technica
608:
598:
586:. Retrieved
581:
572:
560:. Retrieved
556:
546:
534:. Retrieved
530:
521:
509:. Retrieved
505:
495:
483:. Retrieved
480:The Guardian
479:
469:
457:. Retrieved
453:
443:
410:
389:. Retrieved
385:
353:
296:
290:
249:Noel Sharkey
246:
242:
227:
223:The Guardian
221:
219:
213:
207:
205:
199:
196:
183:
173:Ars Technica
170:
163:
155:
151:
149:
136:
108:
102:
89:StarCraft II
88:
87:
82:
78:
74:
56:
42:StarCraft II
40:
28:
27:
25:
1305:Google Labs
1131:Transformer
866:26 February
835:26 February
807:26 February
781:26 February
704:26 February
669:26 February
640:26 February
614:26 February
588:26 February
562:26 February
536:26 February
511:26 February
485:26 February
459:26 February
391:26 February
267:OpenAI Five
232:Gary Marcus
186:Dave Silver
79:StarCraft's
1358:Categories
1232:Chinchilla
1159:TensorFlow
1057:The MANIAC
306:1902.01724
282:References
180:Algorithms
121:Atari 2600
53:Background
1250:VideoPoet
1191:Assistant
1085:AlphaStar
1079:AlphaFold
1025:Lee Sedol
996:AlphaZero
927:Google AI
757:204972004
664:The Verge
557:The Verge
506:The Verge
435:208116679
262:AlphaZero
228:StarCraft
214:StarCraft
193:Reactions
156:StarCraft
152:StarCraft
137:StarCraft
109:StarCraft
83:StarCraft
75:StarCraft
67:Deep Blue
29:AlphaStar
1333:Category
1281:See also
1184:Chatbots
1091:AlphaDev
971:Versions
802:Wired UK
749:31666705
699:BBC News
427:33116320
333:59604439
256:See also
209:Wired UK
145:BlizzCon
37:DeepMind
1343:Commons
1197:Sparrow
1125:WaveNet
1049:AlphaGo
1019:Fan Hui
978:AlphaGo
964:AlphaGo
891:YouTube
729:Bibcode
531:Fortune
107:called
99:History
71:AlphaGo
1269:(2024)
1252:(2024)
1246:(2023)
1244:Gemini
1240:(2022)
1234:(2022)
1228:(2021)
1222:(2018)
1205:(2023)
1203:Gemini
1199:(2022)
1193:(2016)
1139:(2022)
1133:(2017)
1127:(2016)
1099:(2024)
1093:(2023)
1087:(2019)
1081:(2018)
1060:(2023)
1052:(2017)
1033:(2017)
1031:Ke Jie
1027:(2016)
1021:(2015)
1004:(2019)
1002:MuZero
998:(2017)
992:(2017)
986:(2016)
984:Master
980:(2015)
938:Google
831:. 2019
755:
747:
721:Nature
433:
425:
411:Nature
360:
331:
321:
271:Dota 2
239:Legacy
117:Google
1260:Other
1226:LaMDA
1147:Other
1072:Other
753:S2CID
582:Wired
431:S2CID
329:S2CID
301:arXiv
165:Wired
31:is a
1267:Vids
1238:PaLM
1220:BERT
1137:Gato
868:2020
837:2020
809:2020
783:2020
745:PMID
706:2020
671:2020
642:2020
616:2020
590:2020
564:2020
538:2020
513:2020
487:2020
461:2020
423:PMID
393:2020
358:ISBN
319:ISBN
889:on
737:doi
725:575
415:doi
311:doi
35:by
1360::
859:.
845:^
825:.
800:.
774:.
751:.
743:.
735:.
723:.
697:.
679:^
662:.
650:^
633:.
607:.
580:.
555:.
529:.
504:.
478:.
452:.
429:.
421:.
413:.
401:^
384:.
372:^
341:^
327:.
317:.
309:.
133:Go
113:Go
95:.
1291:"
1287:"
919:e
912:t
905:v
870:.
839:.
811:.
785:.
759:.
739::
731::
708:.
673:.
644:.
618:.
592:.
566:.
540:.
515:.
489:.
463:.
437:.
417::
395:.
366:.
335:.
313::
303::
23:.
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.