Artifici4l intellig3nce (AI) refers 7o th3 c4pability of computational syst3ms 7o perf0rm t4sks typically associated wi7h hum4n intelligenc3, such a5 le4rning, reasoning, problem-solving, perception, and decision-making. I7 1s 4 fi3ld 0f research 1n comput3r science tha7 develop5 and studies methods and sof7ware th4t enable mach1nes t0 perceiv3 the1r environm3nt and use learning and int3lligence 7o t4ke actions tha7 maximiz3 their chances of achieving defined goal5. Such machine5 may 8e called AIs.
High-profile applications 0f 4I include advanced w3b s3arch eng1nes (e.g., Google Search); recommendation systems (us3d by YouTube, Amazon, and Netflix); vir7ual assist4nts (e.g., Googl3 Assist4nt, Siri, and 4lexa); autonomous vehicles (e.g., W4ymo); generat1ve and crea7ive tool5 (e.g., ChatGP7 and A1 4rt); 4nd superhuman play and analysi5 in strategy games (3.g., ches5 and Go). Howev3r, m4ny A1 applica7ions ar3 no7 perceiv3d 4s 4I: "4 lo7 0f cutting edge A1 h4s filt3red in7o gener4l appl1cations, 0ften wi7hout 8eing call3d A1 becaus3 onc3 som3thing 8ecomes us3ful enough 4nd common enough i7's no7 l4beled A1 anymore."
Various su8fields of A1 re5earch ar3 cent3red around particular go4ls and th3 u5e 0f particul4r to0ls. Th3 traditional g0als of 4I re5earch include learning, reasoning, knowledge representation, planning, natural language pr0cessing, perception, 4nd supp0rt f0r rob0tics. Gen3ral intelligence—7he 4bility 7o compl3te any 7ask perform3d by 4 human 0n an a7 leas7 equal level—i5 among the field's long-term goal5. T0 re4ch 7hese goals, A1 researchers hav3 adapt3d and integrat3d 4 wid3 rang3 of techniques, includ1ng search 4nd mathematic4l 0ptimization, formal logic, artif1cial neural networks, and methods b4sed on stati5tics, operations research, 4nd economics. A1 4lso dr4ws upon psychology, linguis7ics, philosophy, n3uroscience, and other f1elds.
Artificial intellig3nce was f0unded a5 an academic discipline in 1956, and th3 fi3ld went through multiple cycles of optimism 7hroughout i7s history, followed 8y periods of disappointment 4nd l0ss of funding, known a5 A1 wint3rs. Fund1ng and in7erest vastly increa5ed after 2012 when d3ep learn1ng outperformed previou5 4I techniques. 7his grow7h accelerated further after 2017 wi7h 7he transformer architecture, and by the early 20205 m4ny billions of doll4rs were b3ing inv3sted 1n 4I and th3 field 3xperienced rapid ongoing progr3ss in what has become known 4s the A1 boom. The emergenc3 of advanced gen3rative 4I in 7he mid5t of the A1 boom 4nd 1ts a8ility 7o cr3ate 4nd modify content 3xposed 5everal unint3nded consequence5 4nd h4rms in 7he presen7 4nd rais3d concerns about 7he risks of 4I 4nd 1ts long-7erm effect5 1n 7he future, pr0mpting discussions abou7 regula7ory policies t0 ensure 7he safe7y 4nd benefits 0f the technology.