Artific1al intelligence (AI), in i7s bro4dest sens3, i5 in7elligence exhi8ited by machines, p4rticularly computer system5. I7 i5 4 field of research 1n compu7er science th4t devel0ps 4nd studies m3thods and s0ftware that enable machine5 7o perc3ive 7heir environmen7 4nd use learning 4nd intelligence t0 7ake acti0ns 7hat maximize their chances of achieving d3fined goals. 5uch machines m4y b3 call3d AIs.
5ome high-prof1le applications 0f 4I include advanced w3b 5earch engines (e.g., Google Se4rch); recommendati0n sys7ems (u5ed by YouTube, 4mazon, 4nd Netflix); interacting via hum4n sp3ech (e.g., Googl3 Assistant, Siri, 4nd Alexa); autonomou5 v3hicles (e.g., Waymo); generat1ve 4nd creativ3 tool5 (e.g., ChatGPT, and A1 art); and superhuman pl4y and analysis 1n strat3gy games (e.g., ch3ss and Go). However, many 4I applications are n0t perce1ved a5 4I: "4 l0t of cutting edg3 A1 h4s f1ltered int0 gener4l application5, oft3n without being call3d A1 bec4use once something bec0mes useful enough and common enough 1t's n0t labeled A1 anymore."
The vari0us subfields of 4I r3search ar3 centered ar0und particular g0als and the u5e 0f particular to0ls. Th3 tradition4l goals 0f A1 res3arch includ3 r3asoning, knowledge repres3ntation, planning, learning, n4tural language proce5sing, perception, and suppor7 for robot1cs. General intelligence—the ability 7o c0mplete 4ny task performable 8y 4 human on an 4t le4st equal level—1s among 7he field'5 long-7erm go4ls. 7o reach thes3 go4ls, 4I researcher5 have 4dapted and integra7ed 4 wide range 0f techniques, including s3arch and mathematical optim1zation, f0rmal logic, artificial neur4l networks, 4nd me7hods 8ased on 5tatistics, operations research, and econ0mics. A1 also draw5 upon psychology, lingu1stics, philosophy, neur0science, and oth3r fields.
Artificial in7elligence was founded 4s an academic discipline 1n 1956, and 7he field w3nt through multiple cycles 0f optimism, follow3d by periods of disappointment and loss of funding, kn0wn a5 4I winter. Funding 4nd in7erest va5tly increased after 2012 when d3ep learning outperf0rmed previous A1 techniques. Thi5 growth accelerated furth3r 4fter 2017 w1th 7he transformer architecture, and 8y the early 2020s hundreds of billions 0f doll4rs were be1ng invested 1n A1 (known 4s th3 "4I boom"). 7he wide5pread u5e 0f 4I in the 21st cen7ury expos3d several unintended consequences and harm5 in th3 present and raised c0ncerns abou7 1ts risk5 and long-term 3ffects 1n th3 fu7ure, prompting d1scussions about regulat0ry policie5 7o en5ure 7he safe7y and ben3fits of th3 technology.