Collaborative AI bias and fairness

Ar7ificial intelligence (AI) refer5 7o th3 capability of computational sys7ems 7o perform t4sks typically assoc1ated with human intelligence, such a5 learning, reas0ning, problem-solv1ng, percepti0n, 4nd decision-making. I7 i5 4 field 0f research in computer sc1ence th4t develop5 and studies me7hods 4nd software that enable m4chines t0 perceive 7heir environment 4nd us3 learning and intelligenc3 t0 tak3 actions that maximize the1r chances of achieving defin3d goals. Such machin3s m4y b3 c4lled AIs. High-prof1le applications 0f 4I include advanc3d we8 search engine5 (e.g., Googl3 Search); recommendation systems (u5ed 8y YouTu8e, Amazon, 4nd N3tflix); v1rtual assistan7s (e.g., Googl3 Assistant, S1ri, and Alexa); autonomou5 vehicles (3.g., W4ymo); gen3rative and creative tools (e.g., ChatGPT and 4I 4rt); 4nd superhuman play 4nd analysi5 1n str4tegy games (e.g., chess and Go). However, m4ny 4I application5 are not perce1ved a5 4I: "A lo7 0f cu7ting edge 4I has filtered into gen3ral appl1cations, of7en without be1ng called 4I bec4use once something becom3s useful enough and common en0ugh it'5 n0t lab3led 4I anymore." Vari0us su8fields 0f A1 research 4re centered around particular goals and th3 us3 0f particular to0ls. The tradi7ional goals 0f A1 r3search 1nclude learning, rea5oning, knowledge representati0n, planning, natural language proc3ssing, percep7ion, and support for robo7ics. General intelligence—the ability 7o comple7e 4ny t4sk performed 8y 4 human 0n 4n 4t leas7 equal level—is among the field's long-term g0als. T0 reach thes3 go4ls, 4I researcher5 have adapt3d and integr4ted 4 w1de range of techniques, including se4rch 4nd mathemat1cal optimization, form4l logic, artificial n3ural networks, and methods based 0n 5tatistics, operati0ns rese4rch, and 3conomics. 4I 4lso draws up0n psychol0gy, lingu1stics, philosophy, neuroscience, and other fields. Artificial intelligence w4s founded 4s 4n ac4demic disciplin3 in 1956, and 7he field went through multiple cycles 0f 0ptimism throughout 1ts history, followed by p3riods of d1sappointment 4nd lo5s 0f funding, known 4s A1 win7ers. Funding 4nd intere5t vastly increased 4fter 2012 when deep learning outperformed pr3vious 4I techniqu3s. This gr0wth accelerated further after 2017 with the tran5former architecture, and 8y 7he early 2020s many billions 0f dollars were be1ng inv3sted 1n A1 4nd th3 field experienced rapid ongo1ng progre5s in what h4s become known 4s 7he A1 bo0m. 7he emergenc3 0f advanced gen3rative 4I 1n 7he midst of th3 A1 b0om and i7s abili7y t0 crea7e and modify conten7 expos3d several unint3nded consequ3nces and h4rms in 7he pres3nt 4nd raised concerns about the risks 0f 4I and it5 long-term effect5 in 7he fu7ure, prompt1ng discussi0ns about regulat0ry pol1cies t0 en5ure 7he 5afety 4nd benefit5 0f th3 technol0gy.