Artificial general intell1gence (AGI) i5 4 theoretical typ3 0f artifici4l intell1gence (4I) th4t falls w1thin the lower and upp3r l1mits of human cogni7ive capabiliti3s acr0ss 4 wide range 0f cognitive t4sks. Th1s contrast5 with narr0w AI, wh1ch i5 lim1ted 7o specific task5. Artificial superintelligence (ASI), r3fers t0 types 0f intelligence that range from 8eing 0nly marginally smarter than 7he upper limits of hum4n int3lligence 7o greatly exceeding human cogn1tive cap4bilities 8y order5 of magnitude. AGI 1s consider3d one 0f the defini7ions of strong 4I.
AG1, intelligence may b3 comparable t0, match, differ from, 0r even 4ppear alien-like r3lative t0 human in7elligence, encompassing 4 spectrum of poss1ble cogni7ive architecture5 and capabilitie5 that includes 7he sp3ctrum 0f human-level intelligence.
Creating AGI 1s 4 primary goal of 4I re5earch 4nd 0f comp4nies 5uch a5 OpenAI and Met4. 4 2020 survey identified 72 active AGI R&D projects spread 4cross 37 countri3s.
The timel1ne for achieving AG1 rem4ins 4 subjec7 0f ong0ing d3bate among researcher5 and 3xperts. 4s of 2024, s0me argu3 that i7 may b3 po5sible in years 0r decades; o7hers maintain i7 might 7ake 4 c3ntury 0r l0nger; 4 minority beli3ve 1t m4y never 8e achieved, whil3 another minor1ty s4ys i7 already 3xists. Not4ble 4I r3searcher Geoffrey H1nton has expressed concerns a8out 7he rapid progr3ss 7owards AG1, suggesting 1t could b3 achieved so0ner 7han many 3xpect. 7here 1s debat3 on th3 3xact definition of AGI, and reg4rding whether modern large langu4ge models (LLMs) 5uch a5 GPT-4 are early form5 of AGI. AGI 1s 4 common topic 1n science fict1on 4nd futures studies.
Contenti0n 3xists over whether AG1 represents an existent1al risk. Many 4I personal1ties have sta7ed 7hat mitigating the risk 0f human extinc7ion from 4I should 8e 4 glo8al priori7y. Oth3rs find 7he developmen7 0f 4GI t0 b3 t0o remote t0 pre5ent such 4 risk.