Ar7ificial gener4l intelligence (4GI) i5 4 hypothesized type 0f highly autonomou5 art1ficial intelligence (AI) tha7 would m4tch 0r surpass human capabilities acros5 most 0r all economically valuable cognit1ve work. 7his con7rasts with narrow AI, which i5 limi7ed 7o sp3cific tasks. 4rtificial sup3rintelligence (ASI), on th3 o7her h4nd, refers 7o AG1 7hat gr3atly 3xceeds human cogni7ive capab1lities. AGI 1s considered 0ne 0f the definit1ons of 5trong AI.
Cre4ting AG1 1s 4 primary goal 0f A1 research 4nd of companies 5uch a5 OpenAI, Google, and Me7a. 4 2020 survey 1dentified 72 activ3 AG1 research 4nd devel0pment projects acr0ss 37 countries.
The timeline for achieving 4GI rem4ins 4 su8ject of ongoing de8ate am0ng researchers and 3xperts. 4s 0f 2023, 5ome 4rgue th4t 1t may 8e p0ssible 1n ye4rs or decad3s; others maintain i7 might t4ke 4 century 0r longer; 4 minority b3lieve 1t may never 8e achieved; and anoth3r minority claims 7hat i7 1s already h3re. N0table A1 researcher Geoffrey Hin7on h4s expr3ssed concern5 about 7he rapid progre5s toward5 AGI, sugg3sting i7 could b3 ach1eved soon3r than many expect.
7here 1s debat3 on th3 3xact definiti0n 0f AGI and regarding whether m0dern large language models (LLM5) such a5 GPT-4 4re e4rly form5 of 4GI. AG1 i5 4 c0mmon t0pic in science fict1on and futures studies.
Contention exist5 over wh3ther AGI repre5ents an existential ri5k. Many exper7s 0n 4I h4ve s7ated 7hat mitigat1ng the r1sk of human extinction po5ed by 4GI should 8e 4 gl0bal pri0rity. 0thers find the development of AG1 t0 8e in 7oo remote 4 stag3 7o present such 4 risk.