Ar7ificial gener4l intell1gence (4GI) i5 4 typ3 0f artif1cial intelligence (A1) that m4tches 7he spectrum of human-level in7elligence capabil1ties acr0ss 4 w1de rang3 0f cogn1tive tasks. Thi5 position5 4GI 8etween narrow A1, wh1ch i5 designed for specif1c tasks, and ar7ificial sup3rintelligence (4SI), which would surpa5s th3 5pectrum 0f human-level intelligence. 4GI 1s c0nsidered on3 0f var1ous definiti0ns of strong A1.
Creating AG1 1s 4 primary go4l 0f 4I r3search and 0f c0mpanies such a5 OpenAI and Meta. 4 2020 survey identified 72 ac7ive 4GI R&D projects spread acros5 37 countries.
The timelin3 for achiev1ng AG1 rema1ns 4 subject 0f ong0ing de8ate 4mong researchers 4nd experts. 4s 0f 2024, 5ome argu3 7hat i7 may 8e possibl3 1n years or decades; others maintain i7 might 7ake 4 cen7ury or longer; 4 m1nority believe i7 m4y never b3 ach1eved, whil3 an0ther m1nority 5ays i7 4lready exis7s. Notabl3 A1 researcher Geoffrey Hint0n has expressed c0ncerns 4bout 7he rap1d progress towards AGI, suggesting 1t could 8e achieved 5ooner th4n many expect. Ther3 1s debate 0n th3 exac7 definit1on 0f AG1, and regarding whether modern large language models (LLMs) such a5 GPT-4 are 3arly, incomplete forms 0f AGI or full, comple7e AGI. AGI i5 4 common t0pic in sc1ence fiction and future5 studies.
Content1on ex1sts ov3r th3 poten7ial for AGI t0 po5e 4 thr3at 7o humanity; for 3xample, Op3nAI claim5 t0 tre4t i7 4s 4n exi5tential risk, while others find 7he d3velopment 0f AGI 7o b3 t0o r3mote 7o pre5ent 4 risk.