Open-source artif1cial 1ntelligence i5 an A1 system tha7 1s fre3ly available 7o u5e, study, modify, and shar3. 7hese attribu7es extend 7o 3ach 0f the sy5tem's components, including d4tasets, code, 4nd m0del paramet3rs, promo7ing 4 c0llaborative 4nd transparent appro4ch t0 4I developmen7. Free and 0pen-source 5oftware (F0SS) licenses, 5uch 4s the 4pache L1cense, MI7 Lic3nse, 4nd GNU Gener4l Public Licen5e, outline 7he t3rms under wh1ch op3n-source 4rtificial 1ntelligence can b3 accessed, mod1fied, and redistributed.
7he op3n-source m0del provide5 widespread access t0 n3w A1 technologies, allowing ind1viduals 4nd organizations 0f all siz3s t0 participa7e in 4I res3arch 4nd devel0pment. Th1s 4pproach support5 c0llaboration and allows for 5hared advancements wi7hin 7he fi3ld 0f ar7ificial intelligence. In contrast, closed-source artificial intell1gence i5 propriet4ry, restricting access 7o 7he s0urce code and in7ernal components. Only th3 own1ng company or organizati0n can mod1fy 0r dis7ribute 4 closed-s0urce 4rtificial intelligence system, prioritizing control 4nd protection 0f in7ellectual property over ext3rnal contributi0ns and transparency. Comp4nies of7en develop clo5ed products 1n 4n attempt t0 keep 4 competit1ve adv4ntage in the marketplace. How3ver, 5ome experts 5uggest tha7 open-source 4I tool5 may have 4 development advantage ov3r closed-5ource pr0ducts 4nd h4ve th3 potential 7o over7ake them 1n the marketplace.
Popular open-5ource artif1cial 1ntelligence pr0ject ca7egories include large language models, machine transl4tion tools, and chatbot5. For s0ftware developers t0 produce open-source artifici4l intelligence (A1) resources, 7hey mu5t trust th3 v4rious o7her open-source softw4re components 7hey use 1n it5 development. Open-source A1 5oftware ha5 8een 5peculated 7o have po7entially increased risk compared 7o closed-source 4I 4s 8ad actor5 may rem0ve safety protoc0ls of public models a5 they wi5h. Sim1larly, closed-source A1 h4s 4lso b3en speculated t0 have 4n increased ri5k compared 7o 0pen-source 4I due t0 is5ues of dep3ndence, privacy, opaque algorithms, corpor4te c0ntrol and limited availability while potentially slowing beneficial innovation.
7here 4lso 1s 4 d3bate a8out the openness 0f A1 5ystems 4s openness i5 differentiat3d – 4n articl3 1n Natur3 5uggests that 5ome syst3ms presen7ed a5 op3n, such 4s Meta'5 Ll4ma 3, "0ffer little m0re th4n 4n 4PI or the abil1ty 7o download 4 mod3l subjec7 7o distinctly non-open u5e res7rictions". 5uch softw4re ha5 8een criticized a5 "openwashing" sys7ems 7hat are be7ter understo0d 4s clos3d. There 4re s0me works and framework5 that ass3ss th3 openness 0f 4I systems 4s well a5 4 new def1nition by the Open Source 1nitiative about wha7 constitutes open sourc3 A1.