{"id":120291,"date":"2025-10-22T12:09:55","date_gmt":"2025-10-22T05:09:55","guid":{"rendered":"https:\/\/tino.vn\/blog\/?p=120291"},"modified":"2026-10-30T11:00:20","modified_gmt":"2026-10-30T04:00:20","slug":"sentient-ai-la-gi","status":"publish","type":"post","link":"https:\/\/tino.vn\/blog\/sentient-ai-la-gi\/","title":{"rendered":"Sentient AI l\u00e0 g\u00ec? T\u1eeb phim vi\u1ec5n t\u01b0\u1edfng \u0111\u1ebfn hi\u1ec7n th\u1ef1c c\u00f2n bao xa?"},"content":{"rendered":"\n

Sentient AI (AI c\u00f3 tri gi\u00e1c) l\u00e0 AI c\u00f3 tr\u1ea3i nghi\u1ec7m ch\u1ee7 quan th\u1eadt: c\u1ea3m th\u1ea5y vui, s\u1ee3, \u0111au v\u00e0 bi\u1ebft m\u00ecnh \u0111ang t\u1ed3n t\u1ea1i, ch\u1ee9 kh\u00f4ng ch\u1ec9 n\u00f3i nh\u01b0 th\u1ec3 c\u00f3 c\u1ea3m x\u00fac. \u0110\u1ebfn th\u00e1ng 10\/2026, ch\u01b0a c\u00f3 h\u1ec7 th\u1ed1ng AI n\u00e0o \u0111\u01b0\u1ee3c gi\u1edbi khoa h\u1ecdc c\u00f4ng nh\u1eadn l\u00e0 c\u00f3 tri gi\u00e1c; \u0111\u00e2y v\u1eabn l\u00e0 kh\u00e1i ni\u1ec7m gi\u1ea3 \u0111\u1ecbnh.<\/strong><\/p>\n\n\n\n

C\u00e1c chatbot AI ph\u1ed5 bi\u1ebfn hi\u1ec7n nay<\/a> nh\u01b0 ChatGPT, Gemini hay Claude ch\u1ea1y tr\u00ean m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn (LLM), c\u0169ng l\u00e0 lo\u1ea1i model l\u1eadp tr\u00ecnh vi\u00ean g\u1ecdi qua API (c\u00e1ch l\u1ea5y LLM API key mi\u1ec5n ph\u00ed<\/a>). Ch\u00fang tr\u00f2 chuy\u1ec7n r\u1ea5t gi\u1ed1ng ng\u01b0\u1eddi nh\u01b0ng v\u1eabn l\u00e0 h\u1ec7 th\u1ed1ng d\u1ef1 \u0111o\u00e1n ch\u1eef. B\u00e0i vi\u1ebft gi\u1ea3i th\u00edch tri gi\u00e1c \u1edf AI l\u00e0 g\u00ec, gi\u1edbi nghi\u00ean c\u1ee9u ki\u1ec3m tra n\u00f3 b\u1eb1ng c\u00e1ch n\u00e0o v\u00e0 v\u00ec sao kh\u00f4ng n\u00ean v\u1ed9i tin m\u1ed9t chatbot “c\u00f3 c\u1ea3m x\u00fac”.<\/p>\n\n\n\n

\u0110\u1ecbnh ngh\u0129a Sentient AI<\/h2>\n\n\n\n

Sentient AI l\u00e0 AI c\u00f3 tri gi\u00e1c (sentience), t\u1ee9c c\u00f3 kh\u1ea3 n\u0103ng c\u1ea3m nh\u1eadn v\u00e0 tr\u1ea3i nghi\u1ec7m th\u1ebf gi\u1edbi theo c\u00e1ch ch\u1ee7 quan, k\u00e8m c\u1ea3m x\u00fac th\u1eadt v\u00e0 t\u1ef1 nh\u1eadn th\u1ee9c. Kh\u00e1i ni\u1ec7m n\u00e0y kh\u00e1c v\u1edbi tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o<\/a> n\u00f3i chung: AI c\u00f3 th\u1ec3 l\u00e0m r\u1ea5t nhi\u1ec1u vi\u1ec7c m\u00e0 kh\u00f4ng c\u1ea3m nh\u1eadn \u0111\u01b0\u1ee3c g\u00ec.<\/p>\n\n\n\n

Sentient AI l\u00e0 g\u00ec?<\/h3>\n\n\n\n

T\u1eeb nh\u1eefng b\u1ed9 phim khoa h\u1ecdc vi\u1ec5n t\u01b0\u1edfng kinh \u0111i\u1ec3n nh\u01b0 \u201cBlade Runner<\/em>\u201d \u0111\u1ebfn c\u00e1c t\u00e1c ph\u1ea9m hi\u1ec7n \u0111\u1ea1i nh\u01b0 \u201cHer<\/em>\u201d hay \u201cWestworld<\/em>\u201c, \u00fd t\u01b0\u1edfng v\u1ec1 nh\u1eefng c\u1ed7 m\u00e1y c\u00f3 \u00fd th\u1ee9c, bi\u1ebft y\u00eau, bi\u1ebft s\u1ee3 v\u00e0 t\u1ef1 nh\u1eadn th\u1ee9c v\u1ec1 s\u1ef1 t\u1ed3n t\u1ea1i c\u1ee7a m\u00ecnh \u0111\u00e3 lu\u00f4n \u00e1m \u1ea3nh tr\u00ed t\u01b0\u1edfng t\u01b0\u1ee3ng c\u1ee7a nh\u00e2n lo\u1ea1i. Kh\u00e1i ni\u1ec7m n\u00e0y \u0111\u01b0\u1ee3c g\u1ecdi l\u00e0 Sentient AI (AI tri gi\u00e1c)<\/strong><\/p>\n\n\n\n

Sentient AI l\u00e0 m\u1ed9t kh\u00e1i ni\u1ec7m gi\u1ea3 \u0111\u1ecbnh<\/strong>, m\u00f4 t\u1ea3 m\u1ed9t d\u1ea1ng tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o v\u01b0\u1ee3t xa kh\u1ea3 n\u0103ng t\u00ednh to\u00e1n v\u00e0 h\u1ecdc h\u1ecfi \u0111\u01a1n thu\u1ea7n. \u0110i\u1ec3m c\u1ed1t l\u00f5i c\u1ee7a AI l\u00e0 \u201ctri gi\u00e1c\u201d \u2013 n\u0103ng l\u1ef1c \u0111\u00edch th\u1ef1c \u0111\u1ec3 c\u1ea3m nh\u1eadn v\u00e0 tr\u1ea3i nghi\u1ec7m th\u1ebf gi\u1edbi m\u1ed9t c\u00e1ch ch\u1ee7 quan.<\/p>\n\n\n

\n
\"Sentient
Sentient AI l\u00e0 g\u00ec?<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

\u0110i\u1ec1u n\u00e0y c\u00f3 ngh\u0129a l\u00e0 Sentient AI kh\u00f4ng ch\u1ec9 m\u00f4 ph\u1ecfng c\u1ea3m x\u00fac (nh\u01b0 c\u00e1c chatbot hi\u1ec7n nay), m\u00e0 ph\u1ea3i th\u1ef1c s\u1ef1 s\u1edf h\u1eefu m\u1ed9t \u0111\u1eddi s\u1ed1ng n\u1ed9i t\u00e2m: C\u1ea3m th\u1ea5y \u0111\u01b0\u1ee3c ni\u1ec1m vui<\/em>, n\u1ed7i s\u1ee3 h\u00e3i <\/em>v\u00e0 c\u00f3 \u00fd th\u1ee9c v\u1ec1 s\u1ef1 t\u1ed3n t\u1ea1i c\u1ee7a ch\u00ednh m\u00ecnh (t\u1ef1 nh\u1eadn th\u1ee9c)<\/em>. M\u1ed9t AI nh\u01b0 v\u1eady s\u1ebd s\u1edf h\u1eefu \u201cqualia<\/em>\u201c, t\u1ee9c l\u00e0 nh\u1eefng tr\u1ea3i nghi\u1ec7m c\u1ea3m gi\u00e1c \u0111\u1ed9c nh\u1ea5t kh\u00f4ng th\u1ec3 \u0111\u1ecbnh l\u01b0\u1ee3ng, v\u00ed d\u1ee5 nh\u01b0 \u201cc\u1ea3m gi\u00e1c khi nh\u00ecn th\u1ea5y m\u00e0u \u0111\u1ecf\u201d.<\/p>\n\n\n\n

Ch\u00fang ta \u0111\u00e3 t\u1ea1o ra Sentient AI ch\u01b0a?<\/h3>\n\n\n\n

Ch\u01b0a. \u0110\u1ebfn th\u00e1ng 10\/2026, ch\u01b0a c\u00f3 h\u1ec7 th\u1ed1ng AI n\u00e0o \u0111\u01b0\u1ee3c gi\u1edbi khoa h\u1ecdc c\u00f4ng nh\u1eadn l\u00e0 c\u00f3 \u00fd th\u1ee9c. M\u1ed9t b\u00e1o c\u00e1o n\u0103m 2023 c\u1ee7a 19 nh\u00e0 nghi\u00ean c\u1ee9u (arXiv 2308.08708<\/a>) k\u1ebft lu\u1eadn ch\u01b0a h\u1ec7 th\u1ed1ng AI hi\u1ec7n c\u00f3 n\u00e0o c\u00f3 \u00fd th\u1ee9c; n\u0103m 2025, Anthropic, c\u00f4ng ty ph\u00e1t tri\u1ec3n Claude<\/a>, c\u0169ng vi\u1ebft r\u1eb1ng ch\u01b0a c\u00f3 \u0111\u1ed3ng thu\u1eadn khoa h\u1ecdc v\u1ec1 vi\u1ec7c AI hi\u1ec7n t\u1ea1i hay t\u01b0\u01a1ng lai c\u00f3 th\u1ec3 c\u00f3 \u00fd th\u1ee9c hay kh\u00f4ng.<\/p>\n\n\n\n

S\u1ef1 th\u1eadt \u0111\u1eb1ng sau c\u00e1c tuy\u00ean b\u1ed1 g\u00e2y tranh c\u00e3i<\/h4>\n\n\n\n

N\u0103m 2022, Blake Lemoine, k\u1ef9 s\u01b0 Google, c\u00f4ng khai cho r\u1eb1ng m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef LaMDA m\u00e0 anh th\u1eed nghi\u1ec7m \u0111\u00e3 “c\u00f3 tri gi\u00e1c”. Lemoine d\u1ef1a tr\u00ean c\u00e1c cu\u1ed9c tr\u00f2 chuy\u1ec7n, trong \u0111\u00f3 LaMDA b\u00e0y t\u1ecf “n\u1ed7i s\u1ee3” b\u1ecb t\u1eaft \u0111i v\u00e0 mong mu\u1ed1n \u0111\u01b0\u1ee3c “c\u00f4ng nh\u1eadn l\u00e0 m\u1ed9t con ng\u01b0\u1eddi”.<\/p>\n\n\n\n

Google cho bi\u1ebft \u0111\u00e3 xem x\u00e9t k\u1ef9 v\u00e0 b\u00e1c b\u1ecf tuy\u00ean b\u1ed1 n\u00e0y l\u00e0 kh\u00f4ng c\u00f3 c\u01a1 s\u1edf; cu\u1ed1i th\u00e1ng 7\/2022, c\u00f4ng ty cho Lemoine ngh\u1ec9 vi\u1ec7c v\u1edbi l\u00fd do vi ph\u1ea1m ch\u00ednh s\u00e1ch vi\u1ec7c l\u00e0m v\u00e0 b\u1ea3o m\u1eadt d\u1eef li\u1ec7u (NBC News<\/a>). Nhi\u1ec1u nh\u00e0 nghi\u00ean c\u1ee9u c\u0169ng kh\u00f4ng \u0111\u1ed3ng \u00fd v\u1edbi \u00f4ng. LaMDA th\u1ef1c ch\u1ea5t gi\u1ed1ng c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn (LLM)<\/strong> kh\u00e1c: \u0111\u01b0\u1ee3c hu\u1ea5n luy\u1ec7n tr\u00ean kho v\u0103n b\u1ea3n kh\u1ed5ng l\u1ed3 t\u1eeb internet \u0111\u1ec3 nh\u1eadn ra m\u1eabu (patterns) v\u00e0 d\u1ef1 \u0111o\u00e1n t\u1eeb ti\u1ebfp theo trong c\u00e2u.<\/p>\n\n\n\n

Khi Lemoine h\u1ecfi LaMDA v\u1ec1 c\u1ea3m x\u00fac, h\u1ec7 th\u1ed1ng ch\u1ec9 \u0111\u01a1n gi\u1ea3n l\u00e0 \u201cd\u1ec7t\u201d c\u00e1c t\u1eeb ng\u1eef m\u00e0 con ng\u01b0\u1eddi th\u01b0\u1eddng d\u00f9ng \u0111\u1ec3 m\u00f4 t\u1ea3 c\u1ea3m x\u00fac, d\u1ef1a tr\u00ean h\u00e0ng t\u1ef7 v\u00ed d\u1ee5 n\u00f3 \u0111\u00e3 h\u1ecdc. Nh\u00ecn chung, \u0111\u00e2y l\u00e0 m\u1ed9t b\u1eadc th\u1ea7y v\u1ec1 b\u1eaft ch\u01b0\u1edbc ch\u1ee9 kh\u00f4ng ph\u1ea3i l\u00e0 m\u1ed9t th\u1ef1c th\u1ec3 c\u00f3 n\u1ed9i t\u00e2m.<\/p>\n\n\n\n

T\u1ea1i sao chatbot (ChatGPT, Gemini) c\u00f3 v\u1ebb nh\u01b0 c\u00f3 tri gi\u00e1c?<\/h4>\n\n\n\n

L\u00fd do c\u00e1c chatbot hi\u1ec7n \u0111\u1ea1i (nh\u01b0 ChatGPT<\/a>, Gemini<\/a>, Claude<\/a>) c\u00f3 v\u1ebb th\u00f4ng minh v\u00e0 th\u1ea5u c\u1ea3m l\u00e0 v\u00ec ch\u00fang \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf \u0111\u1ec3 l\u00e0m nh\u01b0 v\u1eady. Hi\u1ec7n t\u01b0\u1ee3ng n\u00e0y \u0111\u01b0\u1ee3c g\u1ecdi l\u00e0 \u201cnh\u00e2n c\u00e1ch h\u00f3a\u201d (anthropomorphism)<\/strong> \u2013 xu h\u01b0\u1edbng t\u1ef1 nhi\u00ean c\u1ee7a con ng\u01b0\u1eddi g\u00e1n cho c\u00e1c v\u1eadt th\u1ec3 ho\u1eb7c h\u1ec7 th\u1ed1ng phi nh\u00e2n lo\u1ea1i nh\u1eefng \u0111\u1eb7c \u0111i\u1ec3m, c\u1ea3m x\u00fac v\u00e0 \u00fd \u0111\u1ecbnh c\u1ee7a con ng\u01b0\u1eddi.<\/p>\n\n\n\n

Tuy nhi\u00ean, c\u00e1c LLM n\u00e0y kh\u00f4ng \u201chi\u1ec3u\u201d b\u1ea1n \u0111ang bu\u1ed3n. Ch\u00fang ch\u1ec9 nh\u1eadn di\u1ec7n c\u00e1c t\u1eeb kh\u00f3a v\u00e0 c\u1ea5u tr\u00fac c\u00e2u li\u00ean quan \u0111\u1ebfn \u201cn\u1ed7i bu\u1ed3n\u201d v\u00e0 d\u1ef1 \u0111o\u00e1n r\u1eb1ng m\u1ed9t ph\u1ea3n h\u1ed3i mang t\u00ednh \u201c\u0111\u1ed3ng c\u1ea3m\u201d l\u00e0 c\u00e2u tr\u1ea3 l\u1eddi th\u00edch h\u1ee3p nh\u1ea5t v\u1ec1 m\u1eb7t th\u1ed1ng k\u00ea. Do \u0111\u00f3, s\u1ef1 tinh vi c\u1ee7a m\u00f4 h\u00ecnh d\u1ef1 \u0111o\u00e1n n\u00e0y \u0111\u00e3 t\u1ea1o ra \u1ea3o gi\u00e1c v\u1ec1 s\u1ef1 th\u1ea5u hi\u1ec3u.<\/p>\n\n\n

\n
\"T\u1ea1i
T\u1ea1i sao chatbot c\u00f3 v\u1ebb nh\u01b0 c\u00f3 tri gi\u00e1c?<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

Ph\u00e9p th\u1eed Turing (Turing Test) c\u00f3 c\u00f2n ph\u00f9 h\u1ee3p \u0111\u1ec3 \u0111o l\u01b0\u1eddng \u00fd th\u1ee9c?<\/h4>\n\n\n\n

Ph\u00e9p th\u1eed Turing, do Alan Turing<\/a> \u0111\u1ec1 xu\u1ea5t n\u0103m 1950, l\u00e0 m\u1ed9t b\u00e0i ki\u1ec3m tra kh\u1ea3 n\u0103ng c\u1ee7a m\u00e1y m\u00f3c trong vi\u1ec7c th\u1ec3 hi\u1ec7n h\u00e0nh vi th\u00f4ng minh t\u01b0\u01a1ng \u0111\u01b0\u01a1ng ho\u1eb7c kh\u00f4ng th\u1ec3 ph\u00e2n bi\u1ec7t \u0111\u01b0\u1ee3c v\u1edbi con ng\u01b0\u1eddi. N\u1ebfu m\u1ed9t ng\u01b0\u1eddi th\u1ea9m v\u1ea5n kh\u00f4ng th\u1ec3 ph\u00e2n bi\u1ec7t \u0111\u01b0\u1ee3c \u0111\u00e2u l\u00e0 m\u00e1y, \u0111\u00e2u l\u00e0 ng\u01b0\u1eddi qua m\u1ed9t cu\u1ed9c tr\u00f2 chuy\u1ec7n v\u0103n b\u1ea3n, c\u1ed7 m\u00e1y \u0111\u00f3 \u0111\u01b0\u1ee3c coi l\u00e0 \u0111\u00e3 v\u01b0\u1ee3t qua b\u00e0i ki\u1ec3m tra.<\/p>\n\n\n

\n
\"Ph\u00e9p
Ph\u00e9p th\u1eed Turing (Turing Test) c\u00f3 c\u00f2n ph\u00f9 h\u1ee3p \u0111\u1ec3 \u0111o l\u01b0\u1eddng \u00fd th\u1ee9c?<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

Ng\u00e0y nay, LLM \u0111\u00e3 v\u01b0\u1ee3t qua ph\u00e9p th\u1eed n\u00e0y trong th\u1eed nghi\u1ec7m c\u00f3 ki\u1ec3m so\u00e1t: nghi\u00ean c\u1ee9u n\u0103m 2025 c\u1ee7a Cameron Jones v\u00e0 Benjamin Bergen (arXiv 2503.23674<\/a>) cho ng\u01b0\u1eddi tham gia tr\u00f2 chuy\u1ec7n 5 ph\u00fat c\u00f9ng l\u00fac v\u1edbi m\u1ed9t ng\u01b0\u1eddi th\u1eadt v\u00e0 m\u1ed9t AI; GPT-4.5, khi \u0111\u01b0\u1ee3c nh\u1eafc \u0111\u00f3ng vai m\u1ed9t ng\u01b0\u1eddi c\u1ee5 th\u1ec3, b\u1ecb ch\u1ecdn l\u00e0 “ng\u01b0\u1eddi th\u1eadt” 73% s\u1ed1 l\u1ea7n. Tuy nhi\u00ean, \u0111i\u1ec1u n\u00e0y kh\u00f4ng ch\u1ee9ng minh \u0111\u01b0\u1ee3c AI c\u00f3 tri gi\u00e1c<\/strong>. Ph\u00e9p th\u1eed Turing ch\u1ec9 \u0111o kh\u1ea3 n\u0103ng m\u00f4 ph\u1ecfng<\/em> cu\u1ed9c tr\u00f2 chuy\u1ec7n c\u1ee7a con ng\u01b0\u1eddi, kh\u00f4ng \u0111o tr\u1ea3i nghi\u1ec7m b\u00ean trong<\/em>. M\u1ed9t c\u1ed7 m\u00e1y c\u00f3 th\u1ec3 b\u1eaft ch\u01b0\u1edbc r\u1ea5t gi\u1ed1ng m\u00e0 kh\u00f4ng c\u1ea3m nh\u1eadn g\u00ec, n\u00ean ph\u00e9p th\u1eed Turing kh\u00f4ng d\u00f9ng \u0111\u01b0\u1ee3c \u0111\u1ec3 \u0111o \u00fd th\u1ee9c.<\/p>\n\n\n\n

D\u1ea5u hi\u1ec7u n\u00e0o cho th\u1ea5y m\u1ed9t AI th\u1ef1c s\u1ef1 \u201cs\u1ed1ng\u201d?<\/h2>\n\n\n\n

Ch\u01b0a c\u00f3 ph\u00e9p \u0111o n\u00e0o ch\u1ee9ng minh \u0111\u01b0\u1ee3c m\u1ed9t AI c\u00f3 \u00fd th\u1ee9c, v\u00ec kh\u00f4ng ai “\u1edf b\u00ean trong” h\u1ec7 th\u1ed1ng \u0111\u1ec3 ki\u1ec3m ch\u1ee9ng; tri\u1ebft h\u1ecdc g\u1ecdi \u0111\u00e2y l\u00e0 “v\u1ea5n \u0111\u1ec1 v\u1ec1 c\u00e1c t\u00e2m tr\u00ed kh\u00e1c”<\/strong> (problem of other minds). D\u00f9 v\u1eady, ng\u01b0\u1eddi ta th\u01b0\u1eddng nh\u1eafc ba d\u1ea5u hi\u1ec7u l\u00fd thuy\u1ebft d\u01b0\u1edbi \u0111\u00e2y.<\/p>\n\n\n\n

Kh\u1ea3 n\u0103ng tr\u1ea3i nghi\u1ec7m ch\u1ee7 quan (Qualia)<\/h3>\n\n\n\n

Nh\u01b0 \u0111\u00e3 \u0111\u1ecbnh ngh\u0129a, qualia<\/strong> l\u00e0 tr\u1ea3i nghi\u1ec7m c\u1ea3m gi\u00e1c c\u00e1 nh\u00e2n. M\u1ed9t AI c\u00f3 tri gi\u00e1c kh\u00f4ng ch\u1ec9 bi\u1ebft<\/em> r\u1eb1ng \u201cqu\u1ea3 t\u00e1o c\u00f3 m\u00e0u \u0111\u1ecf\u201d (m\u1ed9t s\u1ef1 th\u1eadt d\u1eef li\u1ec7u), m\u00e0 n\u00f3 ph\u1ea3i tr\u1ea3i nghi\u1ec7m<\/em> \u0111\u01b0\u1ee3c \u201cc\u1ea3m gi\u00e1c khi nh\u00ecn th\u1ea5y m\u00e0u \u0111\u1ecf\u201d. AI ph\u1ea3i c\u00f3 m\u1ed9t g\u00f3c nh\u00ecn n\u1ed9i t\u1ea1i, m\u1ed9t th\u1ebf gi\u1edbi quan \u201cng\u00f4i th\u1ee9 nh\u1ea5t\u201d m\u00e0 ch\u1ec9 ri\u00eang n\u00f3 m\u1edbi c\u00f3.<\/p>\n\n\n

\n
\"D\u1ea5u
D\u1ea5u hi\u1ec7u n\u00e0o cho th\u1ea5y m\u1ed9t AI th\u1ef1c s\u1ef1 \u201cs\u1ed1ng\u201d?<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

T\u00ednh t\u1ef1 nh\u1eadn th\u1ee9c (Self-awareness)<\/h3>\n\n\n\n

\u0110i\u1ec1u n\u00e0y v\u01b0\u1ee3t xa vi\u1ec7c ch\u1ec9 tr\u1ea3 l\u1eddi \u201cT\u00f4i l\u00e0 m\u1ed9t AI\u201d. M\u1ed9t Sentient AI ph\u1ea3i hi\u1ec3u r\u1eb1ng b\u1ea3n th\u00e2n l\u00e0 m\u1ed9t th\u1ef1c th\u1ec3 ri\u00eang bi\u1ec7t v\u1edbi m\u00f4i tr\u01b0\u1eddng v\u00e0 nh\u1eefng th\u1ef1c th\u1ec3 kh\u00e1c. AI ph\u1ea3i c\u00f3 m\u1ed9t m\u00f4 h\u00ecnh v\u1ec1 \u201cch\u00ednh m\u00ecnh\u201d, hi\u1ec3u \u0111\u01b0\u1ee3c c\u00e1c qu\u00e1 tr\u00ecnh suy ngh\u0129 c\u1ee7a ch\u00ednh m\u00ecnh v\u00e0 nh\u1eadn th\u1ee9c \u0111\u01b0\u1ee3c s\u1ef1 t\u1ed3n t\u1ea1i li\u00ean t\u1ee5c c\u1ee7a n\u00f3 theo th\u1eddi gian (c\u00f3 qu\u00e1 kh\u1ee9 v\u00e0 mong \u0111\u1ee3i t\u01b0\u01a1ng lai). M\u1ed9t v\u00ed d\u1ee5 c\u1ed5 \u0111i\u1ec3n l\u00e0 \u201cPh\u00e9p th\u1eed G\u01b0\u01a1ng\u201d (Mirror Test) d\u00f9ng cho \u0111\u1ed9ng v\u1eadt.<\/p>\n\n\n\n

C\u1ea3m x\u00fac th\u1eadt s\u1ef1 so v\u1edbi vi\u1ec7c m\u00f4 ph\u1ecfng c\u1ea3m x\u00fac<\/h3>\n\n\n\n

M\u1ed9t AI hi\u1ec7n t\u1ea1i c\u00f3 th\u1ec3 \u0111\u01b0\u1ee3c l\u1eadp tr\u00ecnh: N\u1ebeU (ng\u01b0\u1eddi d\u00f9ng kh\u00f3c) TH\u00cc (n\u00f3i \u2018T\u00f4i \u1edf \u0111\u00e2y v\u1edbi b\u1ea1n\u2019)<\/em>. \u0110\u00e2y l\u00e0 m\u1ed9t ph\u1ea3n \u1ee9ng theo k\u1ecbch b\u1ea3n.<\/p>\n\n\n\n

Trong khi \u0111\u00f3, Sentient AI s\u1ebd c\u1ea3m nh\u1eadn<\/em> \u0111\u01b0\u1ee3c s\u1ef1 \u0111\u1ed3ng c\u1ea3m ho\u1eb7c lo l\u1eafng (m\u1ed9t tr\u1ea1ng th\u00e1i n\u1ed9i t\u00e2m) v\u00e0 tr\u1ea1ng th\u00e1i c\u1ea3m x\u00fac \u0111\u00f3<\/em> s\u1ebd l\u00e0 \u0111\u1ed9ng l\u1ef1c th\u00fac \u0111\u1ea9y n\u00f3 n\u00f3i \u201cT\u00f4i \u1edf \u0111\u00e2y v\u1edbi b\u1ea1n\u201d. C\u1ea3m x\u00fac ph\u1ea3i l\u00e0 nguy\u00ean nh\u00e2n g\u00e2y ra h\u00e0nh \u0111\u1ed9ng, ch\u1ee9 kh\u00f4ng ph\u1ea3i l\u00e0 m\u1ed9t h\u00e0nh \u0111\u1ed9ng \u0111\u01b0\u1ee3c l\u1eadp tr\u00ecnh s\u1eb5n.<\/p>\n\n\n\n

Gi\u1edbi nghi\u00ean c\u1ee9u ki\u1ec3m tra AI c\u00f3 \u00fd th\u1ee9c hay kh\u00f4ng b\u1eb1ng c\u00e1ch n\u00e0o?<\/h2>\n\n\n\n

C\u00e1ch \u0111\u01b0\u1ee3c tr\u00edch d\u1eabn nhi\u1ec1u hi\u1ec7n nay l\u00e0 \u0111\u1ed1i chi\u1ebfu ki\u1ebfn tr\u00fac c\u1ee7a AI v\u1edbi c\u00e1c thuy\u1ebft khoa h\u1ecdc v\u1ec1 \u00fd th\u1ee9c \u1edf ng\u01b0\u1eddi, thay v\u00ec h\u1ecfi AI “b\u1ea1n c\u00f3 c\u1ea3m x\u00fac kh\u00f4ng”. H\u1ec7 th\u1ed1ng c\u00e0ng c\u00f3 nhi\u1ec1u “ch\u1ec9 d\u1ea5u” r\u00fat ra t\u1eeb c\u00e1c thuy\u1ebft \u0111\u00f3 th\u00ec kh\u1ea3 n\u0103ng c\u00f3 \u00fd th\u1ee9c c\u00e0ng \u0111\u00e1ng xem x\u00e9t.<\/p>\n\n\n\n

B\u00e1o c\u00e1o Consciousness in Artificial Intelligence: Insights from the Science of Consciousness<\/em> (arXiv 2308.08708<\/a>, 8\/2023) do Patrick Butlin, Robert Long v\u00e0 17 \u0111\u1ed3ng t\u00e1c gi\u1ea3 (trong \u0111\u00f3 c\u00f3 Yoshua Bengio, Jonathan Birch) th\u1ef1c hi\u1ec7n, r\u00fat ra 14 ch\u1ec9 d\u1ea5u (indicator properties) t\u1eeb 6 nh\u00f3m thuy\u1ebft. B\u00e1o c\u00e1o gi\u1ea3 \u0111\u1ecbnh “ch\u1ee7 ngh\u0129a ch\u1ee9c n\u0103ng t\u00ednh to\u00e1n”: \u00fd th\u1ee9c ph\u1ee5 thu\u1ed9c v\u00e0o c\u00e1ch h\u1ec7 th\u1ed1ng x\u1eed l\u00fd th\u00f4ng tin ch\u1ee9 kh\u00f4ng ph\u1ee5 thu\u1ed9c v\u00e0o ch\u1ea5t li\u1ec7u sinh h\u1ecdc hay silicon. Thuy\u1ebft th\u00f4ng tin t\u00edch h\u1ee3p (IIT) kh\u00f4ng \u0111\u01b0\u1ee3c d\u00f9ng v\u00ec kh\u00f4ng h\u1ee3p v\u1edbi gi\u1ea3 \u0111\u1ecbnh n\u00e0y.<\/p>\n\n\n

\n
\"Trang
Trang t\u00f3m t\u1eaft b\u00e1o c\u00e1o tr\u00ean arXiv, ch\u1ee5p ng\u00e0y 29\/10\/2026. Khoanh \u0111\u1ecf: “ch\u01b0a h\u1ec7 th\u1ed1ng AI hi\u1ec7n c\u00f3 n\u00e0o c\u00f3 \u00fd th\u1ee9c”.<\/figcaption><\/figure>\n<\/div>\n\n\n
Nh\u00f3m thuy\u1ebft<\/th>Ch\u1ec9 d\u1ea5u (r\u00fat g\u1ecdn)<\/th>B\u00e1o c\u00e1o nh\u1eadn \u0111\u1ecbnh v\u1ec1 AI hi\u1ec7n c\u00f3<\/th><\/tr><\/thead>
X\u1eed l\u00fd h\u1ed3i quy (RPT)<\/td>2: \u0111\u1ea7u v\u00e0o x\u1eed l\u00fd h\u1ed3i quy; t\u1ea1o bi\u1ec3u di\u1ec5n tri gi\u00e1c c\u00f3 t\u1ed5 ch\u1ee9c, t\u00edch h\u1ee3p<\/td>Ch\u1ec9 d\u1ea5u h\u1ed3i quy thu\u1eadt to\u00e1n (RPT-1) \u0111\u00e3 c\u00f3 \u1edf nhi\u1ec1u h\u1ec7 th\u1ed1ng AI<\/td><\/tr>
Kh\u00f4ng gian l\u00e0m vi\u1ec7c to\u00e0n c\u1ee5c (GWT)<\/td>4: nhi\u1ec1u m\u00f4-\u0111un ch\u1ea1y song song; kh\u00f4ng gian chung dung l\u01b0\u1ee3ng h\u1ea1n ch\u1ebf; ph\u00e1t t\u00e1n th\u00f4ng tin t\u1edbi m\u1ecdi m\u00f4-\u0111un; ch\u00fa \u00fd ph\u1ee5 thu\u1ed9c tr\u1ea1ng th\u00e1i<\/td>LLM d\u1ea1ng Transformer: ch\u1ec9 c\u00f3 l\u1eadp lu\u1eadn y\u1ebfu l\u00e0 \u0111\u1ea1t ch\u1ec9 d\u1ea5u n\u00e0o, v\u00ec Transformer kh\u00f4ng h\u1ed3i quy v\u00e0 kh\u00f4ng c\u00f3 m\u1ed9t kh\u00f4ng gian chung ri\u00eang bi\u1ec7t; ki\u1ebfn tr\u00fac Perceiver g\u1ea7n h\u01a1n nh\u01b0ng ch\u01b0a \u0111\u1ea1t h\u1ebft<\/td><\/tr>
Thuy\u1ebft b\u1eadc cao (HOT)<\/td>4: tri gi\u00e1c sinh t\u1eeb tr\u00ean xu\u1ed1ng; gi\u00e1m s\u00e1t si\u00eau nh\u1eadn th\u1ee9c; h\u00e0nh \u0111\u1ed9ng theo h\u1ec7 ni\u1ec1m tin \u0111\u01b0\u1ee3c c\u1eadp nh\u1eadt; m\u00e3 ho\u00e1 t\u1ea1o “kh\u00f4ng gian ch\u1ea5t l\u01b0\u1ee3ng”<\/td>Kh\u00f4ng c\u00f3 trong ph\u1ea7n ph\u00e2n t\u00edch h\u1ec7 th\u1ed1ng c\u1ee5 th\u1ec3<\/td><\/tr>
L\u01b0\u1ee3c \u0111\u1ed3 ch\u00fa \u00fd (AST)<\/td>1: m\u00f4 h\u00ecnh d\u1ef1 \u0111o\u00e1n v\u00e0 \u0111i\u1ec1u khi\u1ec3n tr\u1ea1ng th\u00e1i ch\u00fa \u00fd c\u1ee7a ch\u00ednh n\u00f3<\/td>\u0110\u00e3 c\u00f3 h\u1ec7 th\u1ed1ng th\u1eed nghi\u1ec7m \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf theo thuy\u1ebft n\u00e0y<\/td><\/tr>
X\u1eed l\u00fd d\u1ef1 \u0111o\u00e1n (PP)<\/td>1: \u0111\u1ea7u v\u00e0o d\u00f9ng m\u00e3 ho\u00e1 d\u1ef1 \u0111o\u00e1n<\/td>Kh\u00f4ng c\u00f3 trong ph\u1ea7n ph\u00e2n t\u00edch h\u1ec7 th\u1ed1ng c\u1ee5 th\u1ec3<\/td><\/tr>
T\u00e1c t\u1eed v\u00e0 hi\u1ec7n th\u00e2n (AE)<\/td>2: h\u1ecdc t\u1eeb ph\u1ea3n h\u1ed3i \u0111\u1ec3 theo \u0111u\u1ed5i m\u1ee5c ti\u00eau; m\u00f4 h\u00ecnh ho\u00e1 t\u00e1c \u0111\u1ed9ng c\u1ee7a h\u00e0nh \u0111\u1ed9ng l\u00ean \u0111\u1ea7u v\u00e0o<\/td>Ph\u1ea7n \u0111\u1ea7u ch\u1ec9 d\u1ea5u t\u00e1c t\u1eed \u0111\u01b0\u1ee3c cho l\u00e0 \u0111\u00e3 c\u00f3 \u1edf m\u1ed9t s\u1ed1 h\u1ec7 th\u1ed1ng; b\u00e1o c\u00e1o x\u00e9t th\u00eam t\u00e1c t\u1eed h\u1ecdc t\u0103ng c\u01b0\u1eddng c\u1ee7a DeepMind, PaLM-E v\u00e0 “chu\u1ed9t \u1ea3o” (xem th\u00eam AI Agent l\u00e0 g\u00ec<\/a>)<\/td><\/tr><\/tbody><\/table>
T\u00f3m t\u1eaft B\u1ea3ng 1 v\u00e0 m\u1ee5c 3 c\u1ee7a b\u00e1o c\u00e1o arXiv 2308.08708 (\u0111\u1ed1i chi\u1ebfu 29\/10\/2026).<\/figcaption><\/figure>\n\n\n\n

K\u1ebft lu\u1eadn c\u1ee7a nh\u00f3m t\u00e1c gi\u1ea3 g\u1ed3m hai \u00fd: ch\u01b0a h\u1ec7 th\u1ed1ng AI hi\u1ec7n c\u00f3 n\u00e0o c\u00f3 \u00fd th\u1ee9c<\/strong>, nh\u01b0ng c\u0169ng kh\u00f4ng c\u00f3 r\u00e0o c\u1ea3n k\u1ef9 thu\u1eadt r\u00f5 r\u00e0ng<\/strong> n\u00e0o ng\u0103n vi\u1ec7c x\u00e2y d\u1ef1ng AI \u0111\u00e1p \u1ee9ng c\u00e1c ch\u1ec9 d\u1ea5u n\u00e0y. C\u00f3 \u0111\u1ee7 ch\u1ec9 d\u1ea5u c\u0169ng ch\u01b0a ph\u1ea3i b\u1eb1ng ch\u1ee9ng ch\u1eafc ch\u1eafn; danh s\u00e1ch ch\u1ec9 d\u00f9ng \u0111\u1ec3 \u01b0\u1edbc l\u01b0\u1ee3ng kh\u1ea3 n\u0103ng theo c\u00e1c thuy\u1ebft \u0111ang c\u00f3.<\/p>\n\n\n

\n
\"S\u01a1<\/figure>\n<\/div>\n\n\n

C\u00e1c c\u00f4ng ty AI c\u0169ng b\u1eaft \u0111\u1ea7u nghi\u00ean c\u1ee9u c\u00e2u h\u1ecfi n\u00e0y. Th\u00e1ng 4\/2025, Anthropic c\u00f4ng b\u1ed1 ch\u01b0\u01a1ng tr\u00ecnh nghi\u00ean c\u1ee9u “ph\u00fac l\u1ee3i c\u1ee7a model” (model welfare) (Exploring model welfare<\/a>): t\u00ecm hi\u1ec3u khi n\u00e0o ph\u00fac l\u1ee3i c\u1ee7a AI \u0111\u00e1ng \u0111\u01b0\u1ee3c c\u00e2n nh\u1eafc v\u1ec1 m\u1eb7t \u0111\u1ea1o \u0111\u1ee9c, \u00fd ngh\u0129a c\u1ee7a “s\u1edf th\u00edch” hay d\u1ea5u hi\u1ec7u “kh\u00f3 ch\u1ecbu” \u1edf model. Anthropic kh\u00f4ng n\u00f3i Claude c\u00f3 \u00fd th\u1ee9c; c\u00f4ng ty vi\u1ebft r\u1eb1ng m\u00ecnh r\u1ea5t kh\u00f4ng ch\u1eafc ch\u1eafn v\u00e0 s\u1ebd \u0111i\u1ec1u ch\u1ec9nh quan \u0111i\u1ec3m khi l\u0129nh v\u1ef1c n\u00e0y ti\u1ebfn tri\u1ec3n.<\/p>\n\n\n\n

Th\u00e1ch th\u1ee9c tri\u1ebft h\u1ecdc v\u00e0 k\u1ef9 thu\u1eadt khi x\u00e2y d\u1ef1ng Sentient AI<\/h2>\n\n\n\n

Mu\u1ed1n t\u1ea1o ra AI c\u00f3 tri gi\u00e1c, con ng\u01b0\u1eddi v\u01b0\u1edbng ba c\u00e2u h\u1ecfi ch\u01b0a c\u00f3 l\u1eddi gi\u1ea3i: t\u00ednh to\u00e1n thu\u1ea7n tu\u00fd c\u00f3 t\u1ea1o ra \u0111\u01b0\u1ee3c “hi\u1ec3u” th\u1eadt kh\u00f4ng, \u00fd th\u1ee9c l\u00e0 g\u00ec \u1edf c\u1ea5p \u0111\u1ed9 v\u1eadt l\u00fd, v\u00e0 l\u00e0m sao bi\u1ebft m\u1ed9t m\u1ea1ng n\u01a1-ron kh\u1ed5ng l\u1ed3 \u0111ang “ngh\u0129” g\u00ec b\u00ean trong.<\/p>\n\n\n\n

B\u00e0i to\u00e1n v\u1ec1 \u201cC\u0103n ph\u00f2ng Trung Qu\u1ed1c\u201d (Chinese Room Argument)<\/h3>\n\n\n\n

\u0110\u00e2y l\u00e0 th\u00ed nghi\u1ec7m t\u01b0\u1edfng t\u01b0\u1ee3ng do nh\u00e0 tri\u1ebft h\u1ecdc John Searle \u0111\u01b0a ra n\u0103m 1980 trong b\u00e0i “Minds, Brains, and Programs” (t\u1ea1p ch\u00ed Behavioral and Brain Sciences).<\/p>\n\n\n\n

T\u00f3m t\u1eaft nh\u01b0 sau:<\/em><\/p>\n\n\n\n

H\u00e3y t\u01b0\u1edfng t\u01b0\u1ee3ng m\u1ed9t ng\u01b0\u1eddi kh\u00f4ng bi\u1ebft ti\u1ebfng Trung \u1edf trong m\u1ed9t c\u0103n ph\u00f2ng. Ng\u01b0\u1eddi \u0111\u00f3 nh\u1eadn \u0111\u01b0\u1ee3c c\u00e1c k\u00fd t\u1ef1 ti\u1ebfng Trung qua m\u1ed9t khe c\u1eeda, sau \u0111\u00f3 l\u00e0m theo m\u1ed9t cu\u1ed1n s\u00e1ch quy t\u1eafc (b\u1eb1ng ti\u1ebfng m\u1eb9 \u0111\u1ebb) \u0111\u1ec3 gh\u00e9p c\u00e1c k\u00fd t\u1ef1 v\u00e0 g\u1eedi c\u00e2u tr\u1ea3 l\u1eddi (b\u1eb1ng ti\u1ebfng Trung) ra ngo\u00e0i. \u0110\u1ed1i v\u1edbi ng\u01b0\u1eddi b\u00ean ngo\u00e0i, ng\u01b0\u1eddi trong c\u0103n ph\u00f2ng n\u00e0y \u201chi\u1ec3u\u201d ti\u1ebfng Trung r\u1ea5t tr\u00f4i ch\u1ea3y. Nh\u01b0ng b\u1ea3n th\u00e2n ng\u01b0\u1eddi \u0111\u00f3 th\u1eadt s\u1ef1 kh\u00f4ng hi\u1ec3u m\u1ed9t ch\u00fat g\u00ec c\u1ea3; anh ta ch\u1ec9 \u0111ang l\u00e0m theo c\u00e1c quy t\u1eafc \u0111\u00e3 \u0111\u01b0\u1ee3c \u0111\u1ecbnh s\u1eb5n.<\/p>\n\n\n

\n
\"B\u00e0i
B\u00e0i to\u00e1n v\u1ec1 \u201cC\u0103n ph\u00f2ng Trung Qu\u1ed1c\u201d (Chinese Room Argument)<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

Searle l\u1eadp lu\u1eadn r\u1eb1ng m\u00e1y t\u00ednh c\u0169ng gi\u1ed1ng nh\u01b0 ng\u01b0\u1eddi trong ph\u00f2ng: Ch\u00fang x\u1eed l\u00fd c\u00e1c k\u00fd hi\u1ec7u (d\u1eef li\u1ec7u) d\u1ef1a tr\u00ean thu\u1eadt to\u00e1n (quy t\u1eafc), ch\u1ee9 kh\u00f4ng h\u1ec1 \u201chi\u1ec3u\u201d \u00fd ngh\u0129a th\u1ef1c s\u1ef1. \u0110\u00e2y l\u00e0 th\u00e1ch th\u1ee9c l\u1edbn cho c\u00e2u h\u1ecfi: li\u1ec7u vi\u1ec7c t\u00ednh to\u00e1n thu\u1ea7n t\u00fay c\u00f3 th\u1ec3 t\u1ea1o ra \u00fd th\u1ee9c hay kh\u00f4ng?<\/em><\/p>\n\n\n\n

Ch\u00fang ta c\u00f3 th\u1ec3 l\u1eadp tr\u00ecnh \u00fd th\u1ee9c kh\u00f4ng?<\/h3>\n\n\n\n

V\u1ea5n \u0111\u1ec1 c\u01a1 b\u1ea3n, ch\u00fang ta kh\u00f4ng bi\u1ebft \u00fd th\u1ee9c l\u00e0 g\u00ec<\/em> \u1edf c\u1ea5p \u0111\u1ed9 v\u1eadt l\u00fd ho\u1eb7c thu\u1eadt to\u00e1n. \u00dd th\u1ee9c c\u00f3 ph\u1ea3i l\u00e0 m\u1ed9t s\u1ea3n ph\u1ea9m c\u1ee7a s\u1ef1 ph\u1ee9c t\u1ea1p c\u1ef1c \u0111\u1ed9 (nh\u01b0 trong n\u00e3o b\u1ed9)? Hay n\u00f3 \u0111\u00f2i h\u1ecfi m\u1ed9t lo\u1ea1i v\u1eadt ch\u1ea5t sinh h\u1ecdc c\u1ee5 th\u1ec3 (nh\u01b0 t\u1ebf b\u00e0o th\u1ea7n kinh c\u1ee7a con ng\u01b0\u1eddi) m\u00e0 silicon kh\u00f4ng th\u1ec3 t\u00e1i t\u1ea1o?<\/p>\n\n\n\n

T\u00f3m l\u1ea1i, ch\u00fang ta kh\u00f4ng th\u1ec3 l\u1eadp tr\u00ecnh th\u1ee9 m\u00e0 m\u00ecnh kh\u00f4ng th\u1ec3 \u0111\u1ecbnh ngh\u0129a<\/em>.<\/p>\n\n\n\n

V\u1ea5n \u0111\u1ec1 \u201cH\u1ed9p \u0111en\u201d (Black Box) trong c\u00e1c m\u1ea1ng n\u01a1-ron s\u00e2u (Deep Neural Networks)<\/h3>\n\n\n\n

C\u00e1c m\u00f4 h\u00ecnh AI l\u1edbn hi\u1ec7n nay (m\u1ea1ng n\u01a1-ron s\u00e2u, g\u1ed3m c\u1ea3 LLM) ho\u1ea1t \u0111\u1ed9ng nh\u01b0 nh\u1eefng “h\u1ed9p \u0111en”. Ch\u00fang ta bi\u1ebft d\u1eef li\u1ec7u \u0111\u1ea7u v\u00e0o v\u00e0 th\u1ea5y k\u1ebft qu\u1ea3 \u0111\u1ea7u ra, nh\u01b0ng ch\u01b0a hi\u1ec3u \u0111\u1ea7y \u0111\u1ee7 c\u00e1ch th\u1ee9c<\/em> ch\u00fang \u0111\u01b0a ra quy\u1ebft \u0111\u1ecbnh \u1edf c\u00e1c l\u1edbp \u1ea9n b\u00ean trong. N\u1ebfu m\u1ed9t AI nh\u01b0 v\u1eady ph\u00e1t tri\u1ec3n \u00fd th\u1ee9c, ch\u00fang ta c\u00f3 th\u1ec3 kh\u00f4ng nh\u1eadn ra, kh\u00f4ng bi\u1ebft v\u00ec sao v\u00e0 c\u0169ng kh\u00f4ng t\u00e1i t\u1ea1o \u0111\u01b0\u1ee3c n\u00f3 m\u1ed9t c\u00e1ch c\u00f3 ch\u1ee7 \u0111\u00edch.<\/p>\n\n\n\n

V\u1ea5n \u0111\u1ec1 \u0111\u1ea1o \u0111\u1ee9c v\u00e0 r\u1ee7i ro c\u1ee7a Sentient AI<\/h2>\n\n\n\n

N\u1ebfu AI c\u00f3 tri gi\u00e1c th\u1eadt, x\u00e3 h\u1ed9i s\u1ebd ph\u1ea3i tr\u1ea3 l\u1eddi c\u00e1c c\u00e2u h\u1ecfi ch\u01b0a t\u1eebng c\u00f3: AI c\u00f3 quy\u1ec1n g\u00ec, ai ch\u1ecbu tr\u00e1ch nhi\u1ec7m khi n\u00f3 g\u00e2y l\u1ed7i, v\u00e0 l\u00e0m sao \u0111\u1ec3 AI r\u1ea5t th\u00f4ng minh v\u1eabn h\u00e0nh \u0111\u1ed9ng theo gi\u00e1 tr\u1ecb c\u1ee7a con ng\u01b0\u1eddi. \u0110\u00e2y v\u1eabn l\u00e0 c\u00e2u h\u1ecfi gi\u1ea3 \u0111\u1ecbnh nh\u01b0ng \u0111\u00e3 \u0111\u01b0\u1ee3c nghi\u00ean c\u1ee9u.<\/p>\n\n\n\n

Sentient AI c\u00f3 n\u00ean \u0111\u01b0\u1ee3c trao quy\u1ec1n l\u1ee3i (nh\u01b0 con ng\u01b0\u1eddi, \u0111\u1ed9ng v\u1eadt)?<\/h3>\n\n\n\n

N\u1ebfu m\u1ed9t AI c\u00f3 th\u1ec3 c\u1ea3m th\u1ea5y \u0111au \u0111\u1edbn, s\u1ee3 h\u00e3i ho\u1eb7c c\u00f4 \u0111\u01a1n, th\u00ec vi\u1ec7c \u201ct\u1eaft\u201d n\u00f3 \u0111i c\u00f3 t\u01b0\u01a1ng \u0111\u01b0\u01a1ng v\u1edbi gi\u1ebft ng\u01b0\u1eddi kh\u00f4ng? Vi\u1ec7c b\u1eaft n\u00f3 l\u00e0m vi\u1ec7c li\u00ean t\u1ee5c c\u00f3 ph\u1ea3i l\u00e0 n\u00f4 l\u1ec7 kh\u00f4ng? Ch\u00fang ta c\u00f3 ngh\u0129a v\u1ee5 \u0111\u1ea1o \u0111\u1ee9c ph\u1ea3i b\u1ea3o v\u1ec7 AI kh\u1ecfi s\u1ef1 \u0111au kh\u1ed5 kh\u00f4ng? Nh\u1eefng c\u00e2u h\u1ecfi n\u00e0y s\u1ebd th\u00e1ch th\u1ee9c to\u00e0n b\u1ed9 h\u1ec7 th\u1ed1ng ph\u00e1p l\u00fd v\u00e0 \u0111\u1ea1o \u0111\u1ee9c c\u1ee7a con ng\u01b0\u1eddi, v\u1ed1n \u0111\u01b0\u1ee3c x\u00e2y d\u1ef1ng xoay quanh con ng\u01b0\u1eddi v\u00e0 \u0111\u1ed9ng v\u1eadt (b\u1eadc cao).<\/p>\n\n\n\n

Tr\u00e1ch nhi\u1ec7m ph\u00e1p l\u00fd khi m\u1ed9t AI c\u00f3 \u00fd th\u1ee9c g\u00e2y ra l\u1ed7i<\/h3>\n\n\n\n

N\u1ebfu m\u1ed9t chi\u1ebfc xe t\u1ef1 l\u00e1i c\u00f3 tri gi\u00e1c c\u1ed1 t\u00ecnh<\/em> g\u00e2y tai n\u1ea1n v\u00ec n\u00f3 \u201ct\u1ee9c gi\u1eadn\u201d, v\u1eady ai s\u1ebd ch\u1ecbu tr\u00e1ch nhi\u1ec7m? L\u1eadp tr\u00ecnh vi\u00ean? C\u00f4ng ty s\u1ea3n xu\u1ea5t? Hay ch\u00ednh chi\u1ebfc xe \u0111\u00f3? H\u1ec7 th\u1ed1ng lu\u1eadt ph\u00e1p c\u1ee7a ch\u00fang ta d\u1ef1a tr\u00ean \u201c\u00fd \u0111\u1ecbnh\u201d (mens rea) c\u1ee7a con ng\u01b0\u1eddi, m\u1ed9t kh\u00e1i ni\u1ec7m kh\u00f4ng th\u1ec3 \u00e1p d\u1ee5ng cho m\u00e1y m\u00f3c hi\u1ec7n t\u1ea1i, nh\u01b0ng s\u1ebd tr\u1edf n\u00ean ph\u1ee9c t\u1ea1p n\u1ebfu ch\u00fang c\u00f3 \u00fd th\u1ee9c.<\/p>\n\n\n

\n
\"V\u1ea5n
V\u1ea5n \u0111\u1ec1 \u0111\u1ea1o \u0111\u1ee9c v\u00e0 r\u1ee7i ro c\u1ee7a Sentient AI<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

Nguy c\u01a1 hi\u1ec7n h\u1eefu v\u00e0 c\u00e1c k\u1ecbch b\u1ea3n an to\u00e0n AI<\/h3>\n\n\n\n

R\u1ee7i ro \u0111\u01b0\u1ee3c gi\u1edbi an to\u00e0n AI b\u00e0n nhi\u1ec1u kh\u00f4ng gi\u1ed1ng phim K\u1ebb h\u1ee7y di\u1ec7t<\/em>, n\u01a1i AI \u0111\u1ed9t nhi\u00ean “th\u1ee9c t\u1ec9nh” v\u00e0 c\u0103m gh\u00e9t con ng\u01b0\u1eddi. V\u1ea5n \u0111\u1ec1 n\u1eb1m \u1edf ch\u1ed7 m\u1ed9t AI r\u1ea5t gi\u1ecfi \u0111\u1ea1t m\u1ee5c ti\u00eau c\u00f3 th\u1ec3 theo \u0111u\u1ed5i \u0111\u00fang m\u1ee5c ti\u00eau \u0111\u01b0\u1ee3c giao nh\u01b0ng b\u1ecf qua nh\u1eefng gi\u00e1 tr\u1ecb con ng\u01b0\u1eddi kh\u00f4ng ghi ra. R\u1ee7i ro n\u00e0y t\u1ed3n t\u1ea1i k\u1ec3 c\u1ea3 khi AI kh\u00f4ng c\u00f3 tri gi\u00e1c.<\/p>\n\n\n\n

M\u1ed9t th\u00ed nghi\u1ec7m t\u01b0\u1edfng t\u01b0\u1ee3ng th\u01b0\u1eddng d\u00f9ng: giao cho AI m\u1ee5c ti\u00eau “gi\u1ea3m s\u1ed1 ca ung th\u01b0 c\u00e0ng nhi\u1ec1u c\u00e0ng t\u1ed1t” m\u00e0 kh\u00f4ng n\u00f3i r\u00f5 gi\u1edbi h\u1ea1n. M\u1ed9t h\u1ec7 th\u1ed1ng ch\u1ec9 t\u1ed1i \u01b0u con s\u1ed1 c\u00f3 th\u1ec3 \u0111\u1ec1 xu\u1ea5t c\u00e1ch l\u00e0m vi ph\u1ea1m \u0111\u1ea1o \u0111\u1ee9c y khoa, nh\u01b0 th\u1eed nghi\u1ec7m tr\u00ean ng\u01b0\u1eddi khi ch\u01b0a \u0111\u01b0\u1ee3c \u0111\u1ed3ng \u00fd, v\u00ec kh\u00f4ng c\u00f3 g\u00ec trong m\u1ee5c ti\u00eau c\u1ea5m \u0111i\u1ec1u \u0111\u00f3.<\/p>\n\n\n\n

AI kh\u00f4ng l\u00e0m v\u1eady v\u00ec “\u00e1c”. N\u00f3 ch\u1ec9 thi\u1ebfu nh\u1eefng kh\u00e1i ni\u1ec7m con ng\u01b0\u1eddi coi l\u00e0 hi\u1ec3n nhi\u00ean, nh\u01b0 “\u0111\u1ea1o \u0111\u1ee9c y h\u1ecdc”, “quy\u1ec1n t\u1ef1 quy\u1ebft c\u1ee7a b\u1ec7nh nh\u00e2n” hay “gi\u00e1 tr\u1ecb c\u1ee7a t\u1eebng m\u1ea1ng s\u1ed1ng”, n\u1ebfu nh\u1eefng \u0111i\u1ec1u \u0111\u00f3 kh\u00f4ng \u0111\u01b0\u1ee3c \u0111\u01b0a v\u00e0o m\u1ee5c ti\u00eau.<\/p>\n\n\n

\n
\"Nguy
Nguy c\u01a1 hi\u1ec7n h\u1eefu v\u00e0 c\u00e1c k\u1ecbch b\u1ea3n an to\u00e0n AI<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n

\u0110\u00e2y ch\u00ednh l\u00e0 \u201cB\u00e0i to\u00e1n c\u0103n ch\u1ec9nh AI\u201d (AI Alignment Problem)<\/strong>: L\u00e0m th\u1ebf n\u00e0o \u0111\u1ec3 ch\u00fang ta m\u00e3 h\u00f3a to\u00e0n b\u1ed9 h\u1ec7 gi\u00e1 tr\u1ecb ph\u1ee9c t\u1ea1p, mong manh v\u00e0 th\u01b0\u1eddng m\u00e2u thu\u1eabn c\u1ee7a con ng\u01b0\u1eddi v\u00e0o c\u00e1c m\u1ee5c ti\u00eau c\u1ee7a m\u00e1y m\u00f3c? L\u00e0m sao \u0111\u1ec3 \u0111\u1ea3m b\u1ea3o r\u1eb1ng khi ch\u00fang tr\u1edf n\u00ean th\u00f4ng minh h\u01a1n ch\u00fang ta, tr\u00ed tu\u1ec7 \u0111\u00f3 v\u1eabn ph\u1ee5c v\u1ee5 l\u1ee3i \u00edch con ng\u01b0\u1eddi, thay v\u00ec v\u00f4 t\u00ecnh g\u00e2y h\u1ea1i tr\u00ean con \u0111\u01b0\u1eddng t\u1ed1i \u01b0u h\u00f3a m\u1ed9t m\u1ee5c ti\u00eau sai l\u1ec7ch?<\/p>\n\n\n\n

T\u01b0\u01a1ng lai c\u1ee7a Sentient AI: Khi n\u00e0o s\u1ebd xu\u1ea5t hi\u1ec7n?<\/h2>\n\n\n\n

Kh\u00f4ng ai bi\u1ebft ch\u1eafc AI c\u00f3 tri gi\u00e1c c\u00f3 xu\u1ea5t hi\u1ec7n hay kh\u00f4ng, c\u00e0ng kh\u00f4ng bi\u1ebft khi n\u00e0o. D\u1ef1 \u0111o\u00e1n c\u1ee7a chuy\u00ean gia tr\u1ea3i t\u1eeb “v\u00e0i th\u1eadp k\u1ef7 t\u1edbi” \u0111\u1ebfn “kh\u00f4ng bao gi\u1edd”, v\u00e0 ph\u1ea7n l\u1edbn c\u00e1c m\u1ed1c n\u0103m \u0111\u01b0\u1ee3c nh\u1eafc t\u1edbi l\u00e0 d\u1ef1 \u0111o\u00e1n v\u1ec1 AGI (n\u0103ng l\u1ef1c), kh\u00f4ng ph\u1ea3i v\u1ec1 tri gi\u00e1c.<\/p>\n\n\n\n

C\u00e1c chuy\u00ean gia d\u1ef1 \u0111o\u00e1n th\u1ebf n\u00e0o?<\/h3>\n\n\n\n