{"id":622,"date":"2017-10-28T10:01:51","date_gmt":"2017-10-28T03:01:51","guid":{"rendered":"https:\/\/rosetta.vn\/nguyenxuanxanh\/?p=622"},"modified":"2022-12-25T17:12:45","modified_gmt":"2022-12-25T10:12:45","slug":"cuoc-dai-chien-giua-may-va-nguoi-trong-co-vay-phan-i","status":"publish","type":"post","link":"https:\/\/rosetta.vn\/nguyenxuanxanh\/cuoc-dai-chien-giua-may-va-nguoi-trong-co-vay-phan-i\/","title":{"rendered":"Cu\u1ed9c \u0111\u1ea1i chi\u1ebfn gi\u1eefa m\u00e1y v\u00e0 ng\u01b0\u1eddi trong c\u1edd v\u00e2y (Ph\u1ea7n I)"},"content":{"rendered":"

 <\/p>\n

CU\u1ed8C \u0110\u1ea0I CHI\u1ebeN GI\u1eeeA<\/span><\/h1>\n

M\u00c1Y V\u00c0 NG\u01af\u1edcI TRONG <\/span>C\u1edc V\u00c2Y<\/span><\/h1>\n

(Ph\u1ea7n I)<\/span><\/strong><\/p>\n

AI \u0111\u1ea5u v\u1edbi AI: AlphaGo Zero t\u1ef1-d\u1ea1y<\/span><\/h2>\n

\u0110\u00e3 \u0111\u00e1nh b\u1ea1i ti\u1ec1n th\u00e2n c\u1ee7a n\u00f3<\/span>1<\/sup><\/h3>\n

T\u00e1c gi\u1ea3: Larry Greenemeier2<\/sup>, 18 th\u00e1ng 10, 2017<\/p>\n

(Courtesy of Scientific American)<\/i><\/p>\n

Ng\u01b0\u1eddi d\u1ecbch<\/i>: V\u00f5 Th\u00e0nh Minh Tu\u1ec7 & Nguy\u1ec5n Xu\u00e2n Xanh<\/h3>\n

L\u1eddi n\u00f3i \u0111\u1ea7u c\u1ee7a ng\u01b0\u1eddi d\u1ecbch<\/i>. \u0110\u00fang 20 n\u0103m tr\u01b0\u1edbc, 1997, trong m\u1ed9t cu\u1ed9c \u0111\u1ed1i \u0111\u1ea7u gi\u1eefa ng\u01b0\u1eddi v\u1edbi m\u00e1y, Deep Blue c\u1ee7a IBM \u0111\u00e3 \u0111\u00e1nh b\u1ea1i \u0111\u1ea1i ki\u1ec7n t\u01b0\u1edbng c\u1edd vua th\u1ebf gi\u1edbi Garry Kasparov, g\u00e2y ch\u1ea5n \u0111\u1ed9ng d\u01b0 lu\u1eadn. Nh\u01b0ng t\u1eeb nhi\u1ec1u th\u1eadp ni\u00ean qua, ng\u01b0\u1eddi ta tin r\u1eb1ng c\u1edd v\u00e2y c\u1ed5 \u0111\u1ea1i kh\u00f3 c\u00f3 th\u1ec3 b\u1ecb \u0111\u00e1nh b\u1ea1i b\u1edfi c\u00e1c m\u00e1y t\u00ednh (m\u00e1y t\u00ednh) do t\u00ednh ch\u1ea5t tinh t\u1ebf v\u00e0 ph\u1ee9c t\u1ea1p c\u1ee7a n\u00f3 m\u00e0 ch\u1ec9 b\u1ed9 \u00f3c con ng\u01b0\u1eddi m\u1edbi l\u00e0m ch\u1ee7 \u0111\u01b0\u1ee3c. C\u1edd vua l\u00e0 m\u1ed9t tr\u1eadn \u0111\u00e1nh, trong khi c\u1edd v\u00e2y \u0111\u01b0\u1ee3c v\u00ed l\u00e0 m\u1ed9t cu\u1ed9c chi\u1ebfn tranh to\u00e0n di\u1ec7n. Nh\u01b0ng r\u1ed3i v\u00e0i n\u0103m qua, c\u00e1c \u0111\u1ea1i ki\u1ec7n t\u01b0\u1edbng c\u1edd v\u00e2y th\u1ebf gi\u1edbi l\u1ea7n l\u01b0\u1ee3c b\u1ecb \u0111\u00e1nh b\u1ea1i b\u1edfi m\u1ed9t h\u1ec7 th\u1ed1ng tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o c\u00f3 t\u00ean AlphaGo, g\u00e2y s\u1eedng s\u1ed1t th\u1ebf gi\u1edbi. V\u00e0 m\u1edbi \u0111\u00e2y, t\u1edbi phi\u00ean m\u1ed9t lo\u1ea1i tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o (AI) c\u00f3 t\u00ean AlphaGo Zero l\u1ea1i c\u00f3 th\u1ec3 \u0111\u00e1nh b\u1ea1i \u0111\u01b0\u1ee3c phi\u00ean b\u1ea3n th\u1ee9 nh\u1ea5t \u0111\u00e3 t\u1eebng chi\u1ebfn th\u1eafng oanh li\u1ec7t. Trong khi AlphaGo \u0111\u01b0\u1ee3c m\u00f4 ph\u1ecfng theo c\u00e1ch c\u1ed5 \u0111i\u1ec3n c\u1ee7a Deep Blue, t\u1ee9c d\u1ef1a l\u00ean v\u00f4 s\u1ed1 data kinh nghi\u1ec7m c\u1ee7a c\u00e1c k\u1ef3 th\u1ee7 c\u1edd vua \u0111i tr\u01b0\u1edbc, th\u00ec AlphaGo Zero l\u1ea1i ho\u1ea1t \u0111\u1ed9ng \u0111\u1ed9c l\u1eadp<\/i> v\u1edbi DNA tr\u00ed tu\u1ec7 t\u1ed5ng h\u1ee3p c\u1ee7a con ng\u01b0\u1eddi. N\u00f3 c\u00f3 th\u1ec3 t\u1ef1 d\u1ea1y<\/i> v\u00e0 t\u1ef1 h\u1ecdc<\/i>, t\u1ef1 t\u01b0\u01a1ng t\u00e1c v\u1edbi n\u00f3, v\u00e0 ho\u00e0n thi\u1ec7n n\u0103ng l\u1ef1c ch\u1ec9 trong v\u00f2ng 40 ng\u00e0y (!) \u0111\u1ec3 c\u00f3 th\u1ec3 \u0111\u00e1nh b\u1ea1i tri th\u1ee9c t\u00edch l\u0169y h\u00e0ng ng\u00e0n n\u0103m c\u1ee7a con ng\u01b0\u1eddi. \u0110\u00f3 l\u00e0 m\u1ed9t s\u1ef1 r\u1ebd nh\u00e1nh c\u1ee7a ng\u00e0nh tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o? C\u1ea3m h\u1ee9ng, ng\u01b0\u1ee1ng m\u1ed9, k\u00ednh s\u1ee3, v\u00e0 lo \u00e2u v\u1ec1 nh\u1eefng \u1ee9ng d\u1ee5ng c\u1ee7a n\u00f3 trong t\u01b0\u01a1ng lai? B\u00e0i vi\u1ebft d\u01b0\u1edbi \u0111\u00e2y n\u00f3i v\u1ec1 s\u1ef1 ki\u1ec7n \u0111\u00f3.<\/p>\n

Ph\u1ea7n m\u1ec1m ch\u01a1i c\u1edd v\u00e2y (Go) c\u1ee7a DeepMind \u2013 c\u00e1i \u0111\u00e3 th\u1ed1ng l\u00e3nh cu\u1ed9c c\u1ea1nh tranh c\u1ee7a n\u00f3 v\u1edbi con ng\u01b0\u1eddi \u2013 nay \u0111\u00e3 t\u1ed1t h\u01a1n.<\/b><\/p>\n

\"AI<\/p>\n

(Saran Poroong Getty Images<\/a>)<\/p>\n

\u0110\u1ea7u n\u0103m nay, ph\u1ea7n m\u1ec1m tr\u00ed tu\u1ec7 th\u00f4ng minh AlphaGo k\u1ebft th\u00fac 2.500 n\u0103m3<\/sup> th\u1ed1ng tr\u1ecb c\u1ee7a con ng\u01b0\u1eddi tr\u00ean b\u00e0n c\u1edd v\u00e2y. Kh\u00f4ng h\u00e0i l\u00f2ng v\u1edbi chi\u1ebfn th\u1eafng 3-0 tr\u01b0\u1edbc \u0111\u1ed1i th\u1ee7 h\u00e0ng \u0111\u1ea7u (Ke Jie) c\u1ee7a th\u1ebf gi\u1edbi, Cty C\u00f4ng ngh\u1ec7 DeepMind, c\u00f4ng ty s\u00e1ng t\u1ea1o n\u00ean AlphaGo, \u0111\u00e3 c\u00f4ng b\u1ed1 v\u00e0o ng\u00e0y th\u1ee9 t\u01b0 m\u1ed9t phi\u00ean b\u1ea3n \u0111\u01b0\u1ee3c n\u00e2ng cao \u2212 AlphaGo Zero \u2013 phi\u00ean b\u1ea3n m\u00e0 c\u00f4ng ty n\u00f3i \u0111\u00e3 \u0111\u00e1nh b\u1ea1i m\u1ed9t c\u00e1ch \u201cs\u1ea1ch s\u1ebd\u201d ti\u1ec1n th\u00e2n c\u1ee7a n\u00f3 trong m\u1ed9t cu\u1ed9c gi\u00e1p m\u1eb7t-AI v\u1edbi nhau, chi\u1ebfn th\u1eafng t\u1ea5t c\u1ea3 100 b\u00e0n \u0111\u00e3 \u0111\u01b0\u1ee3c ch\u01a1i. Nh\u01b0ng \u0111i\u1ec1u c\u00f3 l\u1ebd c\u00f2n quan tr\u1ecdng h\u01a1n c\u00e1c chi\u1ebfn th\u1eafng n\u00e0y l\u00e0 l\u00e0m sao m\u00e0 AlphaGo Zero \u0111\u00e3 l\u1ea5n \u00e1t v\u01b0\u1ee3t tr\u1ed9i nh\u01b0 th\u1ebf. Kh\u00f4ng gi\u1ed1ng nh\u01b0 AlphaGo ban \u0111\u1ea7u, c\u00e1i m\u00e0 DeepMind hu\u1ea5n luy\u1ec7n qua th\u1eddi gian b\u1eb1ng c\u00e1ch s\u1eed d\u1ee5ng c\u00e1c s\u1ed1 l\u01b0\u1ee3ng l\u1edbn c\u1ee7a tri th\u1ee9c con ng\u01b0\u1eddi, c\u00f3 s\u1ef1 gi\u00e1m s\u00e1t, l\u1ea7n n\u00e0y algorit (thu\u1eadt to\u00e1n) c\u1ee7a h\u1ec7 th\u1ed1ng m\u1edbi \u0111\u00e3 t\u1ef1 d\u1ea1y<\/i> cho n\u00f3 \u0111\u1ec3 bi\u1ebft l\u00e0m ch\u1ee7 cu\u1ed9c ch\u01a1i.<\/p>\n

\"\"

AlphaGo \u0111\u00e1nh v\u1edbi Lee Sedol (ph\u1ea3i) (T\u1ea1p ch\u00ed Wired<\/i>)<\/p><\/div>\n

AI (tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o) l\u00e0m cho m\u00e1y t\u00ednh nh\u1eadn d\u1ea1ng \u0111\u01b0\u1ee3c g\u01b0\u01a1ng m\u1eb7t, l\u00e0m nh\u1eefng cu\u1ed9c gi\u1edbi thi\u1ec7u mua h\u00e0ng tr\u1ef1c tuy\u1ebfn, v\u00e0 th\u1ef1c hi\u1ec7n c\u1ea3 vi\u1ec7c \u0111\u1eadu xe song song. M\u00e1y t\u00ednh thu nh\u1eadn nh\u1eefng kh\u1ea3 n\u0103ng n\u00e0y t\u1eeb c\u00e1c \u201calgorit h\u1ecdc\u201d (learning algorithms), \u0111\u01b0\u1ee3c con ng\u01b0\u1eddi vi\u1ebft ra, v\u00e0 n\u1ea1p nh\u1eefng s\u1ed1 l\u01b0\u1ee3ng l\u1edbn c\u1ee7a c\u00e1c data hu\u1ea5n luy\u1ec7n v\u00e0o m\u1ea1ng n\u01a1-ron (th\u1ea7n kinh) nh\u00e2n t\u1ea1o (\u0111\u01b0\u1ee3c \u0111\u1eb7t t\u00ean nh\u01b0 th\u1ebf v\u00ec kh\u1ea3 n\u0103ng c\u1ee7a n\u00f3 x\u1eed l\u00fd th\u00f4ng tin theo c\u00e1ch th\u1ee9c d\u1ef1a tr\u00ean c\u1ea5u tr\u00fac t\u1ebf b\u00e0o n\u01a1-ron c\u1ee7a n\u00e3o m\u1ed9t c\u00e1ch l\u1ecfng l\u1ebbo). Qu\u00e1 tr\u00ecnh n\u00e0y \u0111\u01b0\u1ee3c g\u1ecdi l\u00e0 m\u00e1y h\u1ecdc<\/i> (machine learning). Trong tr\u01b0\u1eddng h\u1ee3p c\u1ee7a AlphaGo, \u0111i\u1ec1u n\u00e0y li\u00ean quan \u0111\u1ebfn vi\u1ec7c ph\u00e2n t\u00edch h\u00e0ng tri\u1ec7u c\u00e1c n\u01b0\u1edbc \u0111i m\u00e0 c\u00e1c chuy\u00ean gia ng\u01b0\u1eddi c\u1edd v\u00e2y \u0111\u00e3 t\u1eebng ch\u01a1i, v\u00e0 th\u00f4ng qua vi\u1ec7c ch\u01a1i \u0111i ch\u01a1i l\u1ea1i r\u1ea5t nhi\u1ec1u v\u00e1n c\u1edd v\u1edbi ch\u00ednh n\u00f3 \u0111\u1ec3 c\u1ee7ng c\u1ed1 nh\u1eefng \u0111i\u1ec1u n\u00f3 \u0111\u00e3 h\u1ecdc \u0111\u01b0\u1ee3c. AlphaGo \u0111\u00e3 \u0111\u00e1nh b\u1ea1i Ke Jie<\/strong>, k\u1ef3 th\u1ee7 c\u1edd v\u00e2y h\u00e0ng \u0111\u1ea7u th\u1ebf gi\u1edbi \u2013 v\u00e0o th\u00e1ng N\u0103m4<\/sup>. Th\u00e1ng 3, 2016, n\u00f3 \u0111\u00e3 \u0111\u00e1nh b\u1ea1i m\u1ed9t k\u1ef3 th\u1ee7 h\u00e0ng \u0111\u1ea7u kh\u00e1c, Lee Sedol5<\/sup>, v\u1edbi s\u1ef1 tr\u1ee3 l\u1ef1c c\u1ee7a c\u00e1c m\u1ea1ng n\u01a1-ron (neural networks) m\u00e0 c\u00e1c m\u00e1y t\u00ednh c\u1ee7a ch\u00fang \u0111\u00f2i h\u1ecfi 48 \u0111\u01a1n v\u1ecb x\u1eed l\u00fd tenx\u01a1 (tensor processing units, TPU) \u2013 (l\u00e0) c\u00e1c microchip chuy\u00ean m\u00f4n h\u00f3a \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf \u0111\u1eb7c bi\u1ec7t cho s\u1ef1 hu\u1ea5n luy\u1ec7n m\u1ea1ng n\u01a1-ron. (Xem: TPU: Chip c\u1ee7a T\u01b0\u01a1ng lai<\/a>.)<\/p>\n

S\u1ef1 hu\u1ea5n luy\u1ec7n c\u1ee7a AlphaGo Zero s\u1eed d\u1ee5ng b\u1ed1n TPU v\u00e0 m\u1ed9t m\u1ea1ng n\u01a1-ron duy nh\u1ea5t, m\u1ea1ng m\u00e0 ban \u0111\u1ea7u ch\u1eb3ng bi\u1ebft g\u00ec v\u1ec1 c\u1edd v\u00e2y. Tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o, AI, \u0111\u00e3 h\u1ecdc m\u00e0 kh\u00f4ng c\u1ea7n s\u1ef1 gi\u00e1m s\u00e1t \u2013 n\u00f3 \u0111\u01a1n gi\u1ea3n t\u1ef1 ch\u01a1i v\u1edbi n\u00f3, v\u00e0 ch\u1eb3ng bao l\u00e2u n\u00f3 \u0111\u00e3 \u0111o\u00e1n tr\u01b0\u1edbc \u0111\u01b0\u1ee3c nh\u1eefng b\u01b0\u1edbc \u0111i c\u1ee7a n\u00f3, v\u00e0 c\u00f3 th\u1ec3 h\u00ecnh dung ch\u00fang s\u1ebd \u1ea3nh h\u01b0\u1edfng \u0111\u1ebfn k\u1ebft c\u1ee5c c\u1ee7a v\u00e1n c\u1edd nh\u01b0 th\u1ebf n\u00e0o. \u201cK\u1ef9 thu\u1eadt n\u00e0y m\u1ea1nh h\u01a1n c\u00e1c phi\u00ean b\u1ea3n tr\u01b0\u1edbc c\u1ee7a AlphaGo b\u1edfi v\u00ec n\u00f3 kh\u00f4ng c\u00f2n b\u1ecb r\u00e0ng bu\u1ed9c b\u1edfi c\u00e1c gi\u1edbi h\u1ea1n c\u1ee7a tri th\u1ee9c con ng\u01b0\u1eddi\u201d, theo nh\u01b0 m\u1ed9t b\u00e0i vi\u1ebft blog c\u1ee7a ng\u01b0\u1eddi \u0111\u1ed3ng s\u00e1ng l\u1eadp Demis Hassabis c\u1ee7a DeepMind, v\u00e0 David Silver, ng\u01b0\u1eddi h\u01b0\u1edbng d\u1eabn nh\u00f3m nghi\u00ean c\u1ee9u h\u1ecdc t\u0103ng c\u01b0\u1eddng (reinforcement learning) c\u1ee7a c\u00f4ng ty. (DeepMind l\u00e0 m\u1ed9t \u0111\u01a1n v\u1ecb c\u1ee7a Alphabet, Inc., c\u00f4ng ty m\u1eb9 c\u1ee7a Google.) M\u1ed9t v\u1ea5n \u0111\u1ec1 v\u1edbi AI l\u00e0 lu\u00f4n lu\u00f4n ph\u1ea3i d\u1ef1a l\u00ean tri th\u1ee9c con ng\u01b0\u1eddi, do \u0111\u00f3 th\u00f4ng tin nh\u01b0 th\u1ebf c\u00f3 th\u1ec3 qu\u00e1 \u0111\u1eaft, qu\u00e1 kh\u00f4ng tin c\u1eady, hay \u0111\u01a1n gi\u1ea3n kh\u00f4ng t\u1ed3n t\u1ea1i trong nh\u1eefng ho\u00e0n c\u1ea3nh nh\u1ea5t \u0111\u1ecbnh. \u201cN\u1ebfu nh\u1eefng k\u1ef9 thu\u1eadt t\u01b0\u01a1ng t\u1ef1 c\u00f3 th\u1ec3 \u0111\u01b0\u1ee3c \u00e1p d\u1ee5ng cho nh\u1eefng b\u00e0i to\u00e1n c\u00f3 c\u1ea5u tr\u00fac kh\u00e1c nh\u01b0 s\u1ef1 cu\u1ed1n g\u1ea5p protein (protein folding), gi\u1ea3m b\u1edbt s\u1ef1 ti\u00eau th\u1ee5 n\u0103ng l\u01b0\u1ee3ng, hay \u0111i t\u00ecm c\u00e1c v\u1eadt li\u1ec7u m\u1edbi c\u00f3 t\u00ednh c\u00e1ch m\u1ea1ng, c\u00e1c b\u1ee9t ph\u00e1 t\u00ecm \u0111\u01b0\u1ee3c s\u1ebd c\u00f3 ti\u1ec1m n\u0103ng \u1ea3nh h\u01b0\u1edfng t\u1ed1t l\u00ean x\u00e3 h\u1ed9i\u201d, b\u00e0i vi\u1ebft blog n\u00f3i.<\/p>\n

AlphaGo Zero ngh\u0129 ra ngay c\u1ea3 c\u00e1c chi\u1ebfn l\u01b0\u1ee3c b\u1ea5t-quy \u01b0\u1edbc c\u1ee7a ri\u00eang n\u00f3. C\u1edd v\u00e2y ti\u00eau bi\u1ec3u ch\u01a1i b\u1eb1ng c\u00e1ch s\u1eed d\u1ee5ng c\u00e1c \u201cvi\u00ean \u0111\u00e1\u201d nhu\u1ed9m m\u00e0u \u0111en hay tr\u1eafng tr\u00ean m\u1ed9t b\u00e0n c\u1edd v\u1edbi 19×19 d\u00f2ng k\u1ebb. M\u1ed7i ng\u01b0\u1eddi ch\u01a1i \u0111\u1eb7t c\u00e1c vi\u00ean \u0111\u00e1 (t\u1ea1i c\u00e1c n\u00fat giao \u0111i\u1ec3m) v\u1edbi m\u1ee5c ti\u00eau bao v\u00e2y c\u00e1c vi\u00ean \u0111\u00e1 c\u1ee7a \u0111\u1ed1i ph\u01b0\u01a1ng. \u201cTrong l\u00fac hu\u1ea5n luy\u1ec7n, AlphaGo Zero \u0111\u00e3 kh\u00e1m ph\u00e1, ch\u01a1i v\u00e0 cu\u1ed1i c\u00f9ng h\u1ecdc c\u00e1ch ch\u1ecdn \u01b0u ti\u00ean cho m\u1ed9t lo\u1ea1t c\u00e1c bi\u1ebfn th\u1ec3 c\u1ee7a \u0111\u1ecbnh th\u1ee9c (joseki, chu\u1ed7i n\u01b0\u1edbc \u0111i) tr\u01b0\u1edbc \u0111\u00e2y ch\u01b0a \u0111\u01b0\u1ee3c bi\u1ebft\u201d, ph\u00e1t ng\u00f4n vi\u00ean Jon Fildes c\u1ee7a DeepMind n\u00f3i. C\u1edd v\u00e2y \u0111i\u1ec3n h\u00ecnh b\u1eaft \u0111\u1ea7u v\u1edbi nh\u1eefng n\u01b0\u1edbc \u0111i trong c\u00e1c g\u00f3c c\u1ee7a \u0111\u01b0\u1eddng k\u1ebb, cho ph\u00e9p m\u1ed9t ng\u01b0\u1eddi ch\u01a1i t\u1ea1o \u0111\u01b0\u1ee3c m\u1ed9t v\u1ecb th\u1ebf to\u00e0n c\u1ee5c t\u1ed1t cho b\u00e0n c\u1edd. \u201cNh\u01b0 n\u01b0\u1edbc \u0111i th\u1ee9 37 trong v\u00e1n th\u1ee9 hai \u0111\u01b0\u1ee3c ch\u01a1i \u0111\u1ea5u v\u1edbi Lee Sedol, nh\u1eefng kho\u1ea3ng kh\u1eafc c\u1ee7a c\u1ea3m h\u1ee9ng thu\u1eadt to\u00e1n cho ch\u00fang ta m\u1ed9t c\u00e1i nh\u00ecn tho\u00e1ng qua c\u1ee7a t\u00ednh s\u00e1ng t\u1ea1o c\u1ee7a AlphaGo v\u00e0 ti\u1ec1m n\u0103ng c\u1ee7a AI\u201d, ng\u01b0\u1eddi ph\u00e1t ng\u00f4n n\u00f3i th\u00eam. An Young-gil, m\u1ed9t tay ch\u01a1i c\u1edd v\u00e2y chuy\u00ean nghi\u1ec7p c\u1ee7a Nam H\u00e0n c\u00f3 tr\u00ecnh \u0111\u1ed9 \u0111\u1ec7-b\u00e1t-\u0111\u1eb3ng (cao nh\u1ea5t l\u00e0 \u0111\u1ec7-c\u1eedu-\u0111\u1eb3ng) \u0111\u00e3 l\u1ecdc ra \u0111\u01b0\u1ee3c n\u01b0\u1edbc th\u1ee9 37 l\u00e0 n\u01b0\u1edbc \u201chi\u1ebfm hoi v\u00e0 l\u00e0m cho m\u00ea h\u1ed3n\u201d ngay sau cu\u1ed9c \u0111\u1ea5u th\u00e1ng 3, 2016.<\/p>\n

Nghi\u00ean c\u1ee9u c\u1ee7a DeepMind m\u00f4 t\u1ea3 \u201cm\u1ed9t k\u1ebft qu\u1ea3 k\u1ef9 thu\u1eadt r\u1ea5t \u1ea5n t\u01b0\u1ee3ng; v\u00e0 c\u1ea3 hai, kh\u1ea3 n\u0103ng c\u1ee7a ch\u00fang \u0111\u1ec3 l\u00e0m \u0111i\u1ec1u \u0111\u00f3 \u2013 v\u00e0 kh\u1ea3 n\u0103ng c\u1ee7a ch\u00fang hu\u1ea5n luy\u1ec7n h\u1ec7 th\u1ed1ng trong v\u00f2ng 40 ng\u00e0y, tr\u00ean b\u1ed1n TPU \u2013 l\u00e0 \u0111\u00e1ng k\u1ec3\u201d, Oren Etzioni, T\u1ed5ng gi\u00e1m \u0111\u1ed1c c\u1ee7a Vi\u1ec7n nghi\u00ean c\u1ee9u Allen v\u1ec1 Tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o (AI2) n\u00f3i, m\u1ed9t t\u1ed5 ch\u1ee9c m\u00e0 ng\u01b0\u1eddi \u0111\u1ed3ng s\u00e1ng l\u1eadp c\u1ee7a Microsoft, Paul Allen, \u0111\u00e3 th\u00e0nh l\u1eadp trong n\u0103m 2014 \u0111\u1ec3 t\u1eadp trung v\u00e0o nh\u1eefng l\u1ee3i \u00edch ti\u1ec1m n\u0103ng c\u1ee7a AI. \u201cTrong khi nhi\u1ec1u ng\u01b0\u1eddi \u0111\u00e3 s\u1eed d\u1ee5ng [h\u1ecdc t\u0103ng c\u01b0\u1eddng] tr\u01b0\u1edbc \u0111\u00e2y, nh\u1eefng m\u1eb7t k\u1ef9 thu\u1eadt c\u1ee7a c\u00f4ng tr\u00ecnh l\u00e0 m\u1edbi m\u1ebb.\u201d<\/p>\n

Th\u00e0nh c\u00f4ng c\u1ee7a AlphaGo Zero b\u00e1o tr\u01b0\u1edbc \u0111i\u1ec1m hay cho s\u1ef1 l\u00e0m b\u00e1 ch\u1ee7 c\u1ee7a AI \u0111\u1ed1i v\u1edbi c\u00e1c tr\u00f2 ch\u01a1i, Etzioni n\u00f3i. Nh\u01b0ng \u201ct\u00f4i v\u1eabn ngh\u0129 s\u1ebd l\u00e0 \u0111i\u1ec1u sai l\u1ea7m n\u1ebfu tin r\u1eb1ng ch\u00fang ta \u0111\u00e3 h\u1ecdc \u0111\u01b0\u1ee3c \u0111i\u1ec1u g\u00ec t\u1ed5ng qu\u00e1t v\u1ec1 t\u01b0 duy v\u00e0 v\u1ec1 s\u1ef1 h\u1ecdc cho tr\u00ed th\u00f4ng minh n\u00f3i chung\u201d, \u00f4ng n\u00f3i. \u201cC\u00e1ch ti\u1ebfp c\u1eadn n\u00e0y s\u1ebd kh\u00f4ng ho\u1ea1t \u0111\u1ed9ng \u0111\u01b0\u1ee3c trong nh\u1eefng b\u00e0i to\u00e1n c\u00f3 c\u1ea5u tr\u00fac kh\u00f4ng th\u00edch h\u1ee3p nh\u01b0 hi\u1ec3u bi\u1ebft ng\u00f4n ng\u1eef t\u1ef1 nhi\u00ean, hay robotic, \u1edf \u0111\u00f3 kh\u00f4ng gian tr\u1ea1ng th\u00e1i (state space) l\u00e0 ph\u1ee9c t\u1ea1p h\u01a1n, v\u00e0 kh\u00f4ng c\u00f3 m\u1ed9t h\u00e0m t\u1ed1i \u01b0u\u201d (objective function, loss function) r\u00f5 r\u00e0ng.<\/p>\n

S\u1ef1 t\u1eadp luy\u1ec7n kh\u00f4ng c\u1ea7n gi\u00e1m s\u00e1t l\u00e0 ch\u00eca kh\u00f3a th\u00e0nh c\u00f4ng \u0111\u1ec3 cu\u1ed1i c\u00f9ng t\u1ea1o ra AI c\u00f3 th\u1ec3 t\u1ef1 suy ngh\u0129 cho m\u00ecnh, Etzioni n\u00f3i, nh\u01b0ng \u201cc\u00f2n c\u1ea7n \u0111\u1ebfn nhi\u1ec1u nghi\u00ean c\u1ee9u h\u01a1n b\u00ean ngo\u00e0i c\u00e1c gi\u1edbi h\u1ea1n c\u1ee7a c\u00e1c tr\u00f2 ch\u01a1i b\u1ea3ng, v\u00e0 nh\u1eefng ch\u1ee9c n\u0103ng kh\u00e1ch quan ti\u1ec1n-\u0111\u1ecbnh (predefined) tr\u01b0\u1edbc khi c\u00e1c m\u00e1y t\u00ednh c\u00f3 b\u1eaft \u0111\u1ea7u t\u01b0 duy ngo\u00e0i chi\u1ebfc h\u1ed9p.\/. (Xem ti\u1ebfp Ph\u1ea7n II<\/a>)<\/p>\n

\n

Ch\u00fa gi\u1ea3i:<\/p>\n

    \n
  1. B\u00e0i g\u1ed1c: <\/span>https:\/\/www.scientificamerican.com\/article\/ai-versus-ai-self-taught-alphago-zero-vanquishes-its-predecessor\/<\/span><\/a>. (Courtesy of Scientific American)<\/span><\/li>\n
  2. Larry Greenemeier l\u00e0 ch\u1ee7 b\u00fat li\u00ean k\u1ebft c\u1ee7a m\u1ee5c c\u00f4ng ngh\u1ec7 cho <\/span>Scientific American<\/span><\/i>, l\u00e3nh v\u1ef1c bao g\u1ed3m m\u1ed9t s\u1ed1 \u0111\u1ec1 t\u00e0i li\u00ean quan \u0111\u1ebfn c\u00f4ng ngh\u1ec7, trong \u0111\u00f3 c\u00f3 biotech, m\u00e1y t\u00ednh, c\u00f4ng ngh\u1ec7 qu\u00e2n s\u1ef1, c\u00f4ng ngh\u1ec7 nano v\u00e0 robot.<\/span><\/li>\n
  3. C\u1edd v\u00e2y t\u1ed3n t\u1ea1i \u1edf Trung Qu\u1ed1c 2.500 n\u0103m<\/span><\/li>\n
  4. Tr\u1eadn \u0111\u1ea5u 3 v\u00e1n t\u1eeb ng\u00e0y 23-27\/5\/2017, di\u1ec5n ra \u1edf Trung Qu\u1ed1c. Ke Jie<\/strong> l\u00e0 \u0111\u1ea1i ki\u1ec7n t\u01b0\u1edbng th\u1ebf gi\u1edbi \u0111\u1ec7 c\u1eedu \u0111\u1eb3ng ng\u01b0\u1eddi Trung Qu\u1ed1c. \u00d4ng m\u1ea5t \u0111i gi\u1ea3i th\u01b0\u1edfng 1.5 tri\u1ec7u \u0111\u00f4 la M\u1ef9. Sau khi AlphaGo chi\u1ebfn th\u1eafng, H\u1ed9i c\u1edd v\u00e2y Trung Qu\u1ed1c trao t\u1eb7ng AlphaGo h\u00e0m \u201c\u0111\u1ec7 c\u1eedu \u0111\u1eb3ng\u201d.<\/span><\/li>\n
  5. \u0110\u00e2y l\u00e0 tr\u1eadn \u0111\u1ea5u t\u1eeb ng\u00e0y 9-15\/3\/2016, di\u1ec5n ra t\u1ea1i H\u00e0n Qu\u1ed1c. Lee c\u00f3 h\u00e0m \u0111\u1ec7 c\u1eedu \u0111\u1eb3ng, l\u00e0 nh\u00e0 v\u00f4 \u0111\u1ecbch th\u1ebf gi\u1edbi 18 l\u1ea7n, m\u1ed9t t\u00e0i n\u0103ng \u0111\u1eb7c bi\u1ec7t trong l\u00e0ng c\u1edd v\u00e2y, \u0111\u01b0\u1ee3c xem l\u00e0 \u201canh h\u00f9ng qu\u1ed1c gia\u201d c\u1ee7a H\u00e0n Qu\u1ed1c. Sau khi nhi\u1ec1u ki\u1ec7n t\u01b0\u1edbng c\u1edd v\u00e2y th\u1ea5t b\u1ea1i, ng\u01b0\u1eddi ta hy v\u1ecdng Lee Sedol c\u00f3 th\u1ec3 th\u1eafng \u0111\u01b0\u1ee3c AlphaGo. Tr\u1eadn \u0111\u1ea5u c\u00f3 5 v\u00e1n. Ba v\u00e1n \u0111\u1ea7u AlphaGo th\u1eafng li\u00ean ti\u1ebfp, tr\u01b0\u1edbc s\u1ef1 kinh ho\u00e0ng c\u1ee7a Lee. V\u00e1n th\u1ee9 4 Lee th\u1eafng l\u1ea1i. Nh\u01b0ng v\u00e1n th\u1ee9 5 AlphaGo th\u1eafng ti\u1ebfp. Lee \u0111\u00e3 vu\u1ed9t m\u1ea5t gi\u1ea3i th\u01b0\u1edfng m\u1ed9t tri\u1ec7u \u0111\u00f4 la M\u1ef9. Sau tr\u1eadn \u0111\u1ea5u, Hi\u1ec7p h\u1ed9i c\u1edd v\u00e2y H\u00e0n Qu\u1ed1c Korea Baduk Association \u0111\u00e3 trao t\u1eb7ng cho AlphaGo h\u00e0m \u201c\u0111\u1ec7 c\u1eedu \u0111\u1eb3ng\u201d. <\/span><\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"

      CU\u1ed8C \u0110\u1ea0I CHI\u1ebeN GI\u1eeeA M\u00c1Y V\u00c0 NG\u01af\u1edcI TRONG C\u1edc V\u00c2Y (Ph\u1ea7n I) AI \u0111\u1ea5u v\u1edbi AI: AlphaGo Zero t\u1ef1-d\u1ea1y \u0110\u00e3 \u0111\u00e1nh b\u1ea1i ti\u1ec1n th\u00e2n c\u1ee7a n\u00f31 T\u00e1c gi\u1ea3: Larry Greenemeier2, 18 th\u00e1ng 10, 2017 (Courtesy of Scientific American) Ng\u01b0\u1eddi d\u1ecbch: V\u00f5 Th\u00e0nh Minh Tu\u1ec7 & Nguy\u1ec5n Xu\u00e2n Xanh L\u1eddi n\u00f3i \u0111\u1ea7u c\u1ee7a ng\u01b0\u1eddi d\u1ecbch.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_mi_skip_tracking":false,"jetpack_post_was_ever_published":false},"categories":[1],"tags":[50,51,52],"jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/posts\/622"}],"collection":[{"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/comments?post=622"}],"version-history":[{"count":4,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/posts\/622\/revisions"}],"predecessor-version":[{"id":5175,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/posts\/622\/revisions\/5175"}],"wp:attachment":[{"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/media?parent=622"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/categories?post=622"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rosetta.vn\/nguyenxuanxanh\/wp-json\/wp\/v2\/tags?post=622"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}