ラベル

AI Forecast 8306MUFG (248) Social (76) life (73) Financial (64) IT (60) Proposal (56) idea (50) Fund Management (49) Trading (43) economics (42) AI (40) Covid-19 (40) Hedge Fund (37) Risk (36) Culture (31) BOJ (29) Science (27) hobby (24) accounting (17) Apps (16) Travel (14) career (11) job (7) Hiking (5) emotion (5) music (3) Statarb (2) piano (2)

2024年5月18日土曜日

汎用人工知能まで、あと少し Few More Steps to AGI(Artificial General Intelligence)

汎用人工知能まで、あと少し

Few More Steps to AGI(Artificial General Intelligence)


AGIとは、汎用人工知能のことで、人間が長らく追い求めてきたものだ。人間と同じことが出来るロボットのようなもの。AGI refers to Artificial General Intelligence, which is something humanity has long pursued. It is like a robot capable of doing everything a human can do.


数年前、形状認識、パターン認識、囲碁、将棋などでニューラルネットによるAIが世界を驚かせた。A few years ago, neural network-based AI surprised the world with its abilities in shape recognition, pattern recognition, and games like Go and Shogi.


しかしながら、人間の知能、という段階になると、遅々として進んでいなかった。知るとは何だろう?知性とは?意識とは?However, when it comes to the stage of human intelligence, progress has been slow. What does it mean to know? What is intelligence? What is consciousness?


論理的に考えようとすると、最初は言語学、最後は哲学の領域。脳のシミュレーターを作らないとダメか、など様々な試みがなされた。しかし、そのような論理的な系統だったやり方は、ことごとく失敗した。When trying to think logically, it starts with linguistics and ends in the realm of philosophy. Various attempts were made, such as creating a brain simulator, but all such systematic logical approaches failed.


そして最後に、言葉の膨大な関係性のみをひたすらニューラルネットで覚えさせたAIが、人間に最も近い応答ができるようになった。会話をする限り、会話の相手がAIであることが分からないレベルまで進化した。これがChat GPTだ。In the end, an AI that was relentlessly trained on the vast relationships of words through a neural network was able to respond in a way closest to humans. It evolved to a level where, in conversation, you couldn’t tell that the conversation partner was an AI. This is ChatGPT.


その仕組みは、ある意味、囲碁で人間を凌駕して世界一となった、GoogleのAlpha-Zeroに似ていた。すなわち、ニューラルネットには膨大な言語を入力したものの、文法など、文法的、論理的な教育は一切行わなかったのだ。In a way, its mechanism is similar to Google’s AlphaZero, which surpassed humans and became the world’s best in Go. Although a vast amount of language was input into the neural network, there was no grammatical or logical instruction at all.


Googleの囲碁AIは当初、定石などを教育したAlpha-GOがあったが、Alpha-Zeroでは、囲碁のルールを教えただけで、定石は一切教えず、その代わり、2台のAlpha-Zeroを3日間何億回も対局させ続けただけなのだそうだ。Google's Go AI initially had AlphaGo, which was educated with standard strategies, but with AlphaZero, it only taught the rules of Go and did not teach any standard strategies. Instead, it let two AlphaZero machines play against each other billions of times over three days.


OpenAIのサムアルトマン自身、この結果に驚き、世界的な発見としてその成果を共有したのは素晴らしい。今では、ChatGPTの成功ロジックは世界に広く共有されることとなった。この成果を見て、背筋が寒くなった人も多い。ChatGPTにあと、視覚、感覚、手足を加えれば、それは「新人類」として成立してしまうレベルにあるからだ。Sam Altman of OpenAI himself was astonished by this result and shared its achievements as a global discovery, which is remarkable. Now, the logic behind ChatGPT's success is widely shared worldwide. Many people felt a chill down their spine seeing these results, as adding vision, senses, and limbs to ChatGPT would establish it as a "new kind of human."


AGIは、すでにあと少しの所にあると言ってよいだろう。It can be said that AGI is already within reach.