溯因推理：人工智能的盲點

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文給當今人工智能界推崇深度學習的現象潑了冷水，指出了人工智能的盲點：溯因推理，並提醒人們不要忽視深度學習的種種問題，否則將會因爲深度學習又面臨第三次“人工智能的冬天”。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最近深度學習領域的進步再次激發了人們對那些通用人工智能新機器的興趣。這一想法認爲，通過構建更大、更好的神經網絡，我們就能夠獲得越來越接近於人類大腦的數字化版本。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但這是一個神話，計算機科學家 Erik Larson 認爲，所有的證據都表明，人類和機器所擁有的智能有本質的不同。Larson 的新書《"},{"type":"link","attrs":{"href":"https:\/\/www.hup.harvard.edu\/catalog.php?isbn=9780674983519","title":null,"type":null},"content":[{"type":"text","text":"The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do"}]},{"type":"text","text":"》（目前尚無中譯本，本文暫譯爲《人工智能的神話：爲什麼計算機不能像我們這樣思考》），討論了廣泛宣傳的關於智能和推理的誤解，是如何將人工智能研究引向狹窄的道路，限制了創新和科學發現。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/ea\/eab65e412547b0856c2dd4f2d1ae2089.jpeg","alt":null,"title":"《人工智能的神話》， Erik J. Larson 著。","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 警告說，除非科學家、研究人員和支持他們工作的組織不改變方向，否則他們將註定“屈服於機器世界的擴張，在機器世界中，真正的發明被邊緣化，轉而支持那些鼓吹現有方法的未來主義言論，而這正是來自根深蒂固的利益集團。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"人工智能的神話"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以科學的觀點來看，人工智能的神話是假定我們將通過在應用領域取得進展，來實現"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/05\/13\/what-is-artificial-general-intelligence-agi\/","title":null,"type":null},"content":[{"type":"text","text":"通用人工智能"}]},{"type":"text","text":"（Artificial General Intelligence，AGI），例如對圖像進行分類、理解語音指令，或玩遊戲。但是，這些"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/04\/09\/what-is-narrow-artificial-intelligence-ani\/","title":null,"type":null},"content":[{"type":"text","text":"狹窄的人工智能系統"}]},{"type":"text","text":"的基礎技術並不能解決通用智能能力所必須解決的更大挑戰，例如進行基本的對話、完成簡單的家務活，或者其他需要常識性的任務。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 寫道：“當我們成功地應用了更簡單、更狹窄的智能版本，並從更快的計算機和大量的數據中獲益時，我們並沒有逐步取得進展，而是在摘取低垂的果實。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人工智能神話的文化後果就是忽視了"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2019\/12\/03\/francois-chollet-arc-ai-measurement\/","title":null,"type":null},"content":[{"type":"text","text":"智能的科學奧祕"}]},{"type":"text","text":"，無休止地談論"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2019\/02\/15\/what-is-deep-learning-neural-networks\/","title":null,"type":null},"content":[{"type":"text","text":"深度學習"}]},{"type":"text","text":"和其他當代技術的不斷進步。這個神話阻礙了科學家們思考新的方法來應對智能的挑戰。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“如果我們選擇忽視一個核心奧祕，而不是正視它，我們就無法獲得創新，”Larson 寫道，“健康的創新文化強調探索未知，而非誇大現有方法的延伸……關於人工智能必然成功的神話，往往會扼殺真正進步所需要的發明文化。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"演繹、歸納和溯因推理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當你走出家門時，你發現街道是溼的。你首先想到的是，一定是下雨了。但現在是晴天，人行道是乾的，所以你立即排除了下雨的可能性。當你往旁邊看時，你看到一輛灑水車停在街道旁。你就斷定，街道之所以是溼的，是因爲灑水車沖洗了街道。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上面是一個“推理”的例子，即從觀察到結論的行爲，也是智慧生物的基本能力。我們不斷地根據我們所知和感知的事物來推理。它們大多發生在潛意識中。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 寫道：“任何能夠進行推理的系統都必須具有一些基本的智能，因爲利用已知和觀察到的事物來更新信念的行爲本身，必然與我們所指的智慧相關聯。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人工智能研究人員將他們的系統建立在兩種類型的推理機上：演繹式和歸納式。演繹式推理是利用先驗知識對世界進行推理。這是"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2019\/11\/18\/what-is-symbolic-artificial-intelligence\/","title":null,"type":null},"content":[{"type":"text","text":"符號人工智能"}]},{"type":"text","text":"的基礎，也是人工智能早期幾十年研究人員關注的重點。工程師通過賦予它們一組預先定義的規則和事實來創建符號系統，人工智能利用這些知識來推理它所接收到的數據。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"溯因推理是通過經驗獲取知識，在過去十年中，溯因推理在人工智能研究人員和科技公司中獲得了更多的關注。"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2017\/08\/28\/artificial-intelligence-machine-learning-deep-learning\/","title":null,"type":null},"content":[{"type":"text","text":"機器學習算法"}]},{"type":"text","text":"是溯因推理引擎，一個在相關實例上訓練的機器學習模型，將找到輸入映射到輸出的模式。最近幾年，人工智能研究人員利用機器學習、大數據和先進的處理器，對超出符號系統能力的任務進行模型訓練。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第三種推理類型，即溯因推理，是由美國科學家 Charles Sanders Peirce 在 19 世紀首次提出的。溯因推理是一種能夠提出直覺和假設的認知能力，作出比隨機猜測真相更好的猜測。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/3d\/3d886caa02d5fa349503f33ba168ba87.jpeg","alt":null,"title":"美國科學家 Charles Sanders Peirce 在 19 世紀提出了溯因推理。資料來源：紐約公共圖書館，公共領域。","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"舉例來說，街道潮溼的原因可能有很多（包括一些我們從未直接經歷過的），但是通過溯因推理，我們可以選擇最有希望的假設，迅速排除錯誤的假設，尋找新的假設，並得出可靠的結論。正如 Larson 在《人工智能的神話》一書中寫道：“我們從實際上無限可能中猜測哪些假設看起來是可能的或可信的。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"溯因推理就是許多人所說的“常識”。這是我們看待事實或數據的概念框架，也是結合其他推理類型的“粘合劑”。這能讓我們時刻注意大腦中大量的信息，以及我們從感官接受到的海量數據中的相關內容。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但問題在於，人工智能界對溯因推理的關注還很少。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"人工智能與溯因推理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"20 世紀 80 年代和 90 年代，由於溯因邏輯程序（Abductive Logic Programming）的嘗試，溯因進入了人工智能的討論中，但是這些努力都存在缺陷，最終被放棄。Larson 告訴 TechTalks：“它們是對邏輯編程的重新表述，是演繹的一種變體。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/77\/77356ed256aa299b64711da86eab01b0.jpeg","alt":null,"title":"Erik J. Larson，《人工智能的神話》一書作者。","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"溯因在 2010 年代得到了另一個機會，那就是"},{"type":"link","attrs":{"href":"https:\/\/en.wikipedia.org\/wiki\/Bayesian_network","title":null,"type":null},"content":[{"type":"text","text":"貝葉斯網絡"}]},{"type":"text","text":"，它是試圖計算因果關係的推理引擎。但是，與早期的方法一樣，較新的方法也有一個缺陷，它不能捕捉到真正的溯因，Larson 說，貝葉斯和其他圖形模型都是“歸納法的變種”。他在《人工智能的神話》一書中稱它們爲“名副其實的溯因”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在很大程度上，人工智能的歷史都以演繹和歸納爲主導。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“當早期人工智能先驅 Alan Newell、Herbert Simon、John McCarthy 和 Marvin Minsky 開始着手解決人工推理（人工智能的核心）問題時，他們認爲編寫演繹式規則就足以產生智能思維和行動，”Larson 說，“事實上事實並非如此，在關於我們如何做科學的討論中，這一點應該更早被認識到。”“這太奇怪了，沒有人真的停下來，明確地說‘等等，這是行不通的！’” Larson 說，“這將使研究直接轉向溯因或假設的生成，或者說，‘上下文敏感推理’。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在過去的二十年裏，隨着數據和計算資源的日益豐富，機器學習算法——特別是"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2021\/01\/28\/deep-learning-explainer\/","title":null,"type":null},"content":[{"type":"text","text":"深度神經網絡"}]},{"type":"text","text":"——已經成爲人工智能領域的研究熱點。深度學習技術開啓了以前超出計算機極限的應用。這也吸引了"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/12\/21\/deepminds-annual-report-why-its-hard-to-run-a-commercial-ai-lab\/","title":null,"type":null},"content":[{"type":"text","text":"世界上一些最富有的公司"}]},{"type":"text","text":"的興趣和資金。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 說：“我認爲，隨着萬維網的出現，經驗式或歸納式（以數據爲中心）的方法佔據了上風，而溯因法和演繹法一樣，基本上被遺忘了。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但是，機器學習系統也存在嚴重的侷限性，包括缺乏因果關係，邊緣情況處理不好和對數據的需求過多。同時，隨着研究人員嘗試將機器學習應用到醫療和金融等敏感領域，這些限制正變得越來越明顯，問題也越來越多。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"溯因推理與人工智能的未來之路"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"包括強化學習先驅 "},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2019\/03\/25\/richard-sutton-artificial-intelligence-research\/","title":null,"type":null},"content":[{"type":"text","text":"Richard Sutton"}]},{"type":"text","text":" 在內的一些科學家認爲，我們應該堅持使用能夠隨着數據和計算的可用性而擴展的方法，即學習和搜索。舉例來說，隨着神經網絡的規模越來越大，數據也越來越豐富，它們最終會超越極限，帶來新的突破。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 駁斥了數據驅動的人工智能的大規模應用，認爲“作爲一種智能模型，其本質是有缺陷的”。他重申，儘管搜索和學習都可以提供有用的應用，但是它們是基於非溯因推理。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“如果沒有推理思維的革命，搜索就無法擴展到常識或溯因推理中，而這還沒有發生。與機器學習類似，學習方法的數據驅動特性意味着推理必須來自數據，可以說，人們經常進行的許多智能推理顯然不是這樣的，”Larson 說，“我們不會僅僅通過觀察過去，比如說，從大型數據集中獲取的數據，就能弄清楚對未來的結論、思考或者推理。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另一些科學家認爲，"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/03\/04\/gary-marcus-hybrid-ai\/","title":null,"type":null},"content":[{"type":"text","text":"混合人工智能"}]},{"type":"text","text":"將符號系統與神經網絡結合起來，將有望克服深度學習的缺陷。IBM Watson 就是一個例子，它在 《危險邊緣》（Jeopardy!）美國電視智力問答節目中擊敗了世界冠軍而聞名。最新的概念證明了混合模型在單獨的符號人工智能和深度學習表現不佳的應用中"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2019\/06\/05\/mit-ibm-hybrid-ai\/","title":null,"type":null},"content":[{"type":"text","text":"有很好的效果"}]},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 認爲，混合系統可以填補僅靠機器學習或僅靠規則方法的空白。身爲自然語言處理領域的研究人員，他目前正致力於將大型與訓練語言模型，如 "},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/08\/17\/openai-gpt-3-commercial-ai\/","title":null,"type":null},"content":[{"type":"text","text":"GPT-3"}]},{"type":"text","text":"，與知識圖譜形式的語義網方面的舊工作相結合，以便在搜索、問題回答和其他任務中創造更好的應用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“但是演繹-歸納組合並不能讓我們達到溯因，因爲這三種類型的推理在形式上是不同的，所以它們不能相互還原，也不能結合起來得到第三種。”他說。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 在《人工智能的神話》一書中將規避溯因的努力稱爲“推理陷阱”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"他寫道：“不管計算機的速度有多快，單純的歸納啓發技術，比如機器學習，還是不夠的。像 Watson 這樣的混合系統，也達不到一般的理解，”“在一個開放的場景，如"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2021\/07\/12\/linguistics-for-the-age-of-ai\/","title":null,"type":null},"content":[{"type":"text","text":"語言理解"}]},{"type":"text","text":"，這需要關於世界的知識，而溯因是核心和不可替代的。正因爲如此，將演繹與歸納策略相結合的嘗試總是註定要失敗……這一領域需要一個關於溯因的基礎理論。與此同時，我們卻被困在了陷阱裏。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"人工智能的商業化"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人工智能社區對數據驅動方法的狹隘關注，使得研究和創新集中於那些擁有"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2021\/05\/31\/microsoft-gpt-3-and-the-future-of-openai\/","title":null,"type":null},"content":[{"type":"text","text":"海量數據存儲和雄厚財力"}]},{"type":"text","text":"的少數組織。當深度學習成爲一種將數據轉換成有利可圖的產品的有效方式，大型科技公司正陷入僱傭人工智能人才的激烈競爭，他們通過給研究人員提供豐厚的薪酬，使得他們離開學術界。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這一轉變使得非營利性實驗室和小公司難以參與到人工智能的研究。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 說：“當你將人工智能研發與大型數據集的所有權和控制權聯繫在一起時，初創企業就會面臨進入壁壘，因爲他們並不擁有數據。”他補充說，數據驅動的人工智能從本質上講就是在商業領域創造了“贏家通喫”的局面。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"反過來，人工智能的壟斷也阻礙了科學研究。由於大型科技公司集中精力創造可以利用其巨大的數據資源以保持競爭優勢的應用，沒有什麼動力去探索其他人工智能方法。這一領域的工作開始傾向於狹隘而有利可圖的應用，而忽視了可以帶來新發明的努力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“目前沒有人知道如果沒有如此龐大的集中式數據集的情況下，人工智能將會是什麼樣，因此，對於那些想要通過設計出不同、更強大的人工智能來競爭的企業家來說，沒有什麼真正的機會。”Larson 說。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 在他的書中警告了當前的人工智能文化，“它在不斷編織人工智能神話的同時，從低垂的果實中榨取利潤。”他寫道，通用人工智能進展的假象可能會導致另一個"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2018\/11\/12\/artificial-intelligence-winter-history\/","title":null,"type":null},"content":[{"type":"text","text":"人工智能的冬天"}]},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不過，儘管人工智能的冬天可能會抑制人們對深度學習和數據驅動的人工智能的興趣，但它也能爲新一代的思想家探索新的途徑開闢道路。Larson 希望科學家們開始超越現有的方法。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Larson 在《人工智能的神話》一書中提供了一種推理框架，它揭示了當今該領域所面臨的挑戰，並幫助讀者看穿通用人工智能或"},{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2020\/06\/29\/artificial-intelligence-singularity\/","title":null,"type":null},"content":[{"type":"text","text":"奇點"}]},{"type":"text","text":"進展的誇大說法。“我的希望是，非專業人士有一些工具與這種非科學的必然想法作鬥爭，我的同事和其他人工智能科學家將此視爲解決這一領域所面臨的現實問題的警鐘。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"作者介紹："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Ben Dickson，軟件工程師，也是 TechTalks 的創始人，作家，撰寫關於科技、商業和政治方面的文章。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"原文鏈接："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/bdtechtalks.com\/2021\/09\/20\/myth-of-artificial-intelligence-erik-larson\/","title":null,"type":null},"content":[{"type":"text","text":"https:\/\/bdtechtalks.com\/2021\/09\/20\/myth-of-artificial-intelligence-erik-larson\/"}]}]}]}