The challenge: Figure out why brains are so good at learning, and use that information to design computer systems that can interpret, analyze, and learn information as successfully as humans.
人工智慧太厲害了，我們該怎麼辦?
我們為什麼要推計算思維呢? 因為未來各行各業都需要與電腦合作，否則有可能被電腦和機器人淘汰，例如這篇【機器人搶工作 律師、藥劑師也遭殃】所談到的(註1)狀況。如果不懂計算思維，很容易就迷惘了。
最近像這樣的文章和書籍很多，研究未來學的人，認為人工智慧是未來的重要趨勢，極盡能事去想像未來，但究竟有多少真實會發生，有多少只是虛無飄渺的幻想? 我想，很少人有能力確定，不過當前許多學生都跑來研究人工智慧相關的議題，則已成為我在台大所看到的事實。
我三十年前在高中時，就對於人工智慧很感興趣，開始學LISP，後來進到台大念電機系，還是修了兩門人工智慧的課，也旁聽過神經網路，到密西根大學念書，也修過人工智慧，但我沒有繼續研究人工智慧，因為我覺得當時研究者走偏了，而且運算速度遠遠不足以支持有意義的人工智慧，所以根本做不出東西。
我猜對了，1990年代之後，人工智慧成為票房毒藥，沉寂了二十年。
我做電腦系統，看著電腦系統的效能持續成長，電腦系統的研究者想出各種方法來收割(harvest)不斷成長的電腦效能，過去這三十年，最忙碌的研究領域之二，是計算機結構和系統軟體，我有幸能優游於兩者之間，探討一些軟硬體整合的議題。
如今，單一處理機的運算能力，約為30年前人工智慧全盛時期的100萬倍(註2)，而且只要願意付些許錢，就可以租用雲端的上百台電腦，運算能力更是30年前的一億倍以上。
要注意到，這一億倍的運算效能，是人工智慧東山再起的關鍵。沒有足夠的效能，電腦很難生出智慧。如今的運算效能，是否足以支持未來學想像中的人工智慧，就是一個大哉問。
大部份未來學專家的預測，都是基於摩爾定律(註3)，但這幾年摩爾定律已經放緩，甚至有可能停滯，主要是成本考量。以往這麼多年透過個人電腦、電子商務、行動運算、雲端服務等應用，半導體產業有足夠的利潤做研發來支撐摩爾定律，但大數據分析和人工智慧是否足以繼續支撐摩爾定律? 如果摩爾定律停滯，那該如何是好?
有的人工智慧應用，需要比目前更高百倍的計算能力，有的要成為產品的前提，需要將龐大的運算能力縮小進到生活周邊，因此我認為我們做計算系統的，在產品化的過程中，還是扮演舉足輕重的角色，將來應該會有做不完的人工智慧系統設計的工作。
要創造出人工智慧的系統，關鍵在於要有能夠密切垂直整合的團隊，必須要有三種專家密切配合:
(1)領域專家，例如找律師、藥劑師來指導或教導電腦該領域的專業技能。
(2)人工智慧專家，綜合運用機器學習、數據分析、資料探勘等方式設計人工智慧。
(3)系統專家，提供人工智慧所需的系統整合、資料蒐集、處理和計算能力。
(1)領域專家，例如找律師、藥劑師來指導或教導電腦該領域的專業技能。
(2)人工智慧專家，綜合運用機器學習、數據分析、資料探勘等方式設計人工智慧。
(3)系統專家，提供人工智慧所需的系統整合、資料蒐集、處理和計算能力。
台灣比諸於其他許多國家，由於有硬體產業的基礎，非常適合發展「(3)系統專家」，加上台灣目前很多學生對人工智慧很有興趣，學得很快，所以我不擔心會短缺「(2)人工智慧專家」，台灣在各行各業也有很多領域專家，但是能否聚集人才成為優質研發團隊，是真正的重點。
我想，很多有識之士已經看到這個局面，這是值得台灣去發展的好機會。我希望國家和社會多投入一些資源鼓勵產學界共同組成「對」的團隊，來把握這樣的機會，讓學術界多做些有益於這類幫助國家產業發展的研發工作。
然而，在謀求發展的機會的同時，我們也應該做好教育的工作，讓未來的世代能夠好好面對電腦和機器人。與其教學生背誦記憶一堆電腦瞬間可解的問題，不如教他們如何活用電腦、想辦法與電腦和機器人共榮。
另外，科技的民主化以及財富的合理分配，也將會是越來越重要的課題。我們絕對不希望大家多年努力的成果，被少數資本家收割，讓科技成為資本家搜刮社會資源和剝奪勞工的打手 -- 這是社會大眾需要慎重看待的議題。
(註1)機器人搶工作 律師、藥劑師也遭殃
http://www.cw.com.tw/article/article.action?id=5073792
(註2)以摩爾定律概算，每18個月電腦效能增加一倍。
(註3)https://zh.wikipedia.org/wiki/摩爾定律
http://www.cw.com.tw/article/article.action?id=5073792
(註2)以摩爾定律概算，每18個月電腦效能增加一倍。
(註3)https://zh.wikipedia.org/wiki/摩爾定律
*******
待查：Simon 是否在1990年代提過一套"心理治療"軟體？
A Computer That Can Hear a Marriage in Trouble
Researchers try to harness technology to help therapists better help struggling couples
By programming a computer to analyze the speech of couples, researchers at the universities of Southern California and Utah could predict whether the relationship would improve, worsen or stay the same. ILLUSTRATION: JUSTIN RENTERIA
The human voice can reveal a great deal. Now, with the help of a computer, it can probably reveal whether your marriage is deteriorating.
By programming a computer to analyze the speech of couples, researchers at the universities of Southern California and Utah could predict whether the relationship would improve, worsen or stay the same. The computerized analysis, which focused entirely on aural qualities such as pitch and intensity, was compared with human assessments that took account of familiar features of the marital landscape, such as blame.
The computer turned out to be able to predict marital improvement or deterioration about as reliably as ratings provided by trained humans—in fact, even a little better.
The work is part of a flurry of research in recent years in which scientists have tried to glean useful information from closely examining therapy patients’ voices, gestures and word choices. The aim is to make talk therapy more effective.
In this case, the scientists worked with video recordings of 134 “chronically distressed couples” who had been married for an average of 10 years and had sought therapy for problems in the relationship. The study focused on three sets of sessions: one before therapy began, another after 26 weeks of therapy and a third after two years of treatment.
The researchers used their computer to rate the recorded voices for 74 acoustic features. These included such familiar ones as loudness but also more esoteric elements such as jitter and shimmer—described by Brian Baucom, one of the scientists, as measures of shakiness. The couples’ videotaped interactions—including words and body language—were also rated for various characteristics by teams of undergraduate psychology students who were extensively trained for the purpose.
Dr. Baucom says that prior research has shown that trained students can do better here than therapists because the students tend to follow the rating manual and rate more consistently for behaviors and feelings such as blame and sadness. Their ratings were then correlated with four possible outcomes from therapy: decline, no change, partial recovery and recovery. These correlations have predictive power, letting researchers anticipate actual outcomes for the couples (which were assessed two years after therapy ended).
But the computer did a bit better across the board in predicting marital changes. In predicting whether a recovery would occur, for instance, the computer achieved nearly 78% accuracy, beating the human ratings by two percentage points.
One big question: Can couples learn—perhaps from a smartphone app—to change their voices in a way that improves their marriage? Perhaps a stand-alone device, placed in the home, could sound an alarm when dialogue takes on troubling tones. Shrikanth S. Narayanan, another scientist who worked on the study, says that it’s conceivable. But the focus for now is simply to harness technology to help therapists do a better job in counseling couples.
“Still Together?: The Role of Acoustic Features in Predicting Marital Outcome,” Md Nasir, Wei Xia, Bo Xiao, Brian Baucom, Shrikanth S. Narayanan, Panayiotis Georgiou, Interspeech 2015 (Sept. 6)
