2017年12月14日 星期四

An 8th Planet Is Found Orbiting a Distant Star, With A.I.’s Help

An 8th Planet Is Found Orbiting a Distant Star, With A.I.’s Help

Trilobites By NICHOLAS ST. FLEUR
DEC. 14, 2017

With eight planets whirling around its sun, our solar system has held the galactic title for having the most known planets of any star system in the Milky Way. But on Thursday NASA announced the discovery of a new exoplanet orbiting a distant star some 2,500 light years away from here called Kepler 90, bringing that system’s total to eight planets as well.

The new planet, known as Kepler-90i, is rocky and hot. It orbits its star about once every 14 days. The finding was made using data collected by NASA’s Kepler Space Telescope, a planet hunter that has spotted more than 2,500 confirmed exoplanets since its launch in 2009. Unlike those previous discoveries, the new exoplanet was detected with the help of an artificial intelligence researcher at Google using a machine learning technique called neural networking.

“This is the first time a neural network specifically has been used to identify a new exoplanet,” said Christopher Shallue, a software engineer at Google who helped make the finding. The technology, which is loosely inspired by the human brain, is designed to recognize patterns and classify images. It can learn to tell the difference between something simple like a cat and a dog, and also to distinguish exoplanets from cosmic noise.

For the project, the computer looked at a small chunk of data gathered by Kepler from 2009 to 2013. Of the 150,000 stars represented in Kepler’s collection, the computer combed through 670 star systems for signs of exoplanets. Astronomers spot exoplanets when the celestial bodies move, or transit, in front of their stars. The interaction causes a dip in brightness that creates a detectable signal.

So far, the data set has about 35,000 such signals. The astronomers trained the program on a set of about 15,000 signals, and it identified planets correctly 96 percent of the time. The neural network learned what was a planet and what was not a planet and was able to find the exoplanet Kepler-90i, as well as a second exoplanet named Kepler-80g around a different star system.

Next, the researchers plan to explore more star systems studied by Kepler. “We plan to search all 150,000 stars in the Kepler data system,” said Mr. Shallue.

Andrew Vanderburg, an astronomer at the University of Texas, Austin, said that Kepler-90i is about 30 percent larger than Earth and about as hot as the planet Mercury, reaching about 800 degrees Fahrenheit. Like the other seven planets in its system, it is packed close to its star. It resembles a miniature version of our solar system, he said, where the most distant known planet is about as far away from its star as the Earth is from our sun. But there could be additional, more distant planets not yet detected because planets close to their stars may be easier for astronomers to find.

Seth Shostak an astronomer with the SETI Institute in Mountain View, Calif., who was not involved in the project said the finding that Kepler 90 has eight planets shows that our solar system is “just another duck in a row.” “The bad news is we’re not quite as special as we thought we were,” he said. “But the good news is we may have a lot of cosmic company.”

It’s possible that the two systems may not be tied for long as astronomers search the outer reaches of our solar system for the elusive Planet Nine. It sets the stage for a new space race: Which team will break the intragalactic deadlock? Will artificial intelligence first detect another planet in the Kepler-90 system, or will astronomers find a distant ninth planet orbiting our sun? “It’s kind of cool to see which one will be proven next,” said Jessie Dotson, Kepler’s project scientist at NASA.


2017年12月11日 星期一

AI時代的職場生存條件

林建甫》AI時代的職場生存條件

20171211
林建甫

最近人工智慧AI的發展,一日千里。現代人的一個痛點,恐怕都要擔心未來的工作會不會被取代掉。國內電子零組件大廠正崴日前表示將強化生產技術,建立自動化產線,提升產品生產良率,預估5年內人力減半。不約而同地,統一企業也證實7-ELEVEN將應用新科技開設第1家無人超商,最快明年上半年就會問世。而全世界金融機構的分行都在裁撤,因為電子轉帳、支付都不再臨櫃,理財真的交給機器人就可以了。他們不收佣金、誠實、沒有情緒,完全理性,表現可能更值得信賴。

AI議題真正引起大家重視的是在1997年時,IBM的深藍(DeepBlue)戰勝西洋棋世界冠軍後,當時的《時代》(Times)雜誌還認為電腦要在更複雜的圍棋上戰勝人類,可能還要再過100年,甚至更長的時間。未料,谷歌的深智(DeepMind)團隊開發的AI圍棋程式AlphaGo,利用深度學習與強化學習,自2016年起擊敗多位世界級的職業圍棋棋士,只用了原本《時代》雜誌預估100年的1/5時間。稍後再度進化的AlphaGo Zero不依靠人類玩家的數據創建,僅透過自我對弈,幾天之內它就發展出擊敗人類頂尖棋手的技能,對比AlphaGo要達到同等水平則需要數月的訓練。

就在前幾天,深智團隊再將AlphaGo中代表圍棋的Go去掉,成為AlphaZero,標榜是通用棋類的人工智慧程式,可以從零自學任何棋藝:圍棋、西洋棋、日本將棋…,其表現已經擊敗AlphaGo Zero舉世現在希望利用AlphaZero研究重大疾病,盼治癒人類數百年來找不到療法的疾病,包括阿茲海默症、帕金森氏症、囊狀纖維症等。

AI發展到今天,最基本的就是電腦的運算跟儲存的進步。近幾年移動裝置本身的儲存、運算與聯網的雲端資訊已經徹底改變我們的生活。這裡面的一個關鍵是演算法的改良及大數據的分析。演算法是利用電腦算數學的學問,是求解的利器,它近年利用類神經網路的原理,產生更有效的輸入跟輸出,得到仿人類的學習,因此可以很快得到準確的結果;而大數據的獲得是利用日益發展的感測裝置與物聯網,收集源源不絕傳統的數字,或非傳統的影像、聲音再萃取其特徵值,有的甚至進入雲端進行儲存及後繼的運算分析。

最新手機的刷臉,視覺辨識燈光打下偵測臉部3D形體的數據,一次就高達上萬筆做立即的計算。而無人自動駕駛車,靠前後左右的鏡頭來感應周遭環境及聯網取得道路資訊,綜合迅速運算來控制行車。

人類輸了棋,其實也不用覺得沮喪。因為人不可能跑贏馬,這是人類天生的限制。我們不可能比電腦記憶容量更大、運算速度更快,也沒有永遠不出錯的精準與不衰退的記憶。人類重要的是要懂得駕馭電腦,人機能協作,還有做好決策與發展創意,精進對美的欣賞與情感的表達,這都是機器所沒辦法取代的。

回到工作的問題。《日本經濟新聞》和英國《金融時報》(Financial Times)今年7月的共同調查指出,在820種職業、2069項工作中,約有34%(約710項工作)的工作可被機器人替代。以勞工為例,製造業77項工作中,高達75%可以自動化;食品加工業則是很有危機,所有工作都可由機械技術代替。至於金融業60項工作中,包括建立檔案與整理數據等,65%工作可自動化。另外,公司高層需處理的63項工作任務中,管理決策與領導魅力無法被取代,只有22%如製作業績報告可由機器代替。至於畫家、演員、音樂家及其他藝術相關職業,65項工作中,只有17%可以應用機械人技術。

總之,在人工智慧AI時代,我們不要再期望「安穩」的工作。因為「安穩」意味著簡單、重複,這些僅靠記憶與練習就可以掌握的工作,將最沒有價值,最快被機器取代。我們要選擇那些具有「不穩定且多變化」的工作。這些工作的背後是人性,是需要創造或管理能力。因此培養創造力、經常學習新知,能旁徵博引、綜合分析與正確下決策是不被淘汰的不二法門。


(作者為台灣經濟研究院院長、國立台灣大學經濟系教授)