Almost Human?
聊天機器人ChatGPT 撼動人工智慧界
AI Program Can Write Stories, Debate and Much More
TOM FOREMAN, CNN CORRESPONDENT
Allowing cars to drive themselves, composing songs that mimic popular artist[s] and producing this digital painting that took the top award at a Colorado art show: this is all the work of artificial intelligence—computers that don’t just do what they’re told, but in a sense, think, learn and create.
And right now, ChatGPT is rattling the AI world, turning out stunningly humanoid writing. Just ask Douglas Rushkoff, a renowned author and professor of media culture.
DOUGLAS RUSHKOFF, MEDIA THEORIST
It is writing better than most of my students write at this point—you know, they’re college freshmen. So, yeah, I am impressed with that.
TOM FOREMAN, CNN CORRESPONDENT
How does it work? ChatGPT has been filled, in a sense, with a massive amount of information—imagine the biggest library you can—then programmed and trained by humans to process and spit it out in conversational phrases.
So, ask for 1,000 words on the early days of automobiles, and in seconds, it responds: “In the late 1800s and early 1900s, automobiles were relatively primitive by today’s standards and were primarily used by wealthy individuals or businesses.”
Ask it to write a sonnet in the style of Jerry Seinfeld. “I’m just a stand-up comic, telling jokes. Onstage, I make ’em laugh, and that’s all I do. But sometimes life’s a joke. It hits me low. And then I take the mic and say ‘who knew?’”
It’s not perfect, but it can debate, compose essays, solve math problems …
Well, that looks right.
… write computer code, answer follow-up questions [and] even admit mistakes. And all that means ChatGPT, or more advanced AI like it, could replace people in all sorts of positions.
DOUGLAS RUSHKOFF, MEDIA THEORIST
This could potentially save time and resources, but it could also lead to a loss of personal connections and a decline in the quality of these types of interactions.
TOM FOREMAN, CNN CORRESPONDENT
We know that because everything Rushkoff said just there was written by ChatGPT when asked about potential problems with itself.
DOUGLAS RUSHKOFF, MEDIA THEORIST
The answer it gave me about the dangers of [Chat]GPT—that sounded like a pretty good television guest to me, you know.
CNN特派員 湯姆.佛曼
讓汽車可以自行駕駛、寫出模仿流行藝人的歌以及創作這幅奪得科羅拉多一項藝術展首獎的數位畫作:這全都是人工智慧的成果──人工智慧是不只做它們被要求的事,而是在某種程度上會思考、學習及創作的電腦。
如今,ChatGPT正撼動著人工智慧界,產出驚人的擬人寫作。問問道格拉斯.洛許考夫就知道,他是一位知名作家暨媒體文化教授。
媒體理論家 道格拉斯.洛許考夫
現階段它寫得比我大部分學生寫得還要好──你懂的,他們是大學新鮮人。所以,對啊,那讓我感到驚豔。
CNN特派員 湯姆.佛曼
它是怎麼運作的呢?在某種程度上,ChatGPT被填入大量的資訊──你能想像的最大的圖書館──然後被人類進行程式設計及訓練,而得以處理和用對話式的詞語表達出來。
那麼,要它針對汽車的早期發展寫一千字,在幾秒鐘之內,它回覆了:「在十九世紀末和二十世紀初,以當今的標準來看,汽車是相對原始的,而且主要是富有的個人或企業在使用。」
要它用傑瑞.賽恩菲爾德的風格寫一首十四行詩。「我只是個脫口秀喜劇演員在講笑話。在舞台上,我讓他們笑,那就是我做的一切。不過有時人生是個笑話。它把我打趴。接著我拿起麥克風,說:『誰知道呢?』」
它不完美,不過它會辯論、寫論文、解數學問題……
嗯,那看起來是對的。
……寫電腦程式碼、回答後續問題,甚至承認錯誤。那一切都意味著ChatGPT或更多類似它的先進人工智慧可能會在各種職位上取代人類。
媒體理論家 道格拉斯.洛許考夫
這可能可以節省時間和資源,不過也可能造成人際連結的流失,以及這些類型互動品質的下降。
CNN特派員 湯姆.佛曼
我們知道那一點,因為洛許考夫那邊剛剛說的一切都是ChatGPT被問到它自身的潛在問題時它寫的。
媒體理論家 道格拉斯.洛許考夫
關於ChatGPT的危險,它給我的答案──在我聽來,那像是一個很不錯的電視來賓,你懂嗎。