《那些古怪又讓人憂心的問題》第29期:人力計算機(4)
Why this is ridiculous
These two ways of benchmarking the brain represent opposite ends of a spectrum. One, the pencil-and-paper Dhrystone benchmark, asks humans to manually simulate individual operations on a computer chip, and finds humans perform about 0.01 MIPS. The other, the supercomputer neuron simulation project, asks computers to simulate individual neurons firing in a human brain, and finds humans perform about the equivalent of 50,000,000,000 MIPS. A slightly better approach might be to combine the two estimates. This actually makes a strange sort of sense. If we assume our computer programs are about as inefficient at simulating human brain activity as human brains are at simulating computer chip activity, then maybe a more fair brain power rating would be the geometric mean of the two numbers. The combined figure suggests human brains clock in at about 30,000 MIPS-right about on par with the computer on which I'm typing these words. It also suggests that the year when Earth's digital complexity overtook its human neurological complexity was 2004.
這兩種不同的人腦基準測試得出了兩個完全相反的結論。第一個紙筆基準測試要求人類模擬計算機芯片上執行的單個指令，得出的結果爲人腦的得分僅爲0.01 MIPS左右。第二個超級計算機神經元模擬項目讓計算機模擬人類大腦中單個突觸的行爲，得出的結果爲人腦得分高達500億MIPS。稍微好一些的做法是把兩個結果合併在一起，但還是感覺怪怪的。如果我們認爲計算機程序模擬人腦和人腦模擬計算機芯片的行爲都一樣不利索，那麼稍微公平一點，人腦基準結果也許是這兩個數字的幾何平均值。這樣得到的結果是人腦的執行效率約爲3萬MIPS，差不多和我現在正在打字用的計算機性能是一個水平。這同時也說明全球計算機的總計算能力在2004年就已經超過所有人類的總計算能力了。Ants
螞蟻
In his paper “Moore's Law at 40,” Gordon Moore makes an interesting observation. He points out that, according to biologist E. O. Wilson, there are 1015 to 1016 ants in the world. By comparison, in 2014 there were about 1020 transistors in the world, or tens of thousands of transistors per ant. An ant's brain might contain a quarter of a million neurons, and thousands of synapses per neuron, which suggests that the world's ant brains have a combined complexity similar to that of the world's human brains. So we shouldn't worry too much about when computers will catch up with us in complexity. After all, we've caught up to ants, and they don't seem too concerned. Sure, we seem like we've taken over the planet, but if I had to bet on which one of us would still be around in a million years-primates, computers, or ants-I know who I'd pick.
戈登•摩爾在《摩爾定律邁入40週年》一文中提出了一個很有意思的發現。他指出，根據生物學家E．O．威爾遜的說法，全世界有1015～1016只螞蟻。相比之下，2014年全世界一共有約1020個晶體管，也就是說平攤下來每隻螞蟻能分到幾萬個晶體管。螞蟻的大腦可能有25萬個神經元，每個神經元上又有幾千個突觸，這意味着全世界所有螞蟻大腦的總複雜度已經和所有人類大腦的總複雜度相當。所以我們沒必要太在意什麼時候計算機會在複雜度上擊敗我們。畢竟，我們追上了螞蟻，但螞蟻一點也沒着急嘛。當然了，雖然我們看上去現在主宰了地球，但如果一定要我從靈長類動物、計算機和螞蟻之中選出一個能在幾百萬年後依然存在的東西的話，我當然知道該選哪個。1 Except Red Delicious apples, whose misleading name is a travesty.
1． 除了蛇果，這玩意兒的名字真是坑人。2 Our house had a lot of vases when I was a kid.
2． 我小時候家裏有許多花瓶。3 Yet.
3． 到目前爲止。4 This figure comes from a list in Hans Moravec's book Robot: Mere Machine to Transcendent Mind.
4． 這個數字來自漢斯•莫拉維克撰寫的《機器人：由機器邁向超越人類心智之路》中的一個列表。5 Although even this might not capture everything that's going on. Biology is tricky.
5．即使是這樣也沒法完全精確地模擬每一個細節，生物學從來都不是這麼簡單的。6 Using 82,944 processors with about 750 million transistors each, K spent 40 minutes simulating one second of brain activity in a brain with 1 percent of the number of connections as a human's.
6．每臺“京”超級計算機配備了82944個處理器以及7.5億個晶體管，連接數量相當於人類大腦的1％，它需要花40分鐘才能模擬出人類大腦僅用時一秒的活動。7 If it's past the year 2036 right now while you're reading this, hello from the distant past! I hope things are better in the future. P.S. Please figure out a way to come get us.
7．如果你讀到這篇文章的時候已經過了2036年，那我在這裏給你打一個來自遙遠過去的招呼！我希望未來科技會更加進步。對了，你們快找個方法回來接我們啊！8 “TPA.”
TPA：每隻螞蟻能分到的晶體管數目。
