幾十年前，美國神經科學家卡爾·拉什利就開始找尋行蹤不定的記憶軌跡。確切地說，他想了解，大腦中是不是有類似保險箱的地方儲存每段記憶？大腦皮層上有有專門的存儲記憶區(qū)域，可以即時提取？

為了解開這個謎，拉什利訓練一群老鼠走迷宮，把食物放在迷宮盡頭刺激老鼠找到重點。然后，科學實驗嘛，他的辦法是分別切除每只老鼠大腦皮層中的不同區(qū)域，康復后再走迷宮。拉什利想借此找出老鼠腦中哪個區(qū)域負責記住走出迷宮的路。結果他得到了驚人的發(fā)現：無論他切除哪塊區(qū)域（或者哪個區(qū)域因為病變而損傷），老鼠都能找到食物。真正影響結果的不是切除哪個部位，而是切掉多少。實際上，只要老鼠熟悉了路徑，哪怕腦部失去一大塊灰質，仍然能順利走完迷宮。

拉什利得到這樣一個結論：記憶并不存儲在任何一處神經元組成的保險箱里，某種意義上說，記憶無影無形卻又無處不在，以不為人知的方式分布在大腦內部。

事實證明，拉什利是對的，只是稍微有點錯：功能性磁共振成像等測試顯示，記憶在大腦內部的確分區(qū)域存儲。不過，時至今日人們還是相信記憶是非常復雜的集合功能。

其實記憶的復雜之處以及難于明確和界定的難點在于：記憶到底有多“鮮活”。記憶決定了我們是誰?！皩θ祟愑洃浀难芯孔顬榻咏到y(tǒng)研究心靈?！敝杏洃浄矫娼滩牡氖ツ复髮W教授加百利·雷萬斯基說。在題為《神經學未能解決的問題》的精彩文章中，加州理工學院教授拉爾夫·阿道夫采用了另一種表述方式：“記憶可能是通過學習預知未來的能力?！?/p>

這正是深度學習的一大挑戰(zhàn)。在10月1日《財富》刊發(fā)的文章中，我的同事羅杰·帕羅夫也詳細寫過這個問題，見解深刻。雖然目前無解，但深度學習終將解開的人類生理和身心健康之謎，包括衰老后記憶變差的殘酷現實。（財富中文網）

譯者：Pessy

審校：夏林

Decades ago American neuroscientist Karl Lashley went looking for the elusive engram: the trace of memory. Where precisely, he wanted to know, was the safe deposit box that held each memory in the brain? Was there a specific spot in the cortex that stored this precious data and kept it ready for retrieval at a moment’s notice?

To find out, he trained rats to complete a maze, drawing them to the end with a food reward. Then—science being science—he methodically snipped out discrete parts of the animals’ cortical tissue and, once healed, had them run the maze again. The aim was to see if, by process of elimination, he could discover which piece of a rat’s brain held the memory of the correct path. What he found was surprising: No matter which part he surgically removed (or destroyed by lesion), the rats still found the food. What mattered wasn’t where he cut, but rather how much. And indeed, the rodents could lose a substantial portion of their gray matter and still complete the maze, once it was learned.

Among his many conclusions: Memory isn’t held in any one storage locker of neurons; in a sense, it is both nowhere and everywhere, distributed across the brain in untold ways.

Lashley was right, it turns out, and a little wrong: We know from functional MRI studies, among other tests, that memory does have a regional component, too. But the idea that memory is a phenomenally complex collective function remains with us today.

Indeed, its complexity—and difficulty to pinpoint or characterize—speaks to how “alive” memory is. Our memories are what make us who we are. “The study of human memory is the closest one can get to a systematic study of the human soul,” writes Gabriel Radvansky, a professor at Notre Dame who has authored textbooks on memory. Caltech professor Ralph Adolphs frames it another way in a fascinating article entitled “The Unsolved Problems of Neuroscience”: “Memory may be the ability to predict the future by learning.”

This is where a big challenge for deep learning comes in, a topic on which my colleague Roger Parloff wrote about eloquently and insightfully in Fortune’s October 1 issue. Ultimately, deep learning will enable us to solve riddles of biology and well-being that may now seem just out of reach—including the sad problem of failing memory as we age.