2023年经济学人 新的联想学习模型与其对人工智能的影响(1)

 

    A model of learning that is decades-old is under fire with implications for AI.

    一种已有数十年历史的学习模式正受到抨击，并对人工智能产生影响。

    The buzz of a notification or the ding of an email might inspire excitement - or dread¹.

    通知的嗡嗡声或电子邮件的叮当声可能会让人们感到兴奋或恐惧。

    In a famous experiment, Ivan Pavlov showed that dogs can be taught to salivate at the tick of a metronome or the sound of a harmonium.

    伊万·巴甫洛夫进行的一项著名实验证明，通过学习，狗会因为节拍器的滴答声或铃声流口水。

    This connection of cause to effect - known as associative, or reinforcement learning - is central to how most animals deal with the world.

    这种因果关系——被称为联想学习，或强化学习——是大多数动物行为的核心。

    Since the early 1970s the dominant² theory of what is going on has been that animals learn by trial and error.

    自20世纪70年代初以来，关于这种因果关系存在一种主要理论，认为动物的学习行为是通过“尝试与错误”实现的。

    Associating a cue (a metronome) with a reward (food) happens as follows.

    将信号（节拍器）与奖励（食物）相关联的过程如下。

    When a cue comes, the animal predicts when the reward will occur.

    信号出现时，动物会预测奖励何时出现。

    Then, it waits to see what arrives.

    然后，它会等待接下来发生的事件。

    After that, it computes³ the difference between prediction and result - the error.

    之后，它会计算预测和结果（误差）之间的不同。

    Finally, it uses that error estimate to update things to make better predictions in future.

    最后，它利用误差估计进行调整，以便在未来做出更好的预测。

    Belief in this approach was itself reinforced in the late 20th century by two things.

    20世纪末，有两件事加强了人们对这种方法的信任。

    One of these was the discovery that it is also good at solving engineering problems related to artificial intelligence (AI).

    其中之一，是人们发现这种方法也能够有效解决与人工智能（AI）相关的工程问题。

    Deep neural⁴ networks learn by minimizing the error in their predictions.

    深度神经网络是以“将预测误差最小化”为依据的一种机器学习技术。

    The other reinforcing observation was a paper published in Science in 1997.

    另一件事是1997年发表在《科学》杂志上的一篇论文。

    It noted⁵ that fluctuations⁶ in levels in the brain of dopamine, a chemical which carries signals between some nerve cells and was known to be associated with the experience of reward, looked like prediction-error signals.

    这篇论文指出，大脑中多巴胺水平的波动看起来像是预测-误差的信号。多巴胺是一种在神经细胞之间传递信号的化学物质，与大脑的奖励机制有关。

    Dopamine-generating cells are more active when the reward comes sooner than expected or is not expected at all, and are inhibited⁷ when the reward comes later or not at all - precisely⁸ what would happen if they were indeed such signals.

    如果奖励比预期来得更早或超出预期，产生多巴胺的细胞会更活跃，而如果奖励来得更晚或根本没有奖励，产生多巴胺的细胞就会受到抑制——如果这些细胞真的是信号，这就是大脑中会发生的事情。

    A nice story, then, of how science works.

    这是一个关于科学如何运作的好故事。

    But if a new paper, also published in Science, turns out to be correct, it is wrong.

    但是，如果同样发表在《科学》杂志上的一篇新论文被证明是正确的，那么它就是错误的。

    Researchers have known for a while that some aspects of dopamine activity are inconsistent with the prediction-error model.

    研究人员早就发现，多巴胺活动的某些方面与预测-误差模型不一致。

    But, in part because it works so well for training artificial agents, these problems have been swept under the carpet.

    但是，因为它在训练人工智能方面效果太好，这些问题在某种程度上被掩盖了。

    Until now.

    直到现在。

    The new study, by Huijeong Jeong and Vijay Namboodiri of the University of California, San Francisco, and a team of collaborators, has turned the world of neuroscience on its head.

    由加州大学旧金山分校的郑慧贞和维贾伊·南博迪里以及一个合作团队完成的这项新研究，彻底颠覆了神经科学世界。

    It proposes a model of associative learning which suggests that researchers have got things backwards⁹.

    这项研究提出了一个联想学习模型，认为研究人员把事情搞反了。

    Their suggestion, moreover, is supported by an array of experiments.

    此外，他们的新想法还得到了一系列实验的支持。

    The old model looks forward, associating cause with effect.

    之前的学习模型着眼于未来，用原因推断结果。

    The new one does the opposite.

    而新的学习模型正好相反。

    It associates effect with cause.

    它用结果反推原因。

    They think that when an animal receives a reward (or punishment), it looks back through its memory to work out what might have prompted this event.

    他们认为，当一只动物受到奖励（或惩罚）时，它会回顾自己的记忆，找出可能导致这一事件的原因。

    Dopamine's role in the model is to flag events meaningful enough to act as causes for possible future rewards or punishments.

    多巴胺在该模型中的作用是标记足够有意义的事件，作为未来可能获得奖励或惩罚的原因。

1 dread   

vt.担忧，忧虑；惧怕，不敢；n.担忧，畏惧

参考例句：

• We all dread to think what will happen if the company closes.我们都不敢去想一旦公司关门我们该怎么办。
• Her heart was relieved of its blankest dread.她极度恐惧的心理消除了。

2 dominant   

adj.支配的，统治的；占优势的；显性的；n.主因，要素，主要的人(或物)；显性基因

参考例句：

• The British were formerly dominant in India.英国人从前统治印度。
• She was a dominant figure in the French film industry.她在法国电影界是个举足轻重的人物。

3 computes   

v.计算，估算( compute的第三人称单数 )

参考例句：

• This article also computes STL model volume and research simply model complexity. 文章还评估了STL实体体积，对模型的复杂度进行了研究。 来自互联网
• This macro computes the length of the day, from sunrise to sunset. 把根据要求确定出来的日出日没间隔作为我们通常直觉上的一个白天。 来自互联网

4 neural   

adj.神经的，神经系统的

参考例句：

• The neural network can preferably solve the non- linear problem.利用神经网络建模可以较好地解决非线性问题。
• The information transmission in neural system depends on neurotransmitters.信息传递的神经途径有赖于神经递质。

5 noted   

adj.著名的，知名的

参考例句：

• The local hotel is noted for its good table.当地的那家酒店以餐食精美而著称。
• Jim is noted for arriving late for work.吉姆上班迟到出了名。

6 fluctuations   

波动，涨落，起伏( fluctuation的名词复数 )

参考例句：

• He showed the price fluctuations in a statistical table. 他用统计表显示价格的波动。
• There were so many unpredictable fluctuations on the Stock Exchange. 股票市场瞬息万变。

7 inhibited   

a.拘谨的，拘束的

参考例句：

• Boys are often more inhibited than girls about discussing their problems. 男孩子往往不如女孩子敢于谈论自己的问题。
• Having been laughed at for his lameness,the boy became shy and inhibited. 那男孩因跛脚被人讥笑，变得羞怯而压抑。

8 precisely   

adv.恰好，正好，精确地，细致地

参考例句：

• It's precisely that sort of slick sales-talk that I mistrust.我不相信的正是那种油腔滑调的推销宣传。
• The man adjusted very precisely.那个人调得很准。

9 backwards   

adv.往回地，向原处，倒，相反，前后倒置地

参考例句：

• He turned on the light and began to pace backwards and forwards.他打开电灯并开始走来走去。
• All the girls fell over backwards to get the party ready.姑娘们迫不及待地为聚会做准备。