Distributed anatomical circuits for decision-making, inference, and learning

每天，我们都在为下一步做什么做决定。我们这样做是因为我们一直在监控和评估事情的进展情况。因此，我们调整自己的行为，使之适合当前的环境，或者我们决定采取一种行动，而不是另一种。有时我们意识到做出这些评估和决定，但通常我们不知道。然而，无论我们多么认为这些能力是理所当然的，当它们在抑郁症等心理疾病中发生变化时，令人震惊。这项提议的目的是开展工作，了解大脑做了什么，使我们能够以我们所做的方式行事。我们关注的是大脑中称为前额叶和扣带回的部分。我们已经知道，这些大脑区域对我们感兴趣的行为特别重要，但我们不知道的是，它们是如何完成自己所扮演的角色的。我们想找出它们的运作机制，以及它们与大脑其他部分相互作用的方式。这项工作的一个关键部分是设计行为任务，以精确的方式探索认知操作，以揭示它们的机械基础。第二个部分是记录大脑活动，并观察它与行为的关系。我们使用磁共振成像(MRI)扫描仪，通常通过进行所谓的功能MRI(FMRI)扫描来实现这一点。功能核磁共振扫描可以告诉我们大脑中的血液氧合情况。这很有用，因为血氧水平依赖(BOLD)信号以非常精确的方式跟踪脑细胞--神经元--的活动。例如，当作出决定或提供反馈以实现行为的调整和改变时，可以估计特定大脑区域中大胆信号的分布随时间的变化。我们在动物身上进行fMRI记录，因为我们还想检查操纵我们记录的活动的后果。这对于找出哪些活动模式是因果驱动行为至关重要。我们可以通过对大脑进行精确和有限度的干预来测试因果关系。我们在麻醉下这样做，就像在人类病人身上做的一样。当动物康复时，我们会监测动物行为的变化。通常情况下，行为没有明显的变化，因为我们进行的干预是微妙的。然而，如果我们仔细地设计我们的行为任务，以便准确地探索特定的认知过程，那么我们或许能够获得同样微妙的行为变化。然后可以对它们进行测量和量化。我们使用猕猴，因为它们提供了人类前额叶和扣带回皮质的许多特征的模型。大多数其他动物缺乏这些特征，因此它们不能被用作模型。我们正在研究的问题之一是，我们应该以多快的速度改变和调整我们的行为。有证据表明，在一些心理疾病中，比如双相抑郁，我们的行为对每一个微小的反馈都变得过于敏感；行为变得过于反复无常。我们还感兴趣的是，当我们收到选择的反馈时，我们如何将反馈正确地归因于真正导致选择的事件。有证据表明，我们并不总是能很好地处理好这项简单的工作，当我们做得很差时，我们可能会得出奇怪的结论，说明我们应该对什么负责，而不是对什么负责。同样，这可能是心理疾病的一个特征。我们感兴趣的第三个过程是推理。神经学家经常研究调节对特定事件或行为的学习的神经机制。一旦学习完成，就可以在下一次遇到事件或需要采取行动时做出正确的决定。然而，在现实世界中，我们经常根据具有一些相似组成元素的一些情况的经验来推断下一步要做什么。我们将试图理解这些推论是如何做出的。

Every day we make decisions about what to do next. We do this because we are constantly monitoring and evaluating how well things are going. As a consequence we adjust our behaviour so it is appropriate for the current context or we decide to take one course of action rather than another. Sometimes we are aware of making these evaluations and decisions but often we are not. Nevertheless, however much we take these abilities for granted, it is striking when they are altered in psychological illnesses such as depression. The aim of this proposal is to undertake work to understand what the brain does to enable us to behave in the way that we do. Our focus is on parts of the brain called the prefrontal and cingulate cortex. We already know that these brain regions are especially important for the behaviour we are interested in but what we do not know is how they accomplish the role they play. We want to find out the mechanisms by which they operate and the way in which they interact with the rest of the brain. A key part of the work is designing behavioural tasks to probe cognitive operations in a precise way to reveal their mechanistic basis. A second component is recording brain activity and seeing how it relates to behaviour. We do this by using a magnetic resonance imaging (MRI) scanner, usually by taking what are called functional MRI (fMRI) scans. FMRI scans tell us about blood oxygenation in the brain. This is useful because the blood oxygen level dependent (BOLD) signal tracks the activity of the brain's cells -- neurons -- in a very precise way. It is, for example, possible to estimate changes in distribution of BOLD signal in specific brain areas from moment to moment as a decision is made or as feedback is provided to enable adjustments and changes in behaviour. We conduct the fMRI recording in animals because we also want to examine the consequences of manipulating the activity we record. This is essential for finding out what activity patterns are causally driving behaviours. We can test causation by making precise and circumscribed interventions in the brain. We do this under anaesthesia in the same way that it would be done with human patients. When the animals recover we monitor changes in behaviour. Usually there are no obvious changes in behaviour because the interventions we carry out are subtle. If, however, we have designed our behavioural tasks with care so as to precisely probe specific cognitive processes, then we may be able to pick up equally subtle alterations in behaviour. They can then be measured and quantified. We use macaques because they provide a model of many features of human prefrontal and cingulate cortex. Most other animals lack these features so they cannot be used as models. One of the questions that we are examining concerns how quickly we should change and adjust our behaviour. There is evidence that in some psychological illnesses, such as bipolar depression, our behaviour becomes too responsive to each minor piece of feedback; the behaviour becomes too volatile. We are also interested in how, when we receive feedback for a choice, we attribute that feedback correctly to the event that really caused it. There is evidence that we do not always manage this simple job well and when we are very poor at it we may draw odd conclusions about what we are, and are not, responsible for. Again this may be a feature of psychological illnesses. A third process we are interested in is inference. Often neuroscientists have studied the neural mechanisms that mediate learning about particular events or actions. Once the learning is done a good decision can be made next time the event is encountered or the action is needed. However, often in the real world we make inferences about what to do next on the basis of experience of some situations with some similar component elements. We will attempt to understand how such inferences are made.