Managing Severe Uncertainty

需要由个人、团体或机构作出的许多重要决定是在这些决定所依赖的因素相当不确定的情况下作出的。这种不确定性有两个基本维度：(i)科学或建模的不确定性（我们不知道世界的状态是什么，以及它将在哪种动态规律下发展）和（ii）期权的不确定性(我们不知道我们的选择是什么，以及如果我们行使其中一个或另一个会发生什么。至关重要的是，这些维度相互作用：例如，建模的不确定性会导致期权的不确定性。气候变化政策制定体现了这两种形式的不确定性。在科学的不确定性方面，我们面临着气候模型不完善和非线性的问题，这破坏了标准的预测程序。特别是，众所周知，气候模型在某种意义上是非常不完美的，因为它们甚至不能近似地反映所有现实世界的目标系统。最近的结果表明，这导致了概率预测的崩溃，从而剥夺了我们最强大和最广泛使用的预测工具。在选择不确定性方面，我们对一些拟议的缓解和适应措施的潜在影响（和副作用）知之甚少——碳捕获和储存、从海水中制造云、在外层大气中放置反射器等等——也不知道可能开发出哪些其他机制或技术。但是，尽管最近对代理人缺乏完全概率信息（通常称为模糊情况）的情况下的理性决策的兴趣重新抬头，理论方法仍然相当不发达。决策理论家对第一种不确定性的讨论甚至还处于萌芽阶段，就像对这些形式的不确定性如何相互关联的讨论一样。以气候科学和政策为主要案例研究，我们将研究两个不同的问题家族。1. 关于气候变化的科学不确定性的原因是什么？这涉及到以下问题：概率预测为什么会失败，又是如何失败的？什么样的模型不完美会“破坏”预测能力？我们怎样才能发现这些缺陷呢？如何改进模型？2. 在不确定的情况下，如何做出决策？这涉及到这样的问题：当我们对关键变量缺乏确定的概率预测时，我们如何做出决定，比如未来日期的当地降水，我们的决定取决于（这涉及科学的不确定性）？当我们不知道哪些政策选择是可行的（这涉及到选择的不确定性），我们应该如何考虑我们可能的政策选择？这个为期三年的项目将汇集伦敦政治经济学院在科学建模、决策和气候政策方面具有专长的学者，将以一种反映它们相互依存关系的方式，为这些问题找到答案。例如，制定政策决定所需的信息种类和准确程度应该为处理科学不确定性的战略提供信息。相应地，制定政策的方式应该承认科学的不确定性，并在面对不确定性的不同可能解决方案时最大化其稳健性。

Many of the important decisions that need to be made by individuals, groups or institutions are made under conditions of considerable uncertainty about the factors upon which these decisions depend. There are two essential dimensions to this uncertainty: (i) scientific or modelling uncertainty (we don't know what the state of the world is and under which dynamical law it will evolve) and (ii) option uncertainty (we do not know what our options are and what would happen if we were to exercise one or another of them. Crucially, these dimensions interact: modelling uncertainty contributes to option uncertainty, for instance. Climate change policy-making exemplifies both forms of uncertainty. On the side of scientific uncertainty, we face the problem that climate models are imperfect and non-linear, which undermines standard forecasting procedures. In particular, climate models are known to be strongly imperfect in the sense that they do not even approximately mirror all the real-world target systems. Recent results suggest that this leads to the breakdown of probabilistic forecasts, thus depriving us of the most powerful and widely used tool for making predictions. On the side of option uncertainty, we know little about the potential effects (and side-effects) of some of the proposed measures for mitigation and adaption - carbon capture and storage, creating clouds from seawater, placing reflectors in the outer atmosphere, and so on - or what other mechanisms or technologies might be developed. But despite a recent resurgence of interest in rational decision making in situations in which agents lack full probabilistic information (commonly termed situations of ambiguity), theoretical approaches remain rather undeveloped. Decision theorists' discussions of the first kind of uncertainty are even more embryonic, as does discussion of how these forms of uncertainty are related to one anotherTaking climate science and policy as our main case study, we will look at two different families of questions. 1. What are the reasons for scientific uncertainty regarding climate change? This involves questions such as: Why and how do probabilistic forecasts break down? What model imperfections 'destroy' predictive power? How might we detect these imperfections? How can models be improved? 2. How should one make policy decisions under uncertainty? This involves questions such as: How can we make decisions when we lack determinate probabilistic predictions for crucial variables, such as local precipitation at future dates, upon which our decisions depend (this concerns scientific uncertainty)? How should we think about our possible policy options when we don't know which of them will be feasible (this concerns option uncertainty)? This three-year project, bringing together academics at the LSE with expertise in scientific modelling, decision-making and climate policy, will develop answers to these questions in a way that reflects their interdependence. The kind of information and level of accuracy required in order to make policy decisions should, for instance, inform strategies for dealing with scientific uncertainty. Correspondingly, policy decisions should be made in a way that recognises scientific uncertainty and maximises their robustness in the face of different possible resolutions of the uncertainty.