Choice Group

Due to the current COVID-19 situation this event will now take place online via Zoom.  Everyone is welcome to join this event using a computer with access to the internet and Zoom. To take part just follow these instructions: Download Zoom Join the event anytime after 4pm via this link: https://lse.zoom.us/j/948376433 At 4:30pm the event begins, with Q&A at 5:30pm. The event will…

Unfortunately, due to the current COVID-19 situation this event will no longer take place as planned. We apologise for any inconvenience caused.

Many pressing social problems are the result of undesirable human behavior. People pollute too much, fail to do enough to help the disadvantaged, and ignore government advice during pandemics. The dominant approach to predicting the effectiveness of interventions aimed at changing such behavior is the evidence-based approach. As Cartwright and Hardie describe it, this approach asks the predictor to construct an argument for the effectiveness of an intervention and then to support each premise of the effectiveness argument with evidence – facts about the world.

Unfortunately, the evidence-based approach is highly unreliable in certain cases (e.g., when the proposed intervention is novel, ambitious, complex, and not conducive to small-scale experimentation). Yet we need not resign ourselves to unreliable effectiveness predictions in such cases. We can use intuition grounded in folk psychology (our informal knowledge of others’ behavior) to predict the effectiveness of behavioral policy interventions.

However, relying on untutored intuition alone is problematic. I advocate instead an approach that relies on both intuition and evidence (while also being systematic and explicit about the non-intuitive inputs into the prediction). I argue that this sophisticated intuitive approach to predicting the effectiveness of behavioral policy interventions is more reliable, at least in certain cases, than the evidence-based approach.

The mathematical foundations of imprecise probability theory (IP) have been in place for 25 years, and IP has proved successful in practice. But IP methods lack rigorous accuracy-centered, philosophical justifications. Traditional Bayesian methods can be justified using epistemic scoring rules, which measure the accuracy of the estimates that they produce. But there has been little work extending these justifications to the IP framework. In this talk, I will first outline some initial work developing scoring rules for imprecise probabilities: IP scoring rules. Then I will explain why a range of impossibility results for IP scoring rules should not concern us. Finally, I will use IP scoring rules to engineer a new method for aggregating imprecise forecasts.

Abstract: TBA

Two of our PhD students present their research to the Choice Group

Suppose that an altruistic and morally motivated agent who is uncertain between evidential decision theory (EDT) and causal decision theory (CDT) finds herself in a situation in which the two theories give conflicting verdicts. We argue that even if she has significantly higher credence in CDT, she should nevertheless act in accordance with EDT. First, we claim that that the appropriate response to normative uncertainty is to hedge one’s bets. That is, if the stakes are much higher on one theory than another, and the credences you assign to each of these theories aren’t very different, then it’s appropriate to choose the option which performs best on the high-stakes theory. Second, we show that, given the assumption of altruism, the existence of correlated decision-makers will increase the stakes for EDT but leave the stakes for CDT unaffected. Together these two claims imply that whenever there are sufficiently many correlated agents, the appropriate response is to act in accordance with EDT.

Formal and social epistemologists have devoted significant attention to the question of how to aggregate the credences of a group of agents who disagree about the probabilities of events. Most of this work focuses on strategies for calculating the mean credence function of the group. In particular, Moss (2011) and Pettigrew (2019) argue that group credences should be calculated by taking a linear mean of the credences of each individual in the group, on the grounds that this method leads to more accurate group credences than all other methods. In this paper, I argue that if the epistemic value of a credence function is determined solely by its accuracy, then we should not generate group credences by finding the mean of the credences of the individuals in a group. Rather, where possible, we should aggregate the underlying statistical models that individuals use to generate their credence function, using "stacking" techniques from statistics and machine learning first developed by Wolpert (1992). My argument draws on a result by Le and Clarke (2017) that shows the power of stacking techniques to generate predictively accurate aggregations of statistical models, even when all models being aggregated are highly inaccurate.

In a series of papers over the past twenty years, and in a new book, Igor Douven has argued that Bayesians are too quick to reject versions of inference to the best explanation or abduction that cannot be accommodated within their framework. In this paper, I survey Douven’s worries and bring to bear a series of pragmatic and purely epistemic arguments to show that Bayes’ Rule really is the only correct way to respond to your evidence.

In his two famous papers of 1953 and 1955 defending Utilitarianism, Harsanyi draws on the same simple idea: that to determine what is morally best we should put ourselves in the shoes of an impartial, but sympathetic, rational evaluator of states of affairs that differ in terms of the wellbeing of the various individuals within them. In this talk, I will pursue a similar thought experiment, but depart from Harsanyi in two ways. Firstly, I will allow that the impartial or social evaluator take attitudes to uncertainty that are ruled out by expected utility theory, but which various rival theories of rationality regard as reasonable; especially in situations of ambiguity in which the evaluator has insufficient information to assign precise probabilities to the various possible states of affairs. Secondly, I will allow that they take attitudes to inequality that are disallowed by Utilitarianism but which are regarded as reasonable (or even mandatory) by rival social ethics. Despite this permissiveness, it turns out that impartial evaluation governed by minimal conditions of rationality and sympathy impose strong constraints on evaluations of individual and social wellbeing.

For those of us interested in developing norms for effective scientific advising, the SAGE minutes (59 sets of which are now publicly available) are a valuable resource. Drawing on these minutes, I consider the wider lessons for norms of scientific advising that can be learned from the UK's initial response to coronavirus, highlighting five key issues: (i) the division of advisory labour, (ii) the grain of recommendations, (iii) the role of worst-case scenarios, (iv) switching cost asymmetries and (v) the difference between independence and neutrality.

Precautionary principles are frequently appealed to both in public policy and in discussions o good individual decision-making. They prescribe omission or reduction of an activity, or taking precautionary measures whenever potential harmful effects of the activity surpass some threshold of likelihood and severity. One crucial appeal of precautionary principles has been that they seem to help guard against procrastinating on confronting and mitigating certain kinds of risk, namely those that are especially hard to quantify. Here I raise a challenge for precautionary principles serving as effective action-guiding tools to guard against (policy) inaction, procrastination, or recklessness. Given the fact that risks that are sufficiently harmful and sufficiently likely to fulfil the antecedent of a precautionary principle typically accumulate over time, precautionary principles are only effective if they constrain an agent’s decision-making over time. On the basis of this observation, I argue for two claims. First, to yield the normative verdicts proponents of precautionary principles would like to make, precautionary principles must be understood to be diachronic principles, which requires some added structure to how they are commonly formulated. And secondly, such diachronic precautionary principles invite policy procrastination and inaction in their own right, due to both the vagueness of thresholds of harm and likelihood, and because agents will often fail to abide by the principles if they ignore bygone risks.

Abstract: TBA

Two of our research students, Sophie Kikkert and Dmitry Ananyev, present their work to the Choice Group.

In this paper, we ask: how should an agent who has incoherent credences update when they learn new evidence? The standard Bayesian answer for coherent agents is that they should conditionalize; however, this updating rule is not defined for incoherent starting credences. We show how one of the main arguments for conditionalization, the Dutch strategy argument, can be extended to devise a target property for updating plans that can apply to them regardless of whether the agent starts out with coherent or incoherent credences. The main idea behind this extension is that the agent should avoid updating plans that increase the possible sure loss from Dutch strategies. This happens to be equivalent to avoiding updating plans that increase incoherence according to a distance-based incoherence measure.

The martingale convergence theorem implies that in certain situations a Bayesian agent is sure to converge to the truth in the limit. Gordon Belot has argued that this constitutes a liability for Bayesian epistemology since it ignores the many ways in which one might fail to identify truth in the limit. In this talk I will study convergence to the truth within a nonstandard probability framework that allows fine-grained distinctions between infinite hypotheses. Within the nonstandard framework, convergence to the truth is expected only for hypotheses that can be finitely approximated. Importantly, this leads not to a revision but a refinement of the standard martingale convergence theorem.

Are there ever good epistemic reasons to misrepresent base rates? I investigate this question in the context of recent legislation restricting the presentation of gender stereotypes, and the representation of minority groups in children’s books. I argue that our hesitancy around certain base rates makes sense in the context of a more general epistemic dilemma we face: between knowledge of acuality and knowledge of possibility. Given this dilemma, there are sound epistemic reasons to behave in ways the may involve wariness or misrepresentation of base rates. This approach has implications for the way in which ethical and epistemic norms interact with one another.