Episodic Memory: Uniquely Human?

我们记得过去的许多事情——从上学的第一天到今天早上吃早餐。我们可以在脑海中“再现”这些事件。这种类型的记忆被称为情景记忆。它不同于其他类型的记忆，比如对事实的记忆（比如乘法表）和对技能的记忆（比如骑自行车）。这个项目解决了关于情景记忆的两个难题。首先，情景记忆的功能是什么？为什么除了对事实和技能的记忆外，情景记忆对我们很有用？研究人员在这个问题上为许多观点辩护，不同意它在保存信息、想象、计划、决策、自我意识、沟通等方面的作用。第二，情景记忆是人类独有的吗？在动物思维研究中，这是一个激烈争论的主题，一些人认为情景记忆在动物中普遍存在（Allen & Fortin, 2013），而另一些人则认为这是人类独有的（suddenendorf & Corballis, 2007）。这些谜题是相关的。对情景记忆功能的不同描述表明了非人类是否拥有情景记忆的不同观点。但许多关于情景记忆功能的描述都是基于人类记忆的知识，因此很难应用于非人类。与此同时，关于记忆在动物和人工智能中的作用的证据可以帮助我们理解情景记忆的功能。与此同时，人工智能研究人员正在为人工智能开发情景记忆“模块”（Botvinick et al., 2019; Nuxoll & Laird, 2012），这可能会对这两个难题提供重要的启示。但是在生物记忆和人工记忆过程之间有重要的不同之处，所以任何比较都必须谨慎进行。该项目将汇集有关人类、动物和人工代理的记忆证据，以发展一种关于情景记忆功能的新理论。这将用于推进关于非人类记忆的辩论，并为情景记忆在人工智能中的新应用提出建议。这将是第一个对人工智能情景记忆给予重大哲学关注的项目，并将比较心理学、人工智能和哲学结合起来，为情景记忆的功能提供信息。项目团队将严格审查人工智能和比较心理学中关于情景记忆的现有文献，并将使用哲学论证和分析工具来探索如何在这些领域理解情景记忆，评估情景记忆是否应归因于非人类代理，并提取关于情景记忆功能的见解。然后，通过跨学科合作，该团队将把比较心理学和人工智能结合起来，推进我们对情景记忆的了解。该团队将与世界领先的比较心理学家尼古拉·克莱顿（Nicola Clayton）教授合作，探索人工智能工具如何推进动物记忆研究。在此基础上，该团队将与“动物-人工智能奥运会”（Crosby等人，2020年）的开发人员合作，这是一个开创性的项目，将人工智能与比较心理学的测试进行对比。他们将在3D模拟环境中实施情景记忆测试，并使用这些测试来测试有或没有情景记忆模块的人工智能体。研究结果将揭示情景记忆在智力中的作用，以及在动物身上检测情景记忆的方法。接下来，研究小组将把这项工作的结果结合起来，开发一种情景记忆功能的新描述，根据这种描述，情景记忆可能有多种形式，包括非人类动物的“最小”形式和人类更复杂的形式。他们将评估这种描述的伦理含义，探索情景记忆在人类道德生活中的作用，它对动物福利的意义，以及制造记忆机器的伦理影响。

We remember many events from our past - from our first day at school to having breakfast this morning. We can 'mentally relive' these events in our mind's eye. This type of memory is called episodic memory. It's distinguished from other kinds of memory, like memory for facts (like the times tables) and memory for skills (like riding a bike). This project addresses two puzzles about episodic memory.First, what is episodic memory's function? Why is it useful for us to have episodic memory in addition to memory for facts and skills? Researchers defend many views on this question, disagreeing about its role in preserving information, imagination, planning, decision making, self-awareness, communication and more. Second, is episodic memory uniquely human? This is a subject of intense debate in the study of animal minds, with some arguing that episodic memory is widespread in animals (Allen & Fortin, 2013), and others that it is uniquely human (Suddendorf & Corballis, 2007). These puzzles are related. Different accounts of episodic memory's function suggest different views on whether nonhumans have it. But many accounts of episodic memory's function are based on knowledge of human memory, so are difficult to apply to nonhumans. At the same time, evidence about the role of memory in animals and AI could inform our understanding of episodic memory's function. At the same time, AI researchers are developing episodic memory 'modules' for artificial agents (Botvinick et al., 2019; Nuxoll & Laird, 2012) - which could shed important light on both of these puzzles. But there are important dissimilarities between biological and artificial memory processes - so any comparisons must be made with caution. This project will bring together evidence about memory in humans, animals and artificial agents to develop a new theory of episodic memory's function. This will be used to advance debates about nonhuman memory and suggest new applications for episodic memory in AI. It will be the first project to give significant philosophical attention to episodic memory in AI, and to bring comparative psychology, AI and philosophy together to inform accounts of episodic memory's function.The project team will critically review the existing literature on episodic memory in AI and comparative psychology, and will use tools of philosophical argument and analysis to explore how episodic memory is understood in these fields, assess whether episodic memory should be ascribed to nonhuman agents, and extract insights about episodic memory's function.Then, through interdisciplinary collaborations, the team will bring comparative psychology and AI together to advance our knowledge of episodic memory. Working with with Professor Nicola Clayton, a world-leading comparative psychologist, the team will explore how tools from AI might advance the study of animal memory. Building on this, the team will collaborate with the developers of the 'Animal-AI Olympics' (Crosby et al., 2020) - a ground-breaking project which pits the intelligence of artificial agents against tests from comparative psychology. They will implement tests of episodic memory in a 3D simulated environment, and use these to test artificial agents with and without episodic memory modules. The results will shed new light on episodic memory's role in intelligence, and the methods for detecting it in animals. Next, the team will draw the results of this work together to develop a new account of episodic memory's function, according to which there may be many forms of episodic memory, including 'minimal' forms in nonhuman animals and more sophisticated forms in humans. And they will assess the ethical implications of this account, exploring episodic memory's role in humans' moral lives, its significance for animal welfare, and the ethical impacts of building machines that remember.