Towards a next-generation computational neuroscience

我们不知道什么？最近，《科学》杂志选出了未来25年科学家面临的25个最大的未解之谜。在弄清宇宙的组成之后，第二个问题是：意识的生物学基础是什么？这确实是一个大问题。对意识经验、意志和主观性的科学描述将步哥白尼和达尔文的后尘，通过重构我们与彼此以及与自然的关系，许多临床和技术应用将随之而来。对意识的科学解释不会在“尤里卡”时刻完全形成。我们需要的是一个多学科的，综合的方法，结合理论和实验，并利用信息/计算科学和神经，心理和医学科学之间的交流。在这种交流的前沿，计算神经科学（CN）使用计算方法来模拟复杂的大脑过程，就像气象学使用计算机预测天气一样。在这种观点中，与早期的“人工智能”（AI）方法相反，大脑不是计算机，智能行为和意识体验来自复杂的大脑-身体-环境相互作用，以时间精确的方式展开。当前的CN主要集中在神经系统的单层次描述上（例如，神经活动如何影响神经元之间的连接），而忽略了连接大脑、身体和行为的多尺度关系。此外，当前的CN在意识本身方面也令人惊讶地保持沉默。通过瞄准和克服这些限制，我们的研究将为适应行为和意识体验的神经机制提供新的见解。我们将遵循三个相互作用的主题：（i）大规模CN模型的设计和分析，以探索多尺度神经相互作用如何塑造脑-体-环境相互作用以及如何被脑-体-环境相互作用塑造;（ii）发展新理论，以确定复杂网络中的因果相互作用（我们称之为“因果网络分析”），以及（iii）创建CN模型，解释意识的功能重要方面，例如，每一个有意识的体验都将不同的信息源整合到统一的场景中。上述主题的理论工作将与来自多个来源的实验数据相互作用。在细粒度的水平上，我们将描述池塘蜗牛完整大脑中的因果相互作用，揭示简单（无意识）生物体与环境相互作用时的综合神经功能。缩小，我们将应用因果网络分析从人类在各种意识状态下获得的脑成像数据，以测试基于CN模型的预测，并指导新模型的设计。细粒度层面的见解将为我们理解更复杂的意识机制提供基础，因果网络贯穿大脑，身体和环境提供了一个共同的理论框架。综合起来，这些研究链将催化意识科学从相关性到解释性的重要转变，以及基础科学的进步，我们的研究将在生物科学和信息科学的界面上产生重要的实际效益。这些将包括人工智能/机器人设备的新设计原则，复杂技术网络设计和控制的新见解，以及管理大规模数据集的新工具。下一代CN还将支持新的临床方法。许多与大脑相关的健康问题，从昏迷到抑郁再到失眠，都可以被理解为意识障碍的表现，而现有的许多临床方法都是姑息性的，缺乏理论基础。我们的研究将为新一代有效的临床干预措施提供理论基础。

What don't we know? Recently, the journal Science selected 25 of the biggest unanswered questions facing scientists over the next 25 years. Number two on the list, right after figuring out the composition of the universe, is: What is the biological basis of consciousness? This indeed is a big question. Scientific descriptions of conscious experience, volition, and subjectivity will follow in the footsteps of Copernicus and Darwin by restructuring our relationship with each other and with nature, and many clinical and technological applications will follow.A scientific account of consciousness will not arrive fully formed in a 'Eureka' moment. What is needed is a multidisciplinary, integrative approach combining theory and experiment and exploiting the interchange between the information/computation sciences and the neural, psychological, and medical sciences. At the front-line of this interchange, computational neuroscience (CN) uses computational approaches to model intricate brain processes in much the same way that meteorology uses computers to forecast the weather. In this view and in contrast to early approaches to 'artificial intelligence' (AI), brains are not computers, and intelligent behavior and conscious experience arise from complex brain-body-environment interactions unfolding in temporally precise ways. Much current CN focuses on single levels of description of neural systems (e.g., how neural activity affects connections among neurons) and neglects the multi-scale relations that connect brains, bodies, and behavior. Moreover, current CN is also surprisingly silent with regard to consciousness itself. By targeting and overcoming these limitations, our research will deliver new insights into the neural mechanisms underlying adaptive behavior and conscious experience. We will follow three interacting themes: (i) design and analysis of large-scale CN models to explore how multi-scale neural interactions shape and are shaped by brain-body-environment interactions; (ii) development of new theory to identify causal interactions in complex networks (what we call 'causal network analysis'), and (iii) creation of CN models that account for functionally significant aspects of consciousness, for example that each conscious experience integrates diverse information sources into unified scenes. Theoretical work in the above themes will interact with experimental data from multiple sources. At a fine-grained level we will characterize causal interactions in the intact brain of a pond snail, shedding light on the integrated neural function of a simple (non-conscious) organism as it interacts with its environment. Zooming out, we will apply causal network analysis to brain-imaging data acquired from humans in various states of consciousness, to test predictions based on CN models, and to guide the design of new models. Insights at the fine-grained level will scaffold our understanding of the more complex mechanisms underlying consciousness, with causal networks cross-cutting brains, bodies and environments providing a common theoretical framework. Taken together, these research strands will catalyze an important shift from correlation to explanation in consciousness science.As well as advances in basic science, our research will have important practical benefits at the interface of the biological and information sciences. These will include new design principles for AI/robotic devices, new insights for the design and control of complex technological networks, and new tools for the management of large-scale datasets. A next-generation CN will also underpin new clinical approaches. Many brain-related health problems, from coma to depression to insomnia, can be understood as expressions of disordered consciousness, and many existing clinical approaches are palliative and lacking in theoretical foundation. Our research will provide a theoretical basis for a new generation of effective clinical interventions.