RI: Small: Foundations of Ethics for Multiagent Systems

现代人工智能(AI)技术令人惊讶的能力引发了人们对智能代理可能对人类福利行使的权力的合理担忧。伦理本质上是一个多主体关注的问题，因此，局限于单一主体的伦理观点是不够的。然而，今天对人工智能伦理的研究主要是对单个代理的研究。这个项目被称为多主体伦理学的MAE，采用了一种社会技术立场，即代理人(作为技术实体)帮助自治的社会实体或委托人(个人和组织)。社会技术系统(STS)的这种多主体概念捕捉到了在多个利益相关者的相互作用中如何产生伦理关切。这些基础将使我们能够实现伦理STSS，其中包括社会和技术控制，以尊重所述的伦理立场。要实现设想的基础，需要在三个方面进行新的思考。首先，Mae将通过融合哲学和政治学的思想，特别是价值理论和社会设计，来开发一个面向STSS的伦理模型。其次，Mae将开发推理技术，以帮助识别道德干预的潜在不道德后果。第三，MAE将开发在上下文中引出利益相关者价值观的技术，不断应用上述分析方法来帮助利益相关者细化他们的要求，以具体说明可接受的STS。这些技术将包括增强正式的建模工具，如模型检查器，以适应关于伦理的推理，以及用于评估由于个人决策而产生的系统结果的伦理的基于主体的社会模拟。MAE的新颖性在于它如何扩展多智能体系统建模，以包括规范和价值观等概念；结合形式推理和社会模拟来验证与所述道德姿态有关的STS规范；并通过识别可选STS规范之间的伦理权衡并在其社会和技术架构中捕获STS的道德姿态来帮助指定STS。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The surprising capabilities of modern artificial intelligence (AI) technologies raise legitimate concerns of the power that intelligent agents potentially wield over human welfare. Ethics is inherently a multiagent concern and thus a view of ethics limited to a single agent is inadequate. However, research into AI ethics today is dominated by work on individual agents. This project—dubbed Mae for Multiagent Ethics—adopts a sociotechnical stance in which agents (as technical entities) help autonomous social entities or principals (people and organizations). This multiagent conception of a sociotechnical system (STS) captures how ethical concerns arise in the mutual interactions of multiple stakeholders. These foundations would enable us to realize ethical STSs that incorporate social and technical controls to respect stated ethical postures.To realize the envisioned foundations requires new thinking, in three thrusts. First, Mae will develop a model of ethics geared toward STSs by incorporating ideas from philosophy and political science, especially value theory and the design of a society. Second, Mae will develop reasoning techniques to help identify potentially unethical outcomes of ethical interventions. Third, Mae will develop techniques for eliciting stakeholders' values in context, continually applying the above-mentioned analysis methods to help stakeholders refine their requirements to specify an acceptable STS. These techniques will include enhancements of formal modeling tools, such as model checkers, to accommodate reasoning about ethics as well as agent-based social simulation applied for evaluating the ethicality of system outcomes due to individual decision making. Mae's novelty lies in how it expands multiagent system modeling to include constructs such as norms and values; combines formal reasoning and social simulation to verify STS specifications with respect to a stated ethical posture; and helps specify an STS by identifying ethical tradeoffs between alternative STS specifications and capturing an STS's ethical posture in its social and technical architectures.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.