This thesis presents a series of results that study the deployment of fair, safe, and productive AI. We begin in Chapter one by both motivating and introducing the three topics of study: algorithmic fairness in performative prediction, weak-to-strong generalization, and benchmarking AI adoption and capability across the economy. Additionally, a brief synopsis of the results we will present in subsequent chapters is included.

In Chapter 2, we study the problem of enforcing fairness under distribution shift caused by human behavior. This distribution shift is known as performativity and is often caused by the reactions of individuals with vested interests in the outcome of the predictive model. While performativity is generally problematic, as it leads to changes in the data at test time, we develop algorithmic fairness practices that leverage performativity to achieve stronger group fairness guaranties in social classification problems (compared to what is achievable in non-performative settings). In particular, we show that it is possible to enforce traditionally conflicting group fairness definitions if performativity is present.

In Chapter 3, we study the problem of superalignment through the analogy of weak to strong generalization; or using weaker (less capable) feedback to train a stronger (more capable) model. We prove that weak-to-strong generalization is possible by eliciting latent knowledge from pre-trained LLMs. In particular, we cast the weak-to-strong generalization problem as a transfer learning problem in which we wish to transfer a latent concept from a weak model to a strong pre-trained model. We prove that a naive fine-tuning approach suffers from fundamental limitations, but an alternative refinement-based approach suggested by the problem structure provably overcomes the limitations of fine-tuning. Finally, we demonstrate the practical applicability of the refinement approach in multiple LLM alignment tasks.

Finally, in Chapter 4, we work towards measuring both AI adoption and the capability of AI to perform discrete labor tasks across various occupations. To measure adoption, we develop an open-source economic index that uses publicly available user-LLM chat data and ONET tasks to replicate studies produced by leading AI labs. To measure capabilities, we build a system that generates benchmark scenarios grounded in ONET occupations, tasks, and model-context-protocol (MCP) servers. We test Kimi-K2-Thinking with an OpenAI harness on scenarios across 13 occupations appearing frequently in our index, finding that there is ample room for improvement across tool calling, workflow completion, and hallucination rates.