Bentham Research
Human intelligence for frontier AI
Bentham Research is an applied research lab that transforms post-doctorate domain scholarship into structured data for AI evaluation and post-training.
Our scholar-centred systems elicit expertise through the research workflows scholars already use to interrogate, critique and improve reasoning — transforming that expert judgement into structured datasets and training signals.
We focus on domains where quality cannot be reduced to automated verification, with ethics and the humanities central to our present work.
Our mission is to bring humanities scholarship into the core feedback loop through which frontier AI is evaluated and improved.
Advanced AI is progressing fastest where outputs can be checked quickly and reliably, as in code and formal mathematics. Many of the questions that matter most in human affairs are fundamentally different.
Reasoning in ethics, history, religion, philosophy and the humanities often admits multiple interpretations and requires contextual judgment depending on standards that may themselves be contested. There may be no single programmable test to determine whether an answer has reasoned well.
Bentham exists to solve this humanities challenge for advanced AI: we build the systems through which expert judgement in these domains can be most effectively elicited, validated, structured and made into a reliable source of human feedback for AI evaluation and post-training.
For millennia, humanities scholars have developed modes of testing arguments, evaluating interpretations, weighing competing considerations, and drawing nuanced conclusions. Yet much of this expertise is difficult to capture through conventional data-labelling workflows. Asking scholars to step outside their normal research practice and work within machine learning annotation schemes can discard precisely the tacit know-how that makes their expertise valuable.
Bentham takes the opposite approach. We build both the scholar-facing research environment in which experts can exercise their knowledge naturally and the technical systems that transform their judgement into structured datasets, rigorous evals, and training signals.
Bentham creates a continuous intellectual feedback loop between frontier models and the world’s leading post-PhD scholarly communities.
Bentham provides a personalised AI research environment that learns from the scholar while the scholar uses it in real time. Our systems continuously convert authentic scholarly interaction into validated reasoning signal for model training and evaluation.
Bentham Research exists to transform the world’s deepest expertise in ethics, philosophy, law, history and the humanities into the evaluations, training signals and alignment infrastructure required to make advanced AI safer, more trustworthy and better aligned with humanity.
Ultimately, Bentham seeks to close the gap between frontier model reasoning and the knowledge, judgment and understanding of the world’s leading experts, so that advanced AI can reason about humanity with the depth and sophistication humanity itself has developed over centuries.