Machine Ethics

Opening — Why this matters now Adaptive AI is quietly rewriting the rules of model evaluation. In regulated domains—especially healthcare—the question is no longer how accurate is your model? but rather what exactly improved, and why? The problem is deceptively simple: when both your model and your data change over time, performance becomes ambiguous. A model might appear to improve simply because the test set got easier. Or worse, it might degrade in real-world deployment despite looking better in controlled evaluation. ...

Opening — Why this matters now There is a quiet but consequential shift happening in AI. We are no longer evaluating models—we are evaluating agents. And agents don’t fail loudly. They fail gradually, politely, and often correctly—until the final step reveals that everything leading up to it was a mistake. The paper AgentHazard fileciteturn0file0 introduces a subtle but uncomfortable truth: the most dangerous behavior in AI systems doesn’t come from a single malicious instruction. It emerges from a sequence of reasonable decisions. ...

Opening — Why this matters now There was a time when we judged AI models by what they knew. Then came multimodal models, and we started judging them by what they could see. Now, quietly but decisively, the benchmark has shifted again: we are judging AI by what it can do. This is not a cosmetic upgrade. It is a structural shift. The emergence of agentic AI — systems that can invoke tools, search the web, manipulate images, and chain decisions — turns models from passive predictors into operational actors. And once an AI starts acting, correctness alone is no longer a sufficient metric. ...

Opening — Why this matters now There is a quiet bottleneck in modern mathematics—and by extension, in any field that depends on complex reasoning: verification doesn’t scale. The academic system still runs on trust. Referees don’t fully check proofs; they check whether proofs feel correct. That works—until it doesn’t. As the paper notes, even widely accepted results can contain errors simply because no one has the time (or incentive) to verify them line by line. fileciteturn0file0 ...

Opening — Why this matters now Everyone likes to say that AI “understands data.” Then you show it a chart. Suddenly, the illusion breaks. Vision-language models (VLMs) can caption images, describe scenes, and even summarize dashboards—but ask them a simple question about a bar chart or trend line, and they start hallucinating like a junior analyst on too much caffeine. ...

Opening — Why this matters now For years, the industry has been obsessed with scale: more data, larger models, longer context windows. The implicit assumption was simple—more exposure leads to better generalization. But something quietly uncomfortable has been emerging: large language models don’t just learn patterns—they sometimes remember too well. And that distinction is no longer academic. ...

Opening — Why this matters now Agentic AI is having a quiet identity crisis. We’ve spent the past two years optimizing outputs—better reasoning, longer context, more coherent plans. And yet, systems still fail in ways that feel oddly primitive: they drift off course after minor disturbances, retrieve the wrong memory at the wrong time, or pass verification checks while quietly violating the real objective. ...

Opening — Why this matters now The current generation of AI agents has an obsession: thinking deeper. Longer chains of reasoning. More steps. More tokens. More “intelligence.” And yet, when asked to do something annoyingly practical—like compiling a complete dataset across dozens of sources—they fail in surprisingly mundane ways: missing entries, duplicated facts, or simply running out of context. ...

Opening — Why this matters now There’s a quiet assumption embedded in modern AI safety: if a model says “Sorry, I can’t help with that,” then something meaningful has been achieved. The paper CRaFT: Circuit-Guided Refusal Feature Selection via Cross-Layer Transcoders challenges that assumption rather directly. fileciteturn0file0 What if refusal is not a principle—but a pattern? Not a rule—but a surface-level artifact of deeper computation? ...

Opening — Why this matters now If you’ve ever asked an AI to recreate a chart from an image, you’ve probably seen the illusion: it almost works. The bars are there, the colors vaguely align, but the labels drift, spacing collapses, and somewhere along the way, precision quietly disappears. This paper addresses a deceptively simple question: what if the model didn’t have to get it right the first time? ...