MIT - Critical Data

Instead of building AI that knows everything, we should be building AI that makes us better: more humble, more curious, more creative. How can we engineer virtues directly into clinical AI systems, equipping them with self-awareness modules that detect overconfidence, flag uncertainty, and prompt clinicians to seek fresh perspectives rather than passively accept a machine’s verdict?

The implications reach far beyond medicine. If we accept that the purpose of AI is not simply to automate cognition but to catalyze our evolution as a species, then the virtues we encode into these systems matter enormously.

The consortium behind this work practices what it preaches. The initiative spans all the continents except Antarctica, deliberately weaving together students, patients, data scientists, clinicians, social scientists, indigenous knowledge holders, and artists. Ultimately, the biases baked into AI are biases baked into who gets to design it. Let us stop building AI that thinks for us and does stuff for us, and start building AI that helps us humans think together, more wisely, and with the kind of courage our most complex challenges demand.

https://news.mit.edu/2026/creating-humble-ai-0324
Image: MIT News; iStock

After decades of conflicting evidence about how tightly to control blood sugar in critically ill patients, our analysis using causal inference methods on the MIMIC-IV database offers clarity, but clarity that must be understood within its specific context. By combining targeted maximum likelihood estimation with joint longitudinal-survival modeling on 8,002 patients from a single academic medical center in Boston, we discovered a U-shaped relationship between glucose and mortality, where aiming for glucose levels between 160-190 mg/dL appeared optimal, with overly aggressive glucose lowering dramatically increasing hypoglycemia risk (77% of patients at 100 mg/dL targets) without improving survival. However, our cohort, median age 66 years, 57% with diabetes, and significantly older and more comorbid than the general global ICU population, reflects the limitations inherent to single-center observational studies. Institutional variations in insulin protocols, glucose monitoring frequency, and clinical workflows at Beth Israel Deaconess Medical Center may not translate to other settings. This finding validates current guidelines recommending liberal glucose ranges and exemplifies how sophisticated analytics on high-resolution health data can help us move beyond the costly cycle of contradictory randomized trials, but it also underscores a critical principle: causal inference in ICU settings should be viewed as causally suggestive rather than definitive, precisely because we can never fully verify the absence of unmeasured confounding or account for the complex, context-dependent practices that shape real-world care. The real innovation here isn’t methodological; it’s about using these frameworks to ask better questions and identify promising directions before launching expensive trials, while maintaining epistemic humility about what observational data can and cannot tell us across diverse healthcare contexts and patient populations.

https://bmjopen.bmj.com/content/16/1/e104916.full