Why machine learning needs humans

Early course failure can have consequences far beyond a single assessment, particularly for first-term students attempting their first objective assessment. Students who fail their first objective exam often leave college with debt but no degree, particularly students who are also first-generation college attenders and/or who come from under-resourced circumstances. Furthermore, faculty members lack time to search through student records and formulate effective proactive outreach. Too often students are not supported until after they have failed. Decision Intelligence helps close this critical gap between identifying risk and taking effective action, giving faculty timely, evidence-informed options to support students when intervention matters most. And most importantly... keeping humans at the center and in control. Contributor: Bernadette Howlett

These presentations highlight a broader opportunity for DI: applying causal reasoning not only to programmatic and operational decisions, but also to business development decisions that determine whether valuable services successfully reach the people they were designed to serve. Decision Intelligence provides a bridge between data and action. Funk’s work demonstrates how causal models can help decision makers identify where to intervene, clarify assumptions about how actions produce outcomes, and learn systematically from results. The business development application extends that opportunity. Growth, partnerships, outreach, and utilization all involve decisions within complex systems. DI can make the causal logic behind those strategies visible and testable, shifting the focus from “What does the data tell us?” to the more consequential question: “Given what we know, what should we do?”. Contributor: Bernadette Howlett, PhD. 

This article highlights the risk that sophisticated decision-support systems are training leaders to defer to systems rather than exercising their own judgment. As organizations embed AI into more decision-making processes, there is a growing risk that human judgment becomes passive, with leaders accepting persuasive-sounding recommendations without sufficient challenge or contextual reasoning due to cognitive offloading and automation bias. This reinforces the importance of designing decision systems that preserve human judgement, make reasoning transparent, and ensure AI augments, rather than replaces, sound judgment. Contributor: Michael Redford.

The democratization of AI-generated analysis is not a new idea. Nor is the argument that the talent race will be won by those with the best understanding of outcomes, and how they relate to actions, rather than those possessing the widest AI toolbox. What stands out about this article is DI's expanding reach, both globally, through its publication in India, and its appearance in a more mainstream news outlet. The shift in audience is as notable as the shift in geography, highlighting DI's growing relevance across the broader business landscape. Contributor: Michael Redford.

Lagally puts the interdisciplinary value of Decision Intelligence into practice by considering one of its fundamental tools, the causal decision model, within the broader context of decision ethics, grounded in realistic examples of ecological policy and intervention. What are the ethical implications of statistical estimation vs. causal inference? What makes a decision ethical in the first place? Lagally navigates these questions and presents a concise argument with an eye for concrete outcomes, relevant examples, and realistic limitations. Contributor: Isaac Kellogg. 

Causal inference shares significant common motivations, ideas, and tools with Decision Intelligence. Through realistic examples, this article discusses what DI readers will recognize as the classic P- vs. C-decision dilemma: the shortcomings of predictive Machine Learning models.
DI's decision models center humans, capturing and aligning their expertise around a multi-link causal structure with a wide variety of applications from interactive simulation to statistical optimization. This article's causal inference approach leads toward a few specialized programming tools somewhat more narrow in scope, aimed at statistical analysis or derivation of particular causal relationships. Look for strong overlapping insights that stem from those shared motivations and familiar ideas: correlations are not levers; predictions do not always map cleanly to outcomes or actions; in a complex and confusing decision-making scenario, modeling the causal structure (even with a simple drawing) is often invaluable. Contributor: Isaac Kellogg.

The article reviews AI- and IoMT-enabled applications across an array of clinical domains, including diabetes management (including gestational diabetes), cancer diagnosis, cardiac monitoring, surgical care, medication administration, and fall prevention in older adults, emphasizing the application of predictive analytics to improve diagnostics and other healthcare decision making. These examples demonstrate how healthcare is evolving from retrospective monitoring toward predictive and prescriptive decision support. By reducing the time between signal detection, risk identification, and intervention, AI-enabled decision systems have the potential to improve outcomes, reduce costs, and prevent harm at both the individual and population levels. Contributor: Bernadette Howlett. 

Many public health (and other types) of organizations have invested heavily in dashboards and analytics, yet decision-makers often feel overwhelmed by volumes of visuals and struggle to convert information into action. This article argues that effective decision support requires more than visualization; it needs timely, data-informed guidance on actions to take. Clear action guidance requires information resources that reduce cognitive overload and provide relevant context. This has direct implications for public health leaders who need systems that help practitioners move from data awareness to timely, evidence-informed decisions. The article points to public health dashboard failures during the COVID-19 pandemic as a reminder that when critical information lacks sufficient context and narrative guidance, organizations struggle to translate available data into effective action. Rather than relying on dashboards that require users to sift through metrics, interpret patterns, and determine next steps, decision-oriented systems can reduce cognitive burden, and time, by delivering actionable recommendations and continuously refining those recommendations as outcomes and new data become available. The time-to-action advantage of decision intelligence is a critical public health capability, where days and even hours can mean the difference between containing a problem and responding after preventable harm has occurred. Contributor: Bernadette Howlett.

Decision intelligence can be used to improve operational readiness in hospitals (and perhaps other settings), which is often overlooked as an intervention. Important positive impacts can be achieved on serious health problems, including maternal morbidity and mortality. As the authors stated, hospitals “do not act on probabilities alone. They act on resources – what is available, what is not, and what can be mobilized in time. A model that predicts hemorrhage risk without informing resource allocation risks becoming, in a sense, informationally rich but operationally incomplete.” This study demonstrated the wisdom of using decision intelligence to translate data and technology to the operational reality of a hospital, and the harm prevention that can be achieved by doing so. Contributor: Bernadette Howlett. 

What is a pattern-based decision? What is a causal decision? What's the difference? Why should you care? This article clearly and concisely answers each of these questions. As more and more organizations explore the role of AI in their decision-making processes, those that take the time to understand this distinction will gain a powerful intuition for guiding that exploration. The concepts explained here sit at the center of DI's ability to reveal where and how various forms of AI fit into decision making, where they do not, and when they don't, what works better instead. Contributor: Isaac Kellogg.

This applies DI to a real-world decision with complex ecological and humanitarian ramifications by walking through the process of building a CDD for the decision. It's especially useful as an example that carefully considers and clearly communicates areas where existing models, datasets, constraints, and metrics naturally fit into the decision model being built. DI is powerful as a connective tissue between people, data, actions, and outcomes, backed by solid engineering at every step, and Lagally demonstrates that here. Contributor: Isaac Kellogg. 

There is a growing pressure for greater transparency in automated decision-making, particularly as government agencies expand the use of AI in citizen services. These pressures are driving a shift in governance models towards explainable AI and retaining human-in-the-loop mechanisms to promote transparency and fairness in decisions that directly impact the public. DI provides a practical framework for building trust by ensuring decision flows remain visible and auditable. Contributor: Michael Redford. 

Wadodkar gives high level insight into the mindset shift facing data and analytics professionals as organizations turn to Decision Intelligence to help close the gap between data and outcomes. Her tips for CDAOs center decisions instead of data, with a focus on actions and measurable outcomes. She writes for executives and c-suite data analytics professionals, so her advice is rooted in the more familiar territory of Business Intelligence, KPIs, and high-level org strategy. There isn't airspace given toward representing decisions causally, or even just visualizing them, but the high-level mindset shift away from data-first and toward decisions-first is crucial, and Wadodkar speaks to that well. Contributor: Isaac Kellogg. 

This is an example of how more decision-centric operating models are delivering value at enterprise scale. While forecasting and planning functions have long existed within supply chain management, Hershey's transformation comes from improving how these activities are connected to operational decision-making across the supply chain. Rather than focusing primarily on technology tooling, the story highlights how better decision orchestration and tighter alignment between planning and execution can materially improve business outcomes. Contributor: Michael Redford. 

Gartner's first Decision Intelligence Platforms Magic Quadrant highlights DI as a significant force addressing meaningful gaps in the business world's approach to both decision-making and AI (and the combination of the two). This article provides a high-level peek into Gartner's analysis, with interview quotes that surface some of the crucial discussions motivating DI, like the value of human expertise, the push and pull of transparency and accountability, and DI's market relationship to the big players in AI. Contributor: Isaac Kellogg

The report underscores the need to shift from focusing on tools to decisions. GenAI delivers value when it enhances how organizations make and execute decisions, not just how they automate tasks or deploy general purpose tools. Teams following a traditional SaaS rollout are missing this. In contrast, the successful builders focus on domain fluency and tightly align AI with specific, high-impact business decisions that address real pain points. Contributor: Michael Redford.

How can Decision Intelligence deliver value as an emerging category in supply chain technology solutions? This post aims to define DI in the context of supply chain strategy, with enough precision to capture what differentiates it from other existing solution categories, in order to spotlight why that difference is valuable and why DI is becoming more relevant to supply chains now. Contributor: Isaac Kellogg. 

A good, practical case study from Bank of America showing how Decision Intelligence has been adopted at the enterprise level to tackle a growing AI-driven management challenge: as investment and use cases expand, AI often struggles to deliver stable, scalable decision-making. Contributor: Michael Redford