> Markdown version of [/videos/72-augmented-intelligence-for-transport-planning-human-in-the-loop-modelling](https://www.wearedevelopers.com/videos/72-augmented-intelligence-for-transport-planning-human-in-the-loop-modelling). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Augmented Intelligence for transport planning: Human in the Loop Modelling

Why do human operators reject mathematically perfect transport plans? Discover how probabilistic programming and human-in-the-loop frameworks bridge the gap between rigid optimization and real-world execution.

- **Speakers:** Stefan Petrov
- **Event:** WeAreDevelopers LIVE
- **Published:** November 17, 2020
- **Duration:** 44:55
- **URL:** https://www.wearedevelopers.com/videos/72-augmented-intelligence-for-transport-planning-human-in-the-loop-modelling

## Summary

Implementing augmented intelligence in transport planning requires resolving the natural conflict between mathematical optimization and human decision-making. Standard mixed-integer programming and other optimization techniques produce rigid global plans that local human operators frequently reject. This resistance occurs because human planners operate imperatively, balancing localized responsibilities, institutional risk aversion, and unrecorded 'soft' constraints that declarative algorithms fail to capture. Furthermore, pure black-box machine learning models inherently lack common sense, often fitting irrelevant confounding artifacts (like the background environment of an image) rather than the actual operational signal, making expert intervention crucial.

To bridge this gap, modern decision support systems must transition from autonomous black boxes to transparent, human-in-the-loop frameworks. Probabilistic programming serves as a powerful foundational technique, merging business logic with statistical models to operate effectively in low-data supply chain regimes. For example, using Bayesian distributions allows teams to confidently estimate missing operational parameters, such as unrecorded package dimensions. Concurrently, data scientists can extract interpretable logic from complex tree-based algorithms using tools like RuleFit, transforming opaque gradient boosting predictions into shallow, expert-validated rules.

The UI and architectural design of these systems must heavily favor operational transparency over absolute algorithmic authority. Openly visualizing system uncertainty and confidence intervals is a proven mechanism for building user trust, as it explicitly signals when a local planner should override the model with real-time floor knowledge. Finally, treating constraint editing as an interactive, online feature rather than a static parameter empowers operators to safely modify optimization requirements. By utilizing warm-start solver features and continuous sensitivity analysis, local decision-makers can view the upstream and downstream financial impacts of their operational exceptions, functionally aligning deep mathematical accuracy with fluid human execution.

**Keywords:** transport planning optimization, human-in-the-loop modeling, probabilistic programming integration, mixed-integer programming, interactive constraint editing, interpretable machine learning, rulefit algorithm extraction, decision support systems, logistics capacity planning, bayesian joint distributions, operational uncertainty visualization, local vs global optimization, warm start solver features, solver sensitivity analysis, imperative vs declarative modeling

## Chapters

1. **Introducing human-in-the-loop decision support for logistics planning** (00:16) — Distributing complex automated processes to local planners requires systems that handle incomplete data alongside human oversight.
1. **Decoupling problem representation and solution using mathematical optimization** (03:40) — Framing business rules as explicit mathematical structures allows programmatic solvers to handle massive configurations efficiently.
1. **Navigating the mismatch between declarative optimization and imperative workflows** (08:08) — Reconciling global algorithm objectives with localized human decision-making prevents rigid automated plans from completely failing during deployment.
1. **Evaluating confounding factors and bias in machine learning models** (13:24) — Identifying underlying data collection errors ensures operators trust algorithmic predictions beyond surface-level accuracy scores.
1. **Integrating algorithmic decision support into local operational planning workflows** (17:35) — Bridging automated operations with partial plan execution capabilities builds trust and adapts to unrecorded logistical conditions.
1. **Handling missing package parameters using probabilistic programming and sampling** (22:53) — Generating Bayesian estimates for unrecorded inputs enables transportation systems to accurately project total truck capacities.
1. **Extracting interpretable operational rules from complex tree-based models** (29:34) — Converting vast random forest structures into shallow logic statements allows domain experts to validate and correct underlying predictive logic.
1. **Displaying predictive statistical uncertainty to empower local decision makers** (32:35) — Exposing statistical confidence intervals helps human operators determine when to override software outputs using real-time situational context.
1. **Evaluating hypothetical business scenarios via interactive linear constraint editing** (34:22) — Allowing users to toggle strict operational limiters enables realistic comparisons between idealized optimization pathways and baseline resource costs.
1. **Validating model assumptions with active learning and expert feedback** (42:23) — Deploying intelligent sampling interfaces captures targeted domain knowledge to systematically refine mathematical boundaries.

## Related Moments

- [Audience questions on practical machine learning operational strategies](https://www.wearedevelopers.com/videos/262-is-my-ai-alive-but-brain-dead-how-monitoring-can-tell-you-if-your-machine-learning-stack-is-still-performing) (from "Is my AI alive but brain-dead? How monitoring can tell you if your machine learning stack is still performing")
- [Commercial applications of modern artificial intelligence and machine learning](https://www.wearedevelopers.com/videos/154-machine-learning-for-software-developers-and-knitters) (from "Machine Learning for Software Developers (and Knitters)")
- [Solving complex engineering challenges in artificial intelligence deployment](https://www.wearedevelopers.com/videos/392-mlops-what-s-the-deal-behind-it) (from "MLOps - What’s the deal behind it?")
- [Scaling intelligent decision making across software automation workflows](https://www.wearedevelopers.com/videos/521-staying-safe-in-the-ai-future) (from "Staying Safe in the AI Future")
- [Transitioning from machine learning models to complex agentic systems](https://www.wearedevelopers.com/videos/1383-the-state-of-genai-machine-learning-in-2025) (from "The State of GenAI & Machine Learning in 2025")
- [Audience Q&A on autonomous driving models and data](https://www.wearedevelopers.com/videos/519-finding-the-unknown-unknowns-intelligent-data-collection-for-autonomous-driving-development) (from "Finding the unknown unknowns: intelligent data collection for autonomous driving development")

## Related Articles

- [Stephan Gillich - Bringing AI Everywhere](https://www.wearedevelopers.com/magazine/489-stephan-gillich-bringing-ai-everywhere)
- [Graph and AI Trends 2026: Why Is AI Running but Not Yet Delivering?](https://www.wearedevelopers.com/magazine/680-graph-and-ai-trends-2026-why-is-ai-running-but-not-yet-delivering)
- [WWC24 Talk - Scott Hanselman - AI: Superhero or Supervillain?](https://www.wearedevelopers.com/magazine/469-wwc24-talk-scott-hanselman-ai-superhero-or-supervillain)
- [MLOps – What’s the deal behind it?](https://www.wearedevelopers.com/magazine/125-mlops-what-s-the-deal-behind-it)

## Related Jobs

- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**
- [AI Operations Manager (all genders)](https://www.wearedevelopers.com/jobs/48263-ai-operations-manager-all-genders) at **envelio**
- [Data Scientist](https://www.wearedevelopers.com/jobs/ext/1351648-data-scientist) at **Almedia**
- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [Head of AI Applications](https://www.wearedevelopers.com/jobs/ext/1231536-head-of-ai-applications) at **ZEISS Group**
- [Head of AI Applications](https://www.wearedevelopers.com/jobs/ext/1456210-head-of-ai-applications) at **ZEISS Group**