Stefan Herzog
LightNews LightNews
banner
stefanherzog.bsky.social
Stefan Herzog
@stefanherzog.bsky.social
3.3K followers 1K following 190 posts
Senior Researcher @arc-mpib.bsky.social MaxPlanck @mpib-berlin.bsky.social, group leader #BOOSTING decisions: cognitive science, AI/collective intelligence, behavioral public policy, comput. social science, misinfo; stefanherzog.org scienceofboosting.org
Posts Replies Media Videos
stefanherzog.bsky.social
Brodeur, A., Lé, M., Sangnier, M., & Zylberberg, Y. (2016). Star Wars: The Empirics strike back. American Economic Journal: Applied Economics, 8(1), 1–32. https://doi.org/10.1257/app.20150044 Using 50,000 tests published in the AER, JPE, and QJE, we identify a residual in the distribution of tests that cannot be explained solely by journals favoring rejection of the null hypothesis. We observe a two-humped camel shape with missing p-values between 0.25 and 0.10 that can be retrieved just after the 0.05 threshold and represent 10–20 percent of marginally rejected tests. Our interpretation is that researchers inflate the value of just-rejected tests by choosing “significant” specifications. We propose a method to measure this residual and describe how it varies by article and author characteristics.
July 8, 2025 at 8:37 AM
stefanherzog.bsky.social
8/

3. Nikolas Zöller (Max Planck Institute for Human Development, Berlin) talks about “Enhancing medical diagnostics with hybrid collective intelligence”
@nikoz.bsky.social
www.mpib-berlin.mpg.de/staff/nikola...
Portrait photo of Nikolas Zöller (Max Planck Institute for Human Development, Berlin)
April 11, 2025 at 2:30 PM
stefanherzog.bsky.social
6/

2. Lucie Flek (University of Bonn) talks about “Pluralistic LLMs: Challenges in modeling and evaluation of perspective taking”.
caisa-lab.github.io/members/luci...
Portrait photo of Lucie Flek (University of Bonn)
April 11, 2025 at 2:30 PM
stefanherzog.bsky.social
4/

1. Jason Burton (Copenhagen Business School) talks about
“How large language models can reshape collective intelligence”.
@jasonburton.bsky.social
www.cbs.dk/en/research/...
Portrait photo of Jason Burton (Copenhagen Business School)
April 11, 2025 at 2:30 PM
stefanherzog.bsky.social
🚨🤖👥
Webinar "Collective Intelligence in the era of LLMs"
April 24 5-6pm CET
www.hacid-project.eu/hacid-webina...

Register
nesta.zoom.us/meeting/regi...

Speakers:
- Jason Burton, Copenhagen Business School
- Lucie Flek, University of Bonn
- Nikolas Zöller, Max Planck Institute for Human Development
A group of six people sits around a large wooden table working collaboratively with laptops, tablets, smartphones, and notebooks. The screens display a presentation slide titled “Discussion Outline.” The table also has coffee mugs, notebooks with handwritten notes, and small potted plants, suggesting a casual but focused team meeting or brainstorming session.
April 11, 2025 at 2:30 PM
stefanherzog.bsky.social
Panda eats[,] shoots, and leaves. 🐼
March 12, 2025 at 6:35 AM
stefanherzog.bsky.social
Panda eats[,] shoots, and leaves. 🐼
Panda eats[,] shoots, and leaves. 🐼
February 26, 2025 at 4:05 PM
stefanherzog.bsky.social
7/ And here's a summary of future issues that should be tackled to move the science and practice of boosting forwards.
List with future issues: 1. How can the boosting approach benefit from other relevant research, including research on empowerment and capability (Conger & Kanungo 1988, Nussbaum 2011, Zimmerman 2000), behavior change in health psychology (Michie et al. 2011), human-centered design (Lyon et al. 2020), engineering psychology (Wickens et al. 2022), cognitive task analysis (Crandall et al. 2006), and implementation science (Bauer et al. 2015)? 2. Boosts, like other behavioral public policy interventions, need to be implemented as part of a policy mix, especially when they respond to complex problems. How can synergies between regulation, boosts, and nudges be identified and exploited? How does boosting interface with well-being public policy (Fabian & Pykett 2022)? 3. How can research that deals with human decision making for teams and organizations—for instance, fast and frugal heuristics (Gigerenzer et al. 2022) and cognitive repairs (Heath et al. 1998)—be better integrated into research on behavioral public policy? 4. How can boosts best be scaled up and disseminated (e.g., apps, fact boxes, games), and what enhances the longevity of the effects of boosting interventions? 5. What would it mean to apply boosting on a collective level (see Hofmann 2024)? 6. How can boosting enable people to better deal with digital media, online challenges (e.g., deepfakes, dark patterns, microtargeting), and emerging AI technologies (e.g., large language models and other generative AI, algorithmic decision making) (see also Herzog & Franklin 2024)? 7. Boosting is one tool to support lifelong and cumulative learning. It is not meant as a verdict on the success or lack thereof of compulsory learning. Nevertheless, one question is whether the competences that boosting addresses (e.g., risk competences, decision-making competences, digital competences) can and should be integrated into school curricula.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
6/ Here's a summary of the main points of our #boosting paper:
List with summary points: 1. Behavioral public policy garnered widespread attention with the introduction of nudging, which aims to steer behavior while maintaining freedom of choice. 2. Criticisms of nudging include that it does not promote agency and competences and that it relies—overly optimistically—on the presence of benevolent choice architects. 3. The proliferation of environments threatening people's autonomy, the slow pace of systemic approaches to tackling societal issues, and the intrinsic benefits of empowerment make empowering citizens an indispensable objective of behavioral public policy. 4. Boosting is a behavioral public policy approach to empowerment grounded in evidence from behavioral science that shows that humans’ boundedly rational decision making is not as flawed as the nudging approach assumes. 5. Boosts are interventions that improve people's competencies to make informed choices that conform to their goals, preferences, and desires. 6. In self-nudging boosts, people learn to use architectural changes in their proximate choice environment to regulate their own behavior—that is, they are empowered to adapt their own choice environments. 7. There are boosts to foster core competences in many domains, including finance, online environments, and health, as well as broader, overarching areas, such as motivation, risk, and judgment and decision making. Boosts should be part of a policy mix that also includes system-level approaches. 8. When implementing boosts, policy makers need to avoid the trap of individualizing responsibility and to be mindful that, due to differences in cognition and motivation, inequalities in the desirable effects across boosted individuals may emerge.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
5/ When to consider #boosting or #nudging? See the box below for guidance.
This image is a screenshot of text that outlines the considerations for choosing between boosting and nudging in public policy. Boosting relies on the target audience having both cognitive ability and motivation to develop competencies, making it unsuitable if these conditions are absent. For nudging, key questions include: whether there is uncertainty about people’s goals, heterogeneity or conflict in those goals, the need for transparency in nudges, risks of nonbenevolent choice architects, potential for toxic choice architectures by the private sector, and whether the aim is to foster generalizable, lasting behaviors. If any of these questions apply, nudging's use may be problematic.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
4/ #Nudges work by subtly influencing choices (e.g., defaults, framing) without requiring much effort or awareness from individuals.

#Boosts, in contrast, are skill-based interventions (e.g., teaching rules of thumb to deal with uncertainty or self-control problems).
This table compares nudging and boosting approaches in public policy. Nudging targets behaviors by addressing cognitive and motivational deficiencies through changes in choice architecture, often relying on dual-system models of cognition. Its effects are typically reversible and focus on correcting specific, localized mistakes, though it may raise concerns about autonomy and transparency. In contrast, boosting aims to build long-term competencies, fostering skills and knowledge while shaping the external environment. It assumes cognitive architectures are malleable, with effects designed to persist after interventions are removed. Boosting emphasizes transparency and voluntary cooperation, equipping individuals with domain-specific or transferable skills. This image illustrates the nurturing of a brain, symbolized by a hand watering it with a watering can, suggesting growth, development, or empowerment. The upward arrow emphasizes progress or improvement, aligning with themes of cognitive development, skill-building, or fostering competencies
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
3/ #Boosting differs from #nudging by focusing on fostering competences rather than altering choice architectures. It equips individuals with tools and knowledge to make better decisions independently. See the table and scienceofboosting.org for examples.

Why is this distinction important?
This image humorously represents a brain leaping towards a thumbs-up gesture, symbolizing empowerment, motivation, and success.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
2/ The #boosting approach uses behavioral science to empower individuals by improving their judgment and decision-making competences.

The article (written by myself & Ralph Hertwig) highlights the untapped potential of behavioral science for public policy.

#behavioralscience #publicpolicy
This image is a screenshot of a definition of "boosting", which defines boosting as a behavioral public policy approach aimed at empowering individuals by improving their competencies to make informed choices that align with their personal goals and preferences. Unlike nudging, boosting respects human decision-making abilities and fosters competence development. It involves both cognitive and environmental strategies to support specific behaviors while preserving personal agency. The emphasis is on transparency and voluntariness, allowing individuals to adopt or reject the interventions based on their own endorsement. This table of contents provides a structured overview of a document focused on the concept of empowerment in behavioral public policy. It includes discussions on the feasibility and need for empowerment, behavioral approaches to empower citizens, fostering core competencies, considerations for implementing these policies, and the limitations of empowerment.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
🌟🧠💪📝
#BOOSTING: Empowering citizens with behavioral science

New, freely available paper in Annual Review of Psychology.
PDF: tinyurl.com/boosting2025

For more: scienceofboosting.org

@arc-mpib.bsky.social @mpib-berlin.bsky.social

@annualreviews.bsky.social
#policy #behavioralscience

1/ 🧵👇
The image is the cover page of an article from the "Annual Review of Psychology" titled "Boosting: Empowering Citizens with Behavioral Science" by Stefan M. Herzog and Ralph Hertwig. It features a brief abstract, keywords, and publication details. The abstract outlines the concept of "boosting" as a behavioral public policy that emphasizes empowering individuals to make informed decisions, in contrast to "nudging," which subtly steers behavior. The abstract reads: Behavioral public policy came to the fore with the introduction of nudging, which aims to steer behavior while maintaining freedom of choice. Responding to critiques of nudging (e.g., that it does not promote agency and relies on benevolent choice architects), other behavioral policy approaches focus on empowering citizens. Here we review boosting, a behavioral policy approach that aims to foster people's agency, self-control, and ability to make informed decisions. It is grounded in evidence from behavioral science showing that human decision making is not as notoriously flawed as the nudging approach assumes. We argue that addressing the challenges of our time—such as climate change, pandemics, and the threats to liberal democracies and human autonomy posed by digital technologies and choice architectures—calls for fostering capable and engaged citizens as a first line of response to complement slower, systemic approaches. List with summary points: 1. Behavioral public policy garnered widespread attention with the introduction of nudging, which aims to steer behavior while maintaining freedom of choice. 2. Criticisms of nudging include that it does not promote agency and competences and that it relies—overly optimistically—on the presence of benevolent choice architects. 3. The proliferation of environments threatening people's autonomy, the slow pace of systemic approaches to tackling societal issues, and the intrinsic benefits of empowerment make empowering citizens an indispensable objective of behavioral public policy. 4. Boosting is a behavioral public policy approach to empowerment grounded in evidence from behavioral science that shows that humans’ boundedly rational decision making is not as flawed as the nudging approach assumes. 5. Boosts are interventions that improve people's competencies to make informed choices that conform to their goals, preferences, and desires. 6. In self-nudging boosts, people learn to use architectural changes in their proximate choice environment to regulate their own behavior—that is, they are empowered to adapt their own choice environments. 7. There are boosts to foster core competences in many domains, including finance, online environments, and health, as well as broader, overarching areas, such as motivation, risk, and judgment and decision making. Boosts should be part of a policy mix that also includes system-level approaches. 8. When implementing boosts, policy makers need to avoid the trap of individualizing responsibility and to be mindful that, due to differences in cognition and motivation, inequalities in the desirable effects across boosted individuals may emerge.
January 23, 2025 at 10:54 AM
stefanherzog.bsky.social
10/ Our paper outlines:
🔑 The limits of XAI.
🌳 The promise of interpretable models.
📊 A framework to study how AI fosters or hinders human learning.
A conceptual diagram illustrating an experimental framework for assessing the influence of various AI systems on decision-making performance. It outlines five conditions: 1. **Control (no AI)**: Decisions made without any AI assistance. 2. **Unassisted baseline performance**: Evaluation of decision-making without external tools. 3. **Interpretable AI**: Support provided by a transparent and inherently interpretable AI model. 4. **Black box AI**: Assistance from an opaque, black-box AI model. 5. **Black box + XAI**: Support from a black-box AI supplemented with explainability tools (XAI). Arrows indicate the progression toward measuring **unassisted follow-up performance** and **AI-assisted performance**, highlighting the experimental process. The flow captures comparisons of decision quality across different AI configurations.
November 20, 2024 at 12:25 PM
stefanherzog.bsky.social
1/ AI is everywhere—healthcare, justice, management—but most systems are opaque black boxes. 🤷‍♂️
Lack of transparency can—or often should—lead to mistrust, especially in high-stakes areas.

Enter **Explainable AI (XAI)** & **Inherently Interpretable AI (IAI)**.
A diagram illustrating four types of explanations for machine learning models in healthcare, divided into global and local explanations, and model-specific versus model-agnostic explanations. Top left (Global, model-specific): A simple decision tree used as an inherently interpretable global model. Users can see decision nodes and simulate how the tree categorizes new patients. Top right (Local, model-specific): A single decision tree applied to a specific patient, showing the Boolean rule used for categorization. Bottom left (Global, model-agnostic): A surrogate model, such as a decision tree, approximates the predictions of an opaque model (e.g., a neural network). Bottom right (Local, model-agnostic): Feature attribution methods like LIME or SHAP explain a neural network’s categorization for a specific patient.
November 20, 2024 at 12:25 PM
stefanherzog.bsky.social
🌟🤖📝 **Boosting human competences with interpretable and explainable artificial intelligence**

How can AI *boost* human decision-making instead of replacing it? We talk about this in our new paper.

doi.org/10.1037/dec0...

#AI #XAI #InterpretableAI #IAI #boosting #competences
🧵👇
Article information Title: Boosting human competences with interpretable and explainable artificial intelligence. Full citation: Herzog, S. M., & Franklin, M. (2024). Boosting human competences with interpretable and explainable artificial intelligence. Decision, 11(4), 493–510. https://doi.org/10.1037/dec0000250 Abstract: Artificial intelligence (AI) is becoming integral to many areas of life, yet many—if not most—AI systems are opaque black boxes. This lack of transparency is a major source of concern, especially in high-stakes settings (e.g., medicine or criminal justice). The field of explainable AI (XAI) addresses this issue by explaining the decisions of opaque AI systems. However, such post hoc explanations are troubling because they cannot be faithful to what the original model computes—otherwise, there would be no need to use that black box model. A promising alternative is simple, inherently interpretable models (e.g., simple decision trees), which can match the performance of opaque AI systems. Because interpretable models represent—by design—faithful explanations of themselves, they empower informed decisions about whether to trust them. We connect research on XAI and inherently interpretable AI with that on behavioral science and boosts for competences. This perspective suggests that both interpretable AI and XAI could boost people’s competences to critically evaluate AI systems and their ability to make accurate judgments (e.g., medical diagnoses) in the absence of any AI support. Furthermore, we propose how to empirically assess whether and how AI support fosters such competences. Our theoretical analysis suggests that interpretable AI models are particularly promising and—because of XAI’s drawbacks—preferable. Finally, we argue that explaining large language models (LLMs) faces similar challenges as XAI for supervised machine learning and that the gist of our conjectures also holds for LLMs.
November 20, 2024 at 12:25 PM
stefanherzog.bsky.social
We need MOAR, ending at...
December 1, 2023 at 4:45 PM
stefanherzog.bsky.social
Even MOAR MOAR manager
December 1, 2023 at 4:45 PM
stefanherzog.bsky.social
Even MOAR manager
December 1, 2023 at 4:44 PM
stefanherzog.bsky.social
MOAR manager...
December 1, 2023 at 4:43 PM
stefanherzog.bsky.social
Asking DALL-E for an image of a manager and asking to make it even more of a manager. 🧵

Starting with...
December 1, 2023 at 4:42 PM
stefanherzog.bsky.social
October 10, 2023 at 7:36 PM