iiinova — where your digital self keeps evolving

Reposted by Florian Keppeler

The People Lab @thepeoplelab.bsky.social · 05/10/2026

What does it take to recruit and retain a strong public workforce, increase access to government programs, and get good evidence into practice? We’re hiring researchers to help us answer questions like these! Learn more and apply 👉 peoplelab.hks.harvard.edu/careers/

013

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

🔗 doi.org/10.1111/puar... @pareview.bsky.social Wilma, PhD (Pretty helpful Dog) and #postdog 🐾 @au.dk End🧵

000

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

➡️ Writing a public sector job ad? Next to the degree requirements, consider saying explicitly that the job is about listening and serving. The authors argue that, otherwise, you may not attract the people that citizens hope to meet at the counter. 5/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

➡️ Citizens rank it differently. Asked about the ideal public servant, citizens put dedication, serviceability and honesty on top, and expertise last. The two rankings barely overlap. 🤷 4/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

➡️ Job ads sell credentials. Expertise and dedication show up in almost all ads. Responsiveness and honesty trail far behind. 📋 3/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

Do public employers look for the same thing citizens do? A new study by Melissa Satzger Grøn and @dominikvogel.bsky.social analyses state job ads and find: 2/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 04/10/2026

#Wilmasreview Wilma 🐶 has an impressive CV: expert nose, years of sniffing experience, efficient removal of food leftovers. But that's not why we love her. It's the head tilt. She listens 👂 1/🧵

110

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

💡 Currently, it looks like we judge human–AI teamwork before anyone has had a chance to learn it. This is a key direction for future research. 🔬 Wilma, PhD (Pretty helpful Dog) and #postdog 🐾 @au.dk 🔗 doi.org/10.1073/pnas... End🧵

doi.org

Human learning is an understudied but promising lever for boosting human–AI synergy | PNAS

Humans collaborating with AI hold the promise of achieving superior outcomes compared to either acting alone (i.e., human–AI synergy). However, the...

033

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

➡️ The twist is that AI explanations on their own backfired. Explanations without feedback went with clearly worse teamwork, while explanations plus feedback went with better teamwork. People need feedback to learn when to trust what the AI tells them. 🤔 6/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

➡️ The studies that did tell people how they were doing tended to show better human–AI teamwork. The evidence is tentative, but it points in the same direction. 📈 5/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

People did one short session with no practice rounds, and only 10 of the 74 studies ever told participants whether their answers were right. 😶 4/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

➡️ A recent meta-analysis found that humans and AI working together usually do no better than the stronger of the two working alone. But most of those studies never gave people a chance to learn. 3/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

#Wilmasreview How did Wilma 🐶 learn to jump on large, felled trees? Not from a manual. She learned it one try at a time and with feedback (🦴). So she was puzzled to hear that most studies of humans working with AI never tell people whether they got it right. 1/🧵

110

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

➡️ The roles changed quietly. Caseworkers became helpdesk staff teaching citizens to self-serve, or screen-level bureaucrats handling exceptions, rejections and complaints. Real discretion moved upstream to designers, IT departments, private vendors who barely appear in the studies at all. 🧩

101

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

5/9 ➡️ Staff does comply and cope. Rule bending, double documentation, rearranging the guided questionnaire to have a real conversation, using private phones, giving extra time to the citizens who need it most. Not sabotage: an attempt to keep human judgement alive in a system that left no room. 🐾

101

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

In contrast, in regulation-oriented organizations such as the police or customs, the same kind of tool tends to enable judgement by providing officers with better information. 🔀

101

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

➡️ The context can be more important than the tool itself. In service-oriented organizations such as benefits offices, employment services and citizen service centers, digital tools tend to restrict discretion.

101

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

When public employees are limited by the digital tools they must use, they try to work around, too. A systematic review by Justine Marienfeldt systematically reviewed the research of how digital tools have changed decision-making at the street level:

101

Florian Keppeler @flokeppeler.bsky.social · 20/09/2026

#Wilmasreview Wilma 🐶 has accepted that there are fences 🚧. A fence doesn't ask, it doesn't make exceptions, and it doesn't care that the neighbor's cat is getting away. So, Wilma dug a hole ... a burdensome workaround. 🕳️ 🧵

101

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

📄 doi.org/10.1093/jopart/muag027 Wilma, PhD (Pretty helpful Dog) and #postdog 🐾 @au.dk End🧵

doi.org

Validate User

000

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

💡 For public managers: signalling "just do what the government wants" does not buy loyalty. It pushes out precisely the people most committed to democratic values, i.e. the ones you would want to keep as a bulwark. Signalling guardianship keeps them in the building. 📚 6/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

➡️ It runs on values: In the US sample, the effect of a manager signaling compliance depended heavily on whether staff shared those values; however, guardian signals worked regardless. Managerial authority ends where an employee's personal interpretation of public service begins. 🧭 5/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

➡️ The manager matters: When the boss signals "always comply with the government," people head for the door. For example, in the Norwegian electoral-fairness scenario, the share considering exit rose from 53% under a "guardian" manager to 67% under a "responsive" one. 🚪 4/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

➡️ Under democratic backsliding, civil servants became less willing to carry out the government's wishes, more willing to obstruct them, more likely to raise concerns internally and publicly, and more likely to consider quitting.🫷 3/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

A new study by Mads Pieter van Luttervelt, Mathias Rask Østergaard-Nielsen, @martinbaekgaard.bsky.social, Zuzana Murdoch and Benny Geys ran studies with Norwegian and US civil servants and find: 2/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 13/09/2026

#Wilmasreview Wilma 🐶 knows the drill when entering the dog forest. She's waiting for the person one level higher in the hierarchy to give a signal. 🦮 Public employees also turn to their managers when politicians start bending the rules of democracy. 1/🧵

111

Florian Keppeler @flokeppeler.bsky.social · 06/09/2026

➡️ But it moved some people a lot. Republicans were markedly less tolerant of fraud when the abstract word "welfare" was used than when a specific program such as unemployment insurance was named. People with personal experience of means-tested programs showed a similar pattern. 🔍 4/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 06/09/2026

➡️ Naming the program did not move the average. Whether people were asked about vague "welfare" or a concrete program, the overall trade-off looked much the same. 🤷 3/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 06/09/2026

Welfare policy has the same challenge. Politicians justify paperwork by pointing at fraud but every hurdle also shuts out people who genuinely qualify. A new study by @drelizabethbell.bsky.social and @sebjilke.bsky.social asked people in the U.S. to actually put a number on that exchange rate.🔬 2/🧵

101

Florian Keppeler @flokeppeler.bsky.social · 06/09/2026

#Wilmasreview Wilma 🐶 takes treats very seriously. There are two ways to get it wrong: hand a treat to a dog who hasn't earned it, or refuse one to a dog who has. Humans lose a lot of sleep over the first mistake. Wilma is much more worried about the second. 🦴 1/🧵

120

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

💡 For policymakers: before publishing salary bands, reorganising teams or announcing cuts, deliberately decide which comparisons to make visible, and put as much effort into designing the procedure as you do into achieving the desired outcome. 6/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

➡️ Not only do institutions respond to fairness preferences, they can also shape them. Schools, workplaces and communities influence what people perceive as fair, and repeatedly enforced norms can become internalised motivations. 🏛️ 5/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

➡️ The procedure carries much weight. Neutrality, respectful treatment and proper explanations can render an unfavourable decision more acceptable. However, when the outcomes are too severe for fair procedures to restore legitimacy, this remains an open question. ⚖️ 4/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

Research suggests that inequality among colleagues and peers in education is considered considerably more unfair than inequality in society at large. 🔍 3/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

➡️ Who do you compare yourself to? Your coworkers, neighbours, or the national income distribution? Pay transparency rules, team structure and workplace design all affect how you make these comparisons. 2/🧵

100

Florian Keppeler @flokeppeler.bsky.social · 30/08/2026

#Wilmasreview Wilma 🐶 has food in her bowl. Pluto next to her does not. Wilma sees absolutely no problem here... 🍖 An article by Ernst Fehr and Keyu Wu argues the old question, whether people care about fairness, is settled. Here is what we need more research about: 🔬 1/🧵

110

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

💡 The authors argue another way to remove the emotional cost is to remove the encounter itself: once-only data, automatic enrolment. 📚 6/🧵

140

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

➡️ Participants with lower education and without German citizenship made fewer errors when the form arrived pre-filled. 🎯 5/🧵

120

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

➡️ Pre-filling the form did not calm anyone down. Emotional arousal was the same in both groups, and overall error rates did not drop either. But people were clearly more satisfied with the procedure afterwards. 🥴 4/🧵

120

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

➡️ Bureaucratic costs shows on a face. Anger, confusion and frustration all rose sharply the moment the paperwork started. Frustration was linked to more mistakes in the forms. 😤 3/🧵

150

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

A new study by Fabian Hattke, Judith Hattke, David Hensel, @pamherd.bsky.social, Dr. Janne Kalucza, @donmoyn.bsky.social and Rick Vogel had citizens complete a real social housing application in the lab, while facial coding software tracked their emotions.🔬 2/🧵

182

Florian Keppeler @flokeppeler.bsky.social · 23/08/2026

#Wilmasreview Wilma 🐶 gets offered a helping hand. She takes a good long look at it, still not convinced. The help is real, the worry too. 🤨 Something similar happens when governments try to make citizens' life easier. 1/🧵

240

Florian Keppeler @flokeppeler.bsky.social · 16/08/2026

➡️ It doesn't matter what the topic is, the evidence comes typically from a richer country. The authors argue this isn't proof that Global North evidence is simply better; it's largely a story of who can afford to make their research and documents visible in the first place. 🐾 5/🧵

101