Ongoing
„Take it apart”: how object presentation affects creativity
Mariusz Urbański, Vira Obshanska, Marta Skorodziievska, Kateryna Shyiko, Aleksandra Ulaszewska
The Alternative Uses Test (AUT) is one of the most widely used tools for measuring divergent thinking: participants are asked to come up with as many unusual uses for an everyday object as they can. But does it matter how the object is presented to them?
In this project, we compare three presentation formats: the object’s name alone, a photograph of the whole object, and a photograph of the object disassembled into its component parts. The idea is that seeing an object broken down may help overcome functional fixation – the tendency to think of things only in terms of their typical use.
Our pilot study showed that participants who saw disassembled objects produced more original responses without generating more responses overall, suggesting that visual decomposition changes the quality rather than the quantity of ideas. We also found that generating more ideas increases average originality but decreases the share of truly unique responses – a trade-off with implications for how AUT results should be scored and interpreted. Finally, different presentation formats shifted the types of uses people came up with, pointing to the role of stimulus design in shaping the space of creative solutions.
We are now running a larger, pre-registered study to confirm these findings and develop practical recommendations for designing and interpreting creativity tasks.
The influence of task sequence on deductive visuospatial reasoning
Mariusz Urbański, Róża Sateja (in collaboration with Agnieszka Nowik)
When we reason about spatial relationships – deciding, for instance, whether objects are arranged consistently with a set of given assumptions – we may rely on different mental models. Some arrangements naturally suggest a linear order, others a circular one. But does the type of model we have just been using carry over to the next task?
In this project, we ask whether prior exposure to tasks that favour a particular normative model (linear or circular) primes people to evaluate conclusions in subsequent, model-neutral tasks. Participants work through three blocks of spatial reasoning problems: one using a Likert-type scale (suggesting a linear model), one using geographical names (suggesting a circular model), and one using abstract coloured symbols (with no implied model). By fully counterbalancing the order of blocks, we can test whether the model activated by one type of material spills over into the interpretation of structurally identical problems presented with neutral symbols.
The project combines online behavioural experiments with a laboratory eye-tracking study. Eye-tracking data allow us to examine whether priming effects are reflected in gaze patterns, fixation durations, and the number of shifts between premises and conclusions, offering a window into the cognitive strategies participants adopt and how these strategies change depending on what they have just been doing.
Evaluating the quality of abductive explanations
Mariusz Urbański, Aleksandra Wasielewska
When faced with a surprising or puzzling event, we naturally try to make sense of it by a process known in logic and philosophy of science as abduction. But what makes one abductive hypothesis better than another? And can we reliably measure this?
We gathered a plethora of data in our previous project on abduction, „Modelling of Abductive Reasoning” (NCN Sonata-BIS grant DEC 2013/10/E/HS1/00172). In this project, we develop a systematic procedure for assessing the quality of abductive explanations produced by participants in a detective-style scenario task. Participants are presented with an ambiguous situation and write a report reconstructing what happened and why, based on limited, partly conflicting evidence. The resulting reports vary widely: some propose creative, well-supported hypotheses that address all the gaps in the story, while others leave key questions unanswered or contain internal contradictions.
To evaluate these explanations, we use a multi-phase approach. First, competent judges provide free-form qualitative assessments of the reports, which we then analyse thematically to identify the dimensions along which people intuitively distinguish good explanations from poor ones, such as completeness, coherence, clarity, and the integration of evidence with conclusions. These dimensions are subsequently operationalised as rating scales and used in a structured evaluation round, allowing us to measure inter-rater reliability and relate expert judgments to objective linguistic and structural features of the texts. The overarching goal is to arrive at an empirically grounded, multi-dimensional account of what constitutes a good abductive explanation.
Syllogistic validity and existential assumptions
Karol Wapniarski, supervised by Mariusz Urbański
When faced with a surprising or puzzling event, we naturally try to make sense of it by a process known in logic and philosophy of science as abduction. But what makes one abductive hypothesis better than another? And can we reliably measure this?
We gathered a plethora of data in our previous project on abduction, „Modelling of Abductive Reasoning” (NCN Sonata-BIS grant DEC 2013/10/E/HS1/00172). In this project, we develop a systematic procedure for assessing the quality of abductive explanations produced by participants in a detective-style scenario task. Participants are presented with an ambiguous situation and write a report reconstructing what happened and why, based on limited, partly conflicting evidence. The resulting reports vary widely: some propose creative, well-supported hypotheses that address all the gaps in the story, while others leave key questions unanswered or contain internal contradictions.
To evaluate these explanations, we use a multi-phase approach. First, competent judges provide free-form qualitative assessments of the reports, which we then analyse thematically to identify the dimensions along which people intuitively distinguish good explanations from poor ones, such as completeness, coherence, clarity, and the integration of evidence with conclusions. These dimensions are subsequently operationalised as rating scales and used in a structured evaluation round, allowing us to measure inter-rater reliability and relate expert judgments to objective linguistic and structural features of the texts. The overarching goal is to arrive at an empirically grounded, multi-dimensional account of what constitutes a good abductive explanation.
The Social Perception of Moral Agency: How Different Actors Are Judged for Their Decisions
Wojciech Zięba, supervised by Mariusz Urbański
This doctoral project examines how people evaluate moral decisions depending on who makes them. Rather than focusing solely on whether a decision is right or wrong, the research investigates how the type of moral agent – such as a human, an artificial intelligence system, or a person occupying a particular social role – influences the way observers judge the same decision.
The project is motivated by the growing presence of AI systems in areas such as healthcare, transportation, law, and finance, where automated systems increasingly participate in decisions that affect human lives. Previous research suggests that moral judgments are shaped not only by outcomes but also by emotions, social norms, and perceptions of the decision-maker’s mind and intentions. The central question of the project is therefore whether identical decisions are evaluated differently when attributed to different kinds of agents.
Empirically, the research combines quantitative and qualitative approaches. Participants evaluate a series of moral dilemma scenarios – based on modified versions of the classic trolley problem – in which the same decisions are made by different agents. This design allows the study to examine how factors such as human vs. artificial agents, gender, professional roles, or underlying ethical principles shape the moral perception of a decision.
The project also investigates how individual differences among observers influence these judgments. In particular, it analyzes the role of beliefs about the uniqueness of human nature, religiosity, prior experience with AI, and cultural attitudes toward technology in shaping moral evaluations.
By integrating perspectives from moral psychology, cognitive science, and AI ethics, the research aims to develop a broader model of how societies attribute moral legitimacy, responsibility, and trust to both human and artificial decision-makers. The findings may contribute to understanding public attitudes toward emerging technologies and inform discussions about the ethical design and social acceptance of AI systems.
Finished (within the Reasoning Research Group):
- Non-cooperative strategies in the Loebner contest
- Test & Research Center for Vienna Test System
- Standardized Epistemological Understanding Assessment
- Deductive Flexibility Test