Leuphana: How human may a chatbot appear?
At a glanceStarting point
Measure comparably, without making the chatbot rigid.
The development of scientifically valid chatbots posed several complex challenges: until now, chatbots in academic studies were designed to be very static and used pre-selected, already prepared answers in order to achieve comparability. This comparability is a central criterion for quantitative studies in academia: it is the only way to compare the extent to which differences in the chatbot trigger different perceptions in students.
Solution
Four variants that differ in exactly two characteristics.
Leuphana University, represented by Prof. Dr. Monika Imschloss and Dr. Lennart Seitz, came to scoreprise.AI with the challenge of having its study supported technically. The aim was to examine the extent to which chatbots can show strong variability and at the same time comparability in an academic context.
The solution lay in developing four chatbots based on a 2x2 design: slow and impersonal, fast and personal, fast and impersonal, and slow and personal. The technical challenge: the chatbot was not allowed to speak arbitrarily on the basis of a prompt, but had to understand the user, remember their characteristics, respond to them personally and adapt its language style. More than 500 students were involved in the project and thereby created a scientifically valid data basis.
Set-up
Four variants in a 2x2 design
Slow and impersonal, fast and personal, fast and impersonal, slow and personal: four chatbots that differ in only two characteristics.
Requirement
Variable and yet comparable
The chatbot was not allowed to answer arbitrarily, but had to understand the user, remember characteristics and adapt its language style, without losing scientific comparability.
Study
More than 500 participants
In the end, each participant received a personal recommendation for their course, tailored to their interests and language style.
The aha moment
A slow, considered answer has a stronger effect on the user than a fast one.
This gives the user the feeling that the AI instance really understands their challenge, thinks about it and develops its own solution. High speed is expected of AI today, but it is not always the best way to speak with a user.
Result
What the study showed for AI employees.
The findings from the Leuphana study clearly show: successful AI implementation depends decisively on how well the communication between people and AI is designed. Through this project, scoreprise.AI gathered valuable experience that feeds directly into the development of human AI employees.
The speed of the answers, the degree of personalisation and the adaptation to individual communication styles are decisive factors that scoreprise.AI has validated through academic research. Through the experience from more than 500 user interactions, scoreprise.AI knows how AI employees must be designed so that they really feel human and are successfully integrated into existing teams.
More use cases.
All use cases
M&A advisory
KERN Unternehmensnachfolge: invoices, proposals and NDAs from a Teams message.
Networking and professional development · Project
Leaders of AI: 60 specialists build five specialised AI teams in one day.
Media and education · Project
campus a: Journalistic feedback for Austria's first nationwide school newspaper.
Next step
Your use case, shown on a comparable one.
Cancellable monthly · Operated in Germany