Exploring Age Adaptation of Conversational Systems


What if conversational agents would be able to adapt in the way they interact with humans? Unfortunately, state-of-the-art AI systems currently lack sophisticated adaptation capabilities. Our long-term goal is to develop artificial agents that can adapt to individuals/user groups at any level (age, expertise, language style, etc.) and that are perceived as trustable by users.

In the following project, we focused on users of different age groups, and asked:

  • Can some degree of adaptation be achieved by training a state-of-the-art system with data targeted to specific age groups?
  • Does this lead to differences in the perceived degree of anthropomorphism, social presence, trust, appreciation, and comprehensibility of the message?

Below you can find an open-source demo that showcases the results of the project regarding the first of these research questions.

This project is a collaboration between the Dialogue Modelling Group at the Institute for Logic, Language and Computation (ILLC) and the Persuasive Communication group at the Amsterdam School of Communication Research (ASCoR) at the University of Amsterdam.


Illustrative image generated with DALL·E 2

System adaptation

Try it yourself. The state-of-the-art language model would first receive a prompt with corresponding criteria to generate an adapted output. Select a combinations from the options below to see what kind of adaptation different systems produced:


Test yourself

To test whether adaptations produced by the language models are detectable, we paired outputs from different models together and asked human evaluators to judge which one sounded more like coming from an older or a younger speaker. In total, we collected 9,000 judgments from 467 different participants.

How would you do yourself? Try to guess the age group of the following examples and see if your guesses agree with our human evaluators.

Which one of the two outputs sounds like a text that could be written by a younger speaker?

Find out more

The work received funding from the University of Amsterdam’s Research Priority Area Human(e) AI and from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 819455).