Today’s society is constantly adapting to change and is immersed in a great technological development. For its part, artificial intelligence (AI) is bringing a very positive impact thanks to the enrichment of capabilities it brings to human beings. However, we must take into account the challenges it poses.
The current social debate on the regulation of the use of technology by minors, the consequences of its abuse and the possible vulnerabilities that this group may present, invite reflection. Telefónica’s Human Experience Hub team has proposed several sessions with the aim of raising awareness and awakening the critical spirit of the digital products consumed by minors, especially the role of artificial intelligence in adolescents.
What does the study on AI and teenagers consist of?
An outreach, experimentation and exploration study on AI was carried out with 12 and 13 year old teenagers, from which interesting conclusions were drawn about how this technology affects young people. The session took place in the Experience Design Lab, a perfect scenario to put teenagers in domestic situations and to analyze their behaviors and habits.
During this experience, the students participated in different activities related to AI:
- They learned about AI: the session talked about the risks and benefits it offers, with the aim of promoting its responsible use.
- They contributed to shaping new responsible AI use cases: they were able to test innovative items that have been designed for speech recognition and elaborated concepts related to the accessibility of a TV experience through LLMs (long language model).
- They explored a technological innovation environment: they visited Telefónica’s Experience Design Lab and experienced an innovation methodology in first person giving their opinions, impressions and ideas.
People at the center of the study
One of the key issues in this research with adolescents was the ethical use of AI. From the design of the research we wanted to set an example so that this pillar underpinned every part of the activities. Consequently, an effort was made to ensure that the research was respectful and beneficial to all participants.
During the proposed activities, the objective was established to reflect the ethical commitment from two aspects: on the one hand, the benefits and risks of Artificial Intelligence were transferred to teenagers, and on the other hand, we talked about how to make a responsible use of AI.
From the Discovery team of Telefónica Innovación Digital, we observed during the dynamics that, given the need to build AI systems, the creation process must always ensure compliance with the following ethical principles of the company:
- Fairness and freedom from bias.
- Transparency and explainability of the systems.
- People as a priority.
- Privacy and security from the design of the systems.
It is also worth mentioning the emergence of new problems related to new ethical regulations, security and privacy, so it would be advisable to have a team specialized in addressing these issues, in order to build fairer systems free of any discrimination.
How to humanize technology?
To begin with, a theoretical presentation on AI was made with the aim of informing and making aware of its presence in the world, to serve as an aid to identify the use of this technology in day to day life. Examples of the different problems that can currently be solved with this advance were also presented, as well as the projection of other ideas about fields that are more distant from their realities, such as robotics or expert systems.
Subsequently, the group worked on the improvement of a Deep Learning system whose mission was the vocal detection of adolescents in the same age group. This work pointed out the importance that data plays in the construction and expansion of AI and also showed how easy it is to extract its own data.
How to build a data learning model
Subscribe to our newsletter!
Find out about our offers and news before anyone else
During the session the teens used the keyword detection mechanism (“OK Aura”) to initiate communication with devices. These models should be robust to diversity, specifically, accessible to adolescents.
The models are built with data sets, which are used during training processes and for evaluation. There are many factors that play a fundamental role in model behavior, such as variations in devices, environments, and speaker characteristics. All of these affect the audio that is produced and, therefore, the behavior of the model. The most common way to provide robustness to stimulus diversity is to have diverse data.
For the construction of a model accessible to all, including adolescents, data collection is needed. In this way, models could be trained and evaluated to minimize bias toward any one group. The data should include audio and metadata. Metadata accompany the audio and minimally describe its characteristics. Specifically, age, gender, accent and a unique speaker identifier should be stored.
This is fundamental information for measuring and working on one’s own biases. The speaker identifier should be anonymized to make it possible to know how many different speakers there are, as well as to guarantee the stratification of different speakers in training and evaluation.
On the one hand, the names of the users and their anonymous identifiers should be stored in parallel. On the other hand, user data should be stored anonymized, i.e. only with their anonymous identifier. This would guarantee, firstly, the security and privacy of the data collected and, secondly, allow users to exercise their right to be forgotten. Also, with this type of ordering, new versions of the models could be generated with different versions of the data.
Main conclusions of the study on the use of artificial intelligence among adolescents
Taking into account the research with artificial intelligence and teenagers carried out a few months ago, differences can be observed regarding the number of AI tools that teenagers have been able to identify, as well as in the main concerns declared around the technology.
AI tools
On the one hand, in the most recent study, teens have been able to recognize a higher number of AI tools and functions among which filters of some RRSS are included.
On the other hand, impersonation along with the creation and dissemination of fake information and videos seem to rise as the main concerns of this group, probably influenced by the cases of image manipulation using AI that have been reported in the news.
Voice assistants
Regarding the use of voice assistants, it can be observed that teenagers seem to be predisposed to use them when they are in emergency situations or temporary disabilities that limit their mobility or manual access to their mobile devices. In these cases, the voice assistants mentioned are mostly those integrated in smartphones, probably due to the association of these devices with communicative actions such as calls.
In addition, through the analysis of the statements and interactions with voice assistants and other applications, it is observed how the concern for voice privacy is given by the content that is expressed with it and not so much by the biometric data that can be obtained from it.
AI-generated content detection
With regard to the detection of AI-generated content, there are indications that there is a certain distrust of the technology and that human judgment is being imposed as the main method of identification .
The recognition of AI-generated content, by means of another tool trained for this purpose, is not a priori among the main options for teenagers. At this point, and being aware that it is a current problem in society, Telefónica has worked on a tool capable of detecting whether images, videos or audios have been manipulated and with what program, VerifAI.
In short, the sessions were of great value for all involved, helping to train young people to face the challenge of AI today, while humanizing technology by reducing biases and making it more accessible.
Also, the results of the research have helped to have more and better information for the design of more human experiences. This post has been written in collaboration with Pablo Gómez Guerrero and William Fernando López Gavilánez from Telefónica’s Discovery area.
