_edited.png){kind=logo}
The advent of the Smart City has opened up new opportunities for citizen-centred learning. These smart cities harness technology to improve citizens' quality of life, but they also offer new opportunities for citizen-centred learning. By collecting and analysing data in real-time, citizens can access relevant information about their immediate environment. Through Steam City, we want to offer an initiative that will give rise to interactive educational applications and provide an innovative experience.
The SteamCity proposal
The "Bot Buddy Adventure - Designing a Chatbot for Urban Fun" SteamCity project is designed to enable students to grasp the concepts associated with artificial intelligence (AI), and thus demystify it while tackling concrete challenges linked to urban life.
The activity presented explores the design and implementation of an autonomous conversational agent. It can be used to provide users with relevant data linked to their geographical location.
To achieve the goal of the activity, several use cases for artificial intelligence will be used:
Voice recognition to acquire the user's request
Speech synthesis to communicate with the user in natural language
Use of language models to obtain personalised information
In order to make this activity usable in a school context from the secondary school level upwards, a visual graphical programming tool has been used. This will make the programming concepts involved more accessible while avoiding the rigour and syntax specificities inherent in a text-based programming language.
Context of the experiment
In this experiment, as mentioned above, the students will use a language model (such as chatGPT) to obtain personalised information:
The city in which the user is located (based on GPS data such as latitude and longitude). The most relevant response possible to a query formulated by the user.
The use of a language model in block programming software is booming.
Until now, the most common uses of artificial intelligence in an educational context, accessible using block-based visual programming tools, have focused on the implementation of machine learning models. In particular, a supervised learning model can be used to categorise data in the form of photos, sounds or postures…
In this example, we wanted to show another approach that can be used quite simply: processing data using a language model to obtain personalised information.
Course of the experiment
The proposed progression is divided into 4 main stages:
Stage 1 - Data collection
In this first stage, students are introduced to the principles of geolocation and GPS. They also learn to use automatic natural language processing (NLP) techniques to analyse text and voice data. This will enable them to extract the information they need to respond to user requests. Based on this data and the definition of an effective prompt, a language model is mobilised to obtain precise results.
This phase introduces an innovative approach using a language model in block programming software. Students explore the subtleties of formulating a query in order to take advantage of a language model.
They will then be able to obtain the city corresponding to the user's GPS position simply and efficiently.
Stage 2 - Displaying data
Once the user's position has been obtained, the students will be able to process the data, store the predictions (responses) given in the language model and formulate a response for the user.
In order to question the 'creative' side of a language model, our chatbot will be able to suggest that the user invent a story to entertain them. To do this, the program will need some information about the protagonists. Once this information has been collected and stored in variables, a prompt will be generated to produce the personalised story.
Stage 3 - Data analysis and learning
At this stage, the students analyse the data collected to evaluate the performance of the conversational agent. They explore how AI can be used to understand the user's request and link this request to geolocated information in real-time. Formulating precise queries and understanding the context remains crucial.
Stage 4 - Using the data to change behaviour/improve the initial situation
Finally, the students will be looking at another type of use of artificial intelligence through text-to-speech.
By exploring different solutions for making the chatbot's responses more natural and convincing, they will select, configure and test the generation of artificial voices.
Lessons learned
This experiment offers an opportunity to discover and mobilise several areas of application of artificial intelligence, ranging from voice recognition, the mobilisation of a language model and text-to-speech synthesis in order to communicate with the user.
Students will be given a real-life opportunity to develop an operational prototype through a project-based approach.
They will also be encouraged to question the strengths and limitations of language models, understand the importance of formulating queries and develop critical thinking skills in evaluating the performance of artificial intelligence.
Particular attention may also be paid to the hallucinations that a language model may generate and the biases it may have.
Conclusion
This learning experience offers a holistic approach to integrating advanced artificial intelligence concepts into the practical context of smart cities. It prepares students to meet future challenges by combining the latest technologies with concrete applications to improve the lives of every citizen. We hope that this experiment will help train the next generation of citizens who will be able to shape the future of smart cities. Happy experimenting!