Artificial intelligence applied to the ECONOMY

The Institute of Informatics is a key player in the digital transformation of companies, enhancing their efficiency, agility and competitiveness through targeted and practical research projects. Our labs contribute to the digitalisation of business processes and streamline the organisation of companies of different sizes. Our research focuses on areas such as business process modelling, automation, organisational optimisation and continuous process improvement. Our experts provide support in fields including:.

1. Business process modelling enables the visualisation of tasks, processes and resources. The aim is to streamline processes and prepare them for automation.BPMN is an industry standard for graphical representation and modelling. It sequences workflows, tasks, and information flows, providing a comprehensive view of business activities. This approach improves the efficiency of management, operational or support processes and creates a competitive advantage. BPMN also identifies effective and immature process areas in order to transform them into value-adding elements for the organisation. Implementing a BPM solution helps organisations adapt to change and align strategy with IT requirements. It is essential for modelling and optimising a company's business processes.
   
   
2. Business process automation (BPA) aims to increase productivity by automating repetitive tasks, shifting the focus to high-value activities that improve customer experience and productivity. An essential component of digital transformation, it relies on technologies such as artificial intelligence and machine learning.
   Robotic Process Automation (RPA) uses software robots to automate administrative tasks such as filling out forms or extracting data. This tool, which includes AI algorithms, is often used to streamline business processes.
   Workflow automation uses software to autonomously manage tasks and data flows following business processes. This can simplify processes, reduce manual work and increase operational efficiency.
   Intelligent process automation combines artificial intelligence, business process management and process automation to streamline and scale decision-making in companies.
   Low-code/no-code automation involves creating software applications that require little or no manual programming to accelerate the development of critical business applications.
   
   
3. Optimisation of the internal organisation: Introducing an agile organisation reduces the risk of errors and improves information flow. Processes are simplified to better meet customer needs.
   EPC software (Enterprise Process Centre) provides methods, tools and approaches for managing processes and other information models in a single repository. It supports companies in streamlining operations and increasing efficiency.
   Intelligent ERP systems use artificial intelligence and machine learning to automate and manage daily operations such as accounting, procurement, project management, risk management, and compliance.
   
   
4. Continuous improvement: Our teams assist companies in defining clear goals for the digitalisation of business processes. We identify key processes for transformation, develop project plans, and provide the necessary tools for successful implementation.

Virtual reality (VR) is an immersive technology that enables users to enter a computer-generated world using a VR headset, allowing interaction with virtual objects in a fully immersive environment. At our institute, VR research projects have been developed for applications in teaching and medicine. For young people with disabilities, VR provides a safe, simulated environment to acquire and practice skills that can later be applied in real-life situations. In the medical field, VR-based therapies enable doctors to detect conditions such as autism spectrum disorders in young girls at earlier stages..

Augmented reality (AR) overlays digital elements—such as images, videos, or information—onto the real world, enhancing the user's environment without isolating them. AR enables maintenance specialists to access instructions through AR glasses when repairing infrastructure. In medicine, this technology enables the real-time visualisation of critical information during a surgical procedure.

 

Mixed reality (MR) combines elements of the digital and physical worlds to enable interaction between them. This technology is used in various areas. For example, architectural firms can present a virtual model of the planned building at the future location. For medical students, MR offers them the possibility to explore virtual organs in 3D while working on them in a real laboratory.

The diverse applications of these technologies open opportunities for cost and risk reduction. Skills can be practiced in a virtual setting before being applied in the real world, making these tools invaluable for students, trainees, and companies launching new products. Additionally, these technologies provide a controlled environment for individuals with disabilities, autism spectrum disorders, or attention deficit disorders to engage with complex situations virtually before facing them in reality.

The vast amounts of data generated by companies and individuals represent a major economic challenge. Big data is an area of computer science that deals with evaluating, analysing and using this data. The aim is to enable companies and industries to analyse and use their data with the help of machine learning algorithms. 

The concept of big data is often defined by the "3 Vs":

• Volume: Data quantities that exceed the limits of conventional technologies and require processing processing through artificial intelligence.
• Variety: Data originating from various internal sources (e.g., transactions, customer databases, patient records, website clicks and logins) or external sources such as mobile apps or social networks.
• Velocity: The speed at which data must be generated and processed in real-time 

 

Our institute's research projects focus on aggregating, analysing, structuring and exploiting this data using customised AI algorithms tailored to the needs of our partner companies. Even structured data that can be stored and processed in a specific format often requires machine learning, as its volume poses a huge challenge for companies in terms of storage and processing. Our researchers also specialise in data mining (recognising data correlations and transforming them into usable information) and process semi-structured or unstructured data from diverse sources (texts, images, videos, etc.). In business intelligence we provide processes, technologies, expertise and applications that deliver critical insights to managers of SMEs or industrial companies. 

Data-supported decision-making is of great importance for all areas of society. 

• Medicine: Big data helps identify disease risk factors and develop diagnostic support tools.
• Energy: Big data enables real-time monitoring of electricity grid data for easier management.
• Agriculture, Finance, Transportation, and Business: Big data facilitates efficient resource management and the development of innovative products and services.

Natural language processing (NLP)

At our institute, we have developed expertise in natural language processing (NLP), a dynamic field that integrates computational linguistics, statistical modelling, and machine learning. This technology enables machines to recognise, understand, and generate human language, both in text and speech. Typical applications include voice-operated GPS systems, digital assistants, and speech recognition software. Businesses also rely on NLP to streamline operations and automate tasks such as spam detection, machine translation, chatbot interactions, and text summarisation.

By breaking down language into manageable components, NLP allows machine learning algorithms to interpret and respond to complex linguistic inputs. This process includes tasks such as:

• Speech Recognition (Speech-to-Text): Transforms spoken words into written text.
• Grammatical Tagging: Identifies parts of speech or syntactic roles within a sentence based on context.
• Semantic Analysis: Determines the precise meaning of words in specific contexts.
• Named Entity Recognition (NER): Identifies and categorises entities like names, locations, and organisations within a text.
• Coreference Resolution: Discerns when multiple words refer to the same entity.
• Sentiment Analysis: Detects subjective elements such as emotions, sarcasm, or irony in a text.
• Natural Language Generation: Converts structured data into coherent, human-like language.

Our researchers have developed an algorithm to detect fake news using machine learning and NLP. Our tools enable advanced text processing tasks, including syntax analysis, word segmentation, stemming, lemmatisation (reducing words to their root forms), and language unit segmentation. Additionally, these tools support the extraction of logical insights from text.

By combining statistical NLP, machine learning, and deep learning, we have optimised systems that automatically extract, classify, and label text data with exceptional precision and efficiency. The applications of NLP are extensive, ranging from enhancing user experiences to fostering innovation in data analysis and communication technologies.

Large Language Models (LLMs)

Large Language Models (LLMs) are systems trained on vast amounts of data to understand and generate natural language or other types of content, such as images and videos. These models have brought generative artificial intelligence into the mainstream, making it accessible to a wider audience. LLMs excel at capturing the context of a query to produce coherent and relevant responses, enabling tasks such as translation, text summarisation, answering questions, and generating content - including computer code.

Built on deep learning techniques, LLMs use neural network layers to predict the next word in a sequence based on the context of preceding words. While these outputs are not human conversations, they represent sophisticated prediction models driven by mathematical probabilities. Through training on extensive corpora, LLMs learn grammar, semantics, and conceptual contexts using techniques like zero-shot learning and self-supervised learning. 

Our researchers have expertise in key areas such as prompt engineering, fine-tuning, and reinforcement learning from human feedback (RLHF). These skills enable the refinement of LLM outputs to reduce biases and enhance reliability. Such AI technologies offer tangible benefits for businesses, from optimising chatbots and virtual assistants to synthesising data, creating content, and writing computer code. LLMs also contribute to accessibility through speech synthesis and the development of inclusive content for people with disabilities. 

Applications in SMEs and Industry

In SMEs or industrial contexts, LLMs can be used for a variety of tasks, including:

• Creating coherent and relevant content: emails, blogs, etc.
• Synthesising information: Summarising long texts or extracting insights from large datasets.
• Improving customer service: Answering customer enquiries via virtual assistants or chatbots.
• Supporting developers: Generating code, identifying errors, and addressing security issues.
• Analysing customer feedback: Understanding customer perceptions to improve reputation management.
• Providing translations: Delivering fluent translations across multiple languages.

Generative artificial intelligence is reshaping the media landscape as profoundly as the internet transformed society. In a world increasingly challenged by fake news, deepfakes, and disinformation, the media must rely on transparent and reliable tools to uphold its role as a provider of credible information. Research institutions can play a key role in developing and offering such tools. The Institute of Informatics aims to position itself as a centre of excellence in this domain, supporting the media industry by integrating European values of ethical and trustworthy AI into journalistic practices through research and tailored training programmes. 

Our researchers have worked on projects that leverage large language models (LLMs) to analyse user behaviour, helping media professionals better understand audience expectations. For example, by creating data-driven personas, journalists and editors can gain insights into readers’ needs and preferences. This technology also enables the efficient analysis of large datasets, such as extensive collections of newspaper articles.

One notable project involved analysing media coverage of sexist and sexual violence in the French-speaking Swiss press. The results were used to develop a decision-making tool. Ethical considerations are central to our approach, ensuring that LLMs in social research projects are used responsibly to create innovative tools built on robust and trustworthy models. 

To support the media industry, our institute provides the following resources:

1. Understanding recommendation algorithms 
   Recommendation systems are often opaque and create information bubbles. Our solutions focus on improving transparency. 
2. Fact checking and detecting fake news 
   The International Fact-Checking Network (IFCN) fights misinformation by promoting transparent and unbiased fact-checking. The number of certified experts is insufficient in view of the increasing demand. Artificial intelligence tools, including social network analysis and automated speech detection, can support this process.
3. Building confidence in public media 
   Media literacy is essential for informed participation in democratic societies. Misinformation, populist rhetoric and the conflation of opinion with scientific fact are growing problems. AI technologies capable of analysing and verifying arguments in real-time during political debates could help citizens evaluate the credibility of public figures.
4. Supporting public media to diversify sources of information 
   Recommendation systems, while promising for the media and their vast libraries of content, have their drawbacks. They often favour the preference of majority groups, marginalizing minority groups. These "algorithmic black boxes" limits transparency and accountability. Tools could be developed that offer other solutions for public media. 
5. A better understanding of algorithms
   Artificial intelligence is a highly powerful tool, but it often provides explanations that remain too opaque. New AI tools aim to help us better understand these systems. 

Transparency and explainability are essential to ensure privacy and security in AI systems. As AI progresses from simple automated tasks to complex autonomous decision-making, understanding these systems’ actions becomes increasingly challenging due to the complex algorithms involved. Responsibility for the development of autonomy and decision-making capabilities in AI lies with both the human creators, such as computer scientists, developers, and other stakeholders, and the AI systems themselves.

Our research teams include specialists in AI ethics and explainability, particularly in health, who try to gain insight into how decisions are made—such as why an AI might arrive at a specific diagnosis. This becomes more challenging in multi-agent systems, where several subsystems are combined (e.g. gossip learning). Even though multi-agent systems are cheaper and more reliable, their overall function and accuracy of decisions are harder to assess due to the varying architectures and algorithms used.

Ethical issues affect the entire research process, from modelling and data collection to decision-making and implementation. The goals of explainable AI (XAI) include:

• Identifying causal relationships between inputs and decisions;
• Determining whether the decision was influenced by an undesirable event or the lack of alternatives;
• Identifying the measures for evaluating the quality of a decision.

Both external and internal explanatory approaches are used. External methods simulate system inputs and observe results to generate explanations, often using statistical analysis or decision trees. Internal approaches focus on designing autonomous systems with self-reflective properties, enabling them to monitor and explain their behaviour as they make decisions rather than retroactively. 

Explaining highly predictive models is challenging because they learn relationships between training data (patient information, production metrics), outcomes (diagnoses, final products), and features (biological data, product characteristics). The complexity of these models, often involving huge volumes of data points and parameters, makes them resemble "black boxes" in terms of interpretability.