Moving towards open AI

Artificial Intelligence is emerging as one of the most disruptive technologies, promising to bring about radical changes across all value chains, the economy and society as a whole. For businesses, it offers the potential to create value and solve complex problems.

Artificial Intelligence (AI) encompasses a wide range of different technologies, such as text detection, speech recognition and generation, translation programmes, chatbots and image classification. AI can be used for a wide range of applications such as quality control, supply chain optimization, customer service and fraud detection, boosting the performance and results of the industries and organizations that use it. However, companies often find themselves dependent on their suppliers' solutions, with their IT systems sometimes obsolete and a lack of in-house expertise. So, how can they embrace AI over the long term? How can they make the most of the changes AI brings?

AI in business: between hopes and constraints

The benefits of AI are manifold. Despite this, we have to admit that it has not yet been adopted massively. According to a survey by PwC Luxembourg (Use of Data Analytics and Artificial Intelligence in Luxembourg - 2023 edition), the level of maturity of data governance practices is still relatively low in Luxembourg.

The obstacles to its implementation are often linked to the company itself: management's level of understanding about the added value of AI, existing resources and expertise, the availability and quality of data, as well as concerns about current regulations, such as GDPR, and those to come, in particular the EU AI Act. External factors are also hindering wider and effective adoption. These include the fear of becoming too dependent on a single supplier (vendor lock-in), as well as ongoing technological and regulatory developments.

In fact, creating value from AI requires not only advanced AI and Machine Learning algorithms, but also vast quantities of data and its quality, managed and exchanged between the various stakeholders, as well as the integration of AI into user experiences. For example, the first three versions of OpenAI's GPT were very promising models of AI, but had little impact until GPT-3.5 was trained on a much larger amount of data and integrated into a chat user experience that was easily accessible without any technical training.

LIST paves the way for reliable and trustworthy AI

The BigTech companies are investing massively in AI, publishing new and improved models, libraries and environments to develop application solutions. This observation has led the Luxembourg Institute of Science and Technology (LIST) to focus on the application of AI to create economic and societal value in a responsible, reliable, trustworthy and sustainable manner.

One of LIST's main objectives is to reduce the technological risks faced by partner companies and organizations. As a neutral player, LIST is totally independent when it comes to technology, preferring open standards and interoperable solutions. Indeed, LIST partners have the opportunity to decide, with the help of scientists and engineers, which combination of technologies best meets their various needs, such as ethics, regulatory compliance, cost reduction, performance and scalability. Depending on the partners' needs and maturity in AI, LIST helps them to focus their actions on various key elements: data, models and services for the various stakeholders.

To do this, LIST offers a toolbox where partner companies can test before investing in a specific AI solution. For example, partners can test, in a no-code environment, existing resources, flows and/or basic components, either directly via the Institute's infrastructure, or by exporting this processing to their premises or to a cloud infrastructure.

The Institute also provides a low-code framework that facilitates the specification and design of AI-enhanced software components, based on the company's data and requirements, to be integrated into the company's information system. This platform is then responsible for generating the running system and its monitoring environment.

In the AI Sandbox, developed by LIST, partners can test different AI models against a number of biases, compare their own AI models with others available on the market, or be assisted in selecting the AI model that best meets their criteria.

Digital twins can also be deployed, enabling organizations to move AI from specialized, siloed use cases to transformative initiatives that generate value and competitive advantage.

This set of tools, processes and best practices for managing and leveraging data helps organizations manage their data. It's about treating data with the same diligence and care as traditional physical or financial assets. This includes internal activities such as governance, management, quality assurance, integration, privacy and security management, and data lifecycle management to ensure the reliability, security and usefulness of the data for the organization, and therefore fuel the AI value creation cycle.

All these projects are interconnected and organizations have the choice of selecting the relevant elements according to their needs and maturity in terms of AI. As the environment is multimodal, any type of data, from sensor data to time series, to heterogeneous data catalogues/collections, including cross-sources, textual data but also images and sound can be processed.

By working in a market-neutral way, LIST meets the specific requirements of digital sovereignty and strategic autonomy of private companies and certain sectors such as public administration, critical infrastructures (energy, transport, telecommunications), healthcare and defence.

Find out more about Generative AI, listen to Tech Advantage's Conversation #5 - Generative AI: Veni, vidi, vici?

Article translated from Paperjam.lu