II.a Context, policy and trends - BDV Big Data Value Association

After the two so-called “AI winters” (70’s and 90s of previous century) and the boom in-between (80’), AI experienced significant progress around 2010 and onwards, due basically to the combination of three factors: (i) advanced machine learning (ML) techniques, and more specifically deep learning that combines several processing layers addressing different levels of abstraction in neural networks that try to mimic to certain extent the behaviour of the human brain, (ii) access to a huge amount of data worldwide, thanks basically to digitalisation in all domains, proliferation of Internet of Things (IoT) devices, and data available on the web, social media platforms, … to accurately train these complex networks, and (iii) availability and access to more powerful and faster computing infrastructures to train and fine-tune the ML models. Deep learning models have become increasingly sophisticated, overcoming human performance in many fields (computer vision, Natural Language Processing (NLP), etc.). Additionally, the combination of several neural networks led to systems that were not only capable of identifying, classifying, predicting, or prescribing information, but could also generate content (generative adversarial networks). When these systems have been equipped with some “memory capabilities” (the so-called transformers[1]), generative AI[2] (GenAI) has exploded with conversational chatbots able to generate from a prompt text a wide range of content (text, images, videos), which is getting always more and more accurate and powerful. These novel technologies are therefore substantially impacting, and in perspective also revolutionizing, all aspects of our daily lives: education, work, industry, etc.

In this regard, AI is a crucial pillar for the evolution of metaverse and virtual worlds. Coming from previous versions of the web (“read-only web” and “social web”)^[3], Web 3.0 (“Semantic Web”) already integrated ground-breaking technologies (virtual and augmented reality, AI, IoT and faster wireless communication) to provide connected, immersive and autonomous web experiences. We are now transitioning to Web 4.0, where physical and digital worlds will seamlessly blend, enabling more intuitive and immersive experiences, leading to next-generation virtual worlds with potential users’ experiences that are challenging to predict. These virtual worlds open many possibilities for human-machine interaction, education and training, healthcare and well-being, manufacturing and the public sector[4].

Despite the advantage acquired by the USA and China in most of the fields mentioned above, during the last years, the European Union (EU) has made a great effort to reduce the technology gap starting a set of strategic initiatives. The European AI Alliance^[5] was established in 2018 to foster an open policy dialogue on AI with around 600 stakeholders. One of the main objectives of the Alliance was to drive the High-Level Expert Group on Artificial Intelligence^[6], which provided, among others, ethics guidelines for Trustworthy AI, a method to assess Trustworthy AI (ALTAI), and policy and investment recommendations. In 2021, the EC published its Coordinated Plan on AI^[7] to accelerate investments, define strategies and align AI policies among its member states. The European approach to trustworthy AI^[8] (risk-based approach) paved the way for the first EU regulation on AI (AI Act), adopted by the European Parliament in March 2024, which will become fully applicable 24 months later. The AI Act provides a common regulatory and legal framework for AI to ensure that the development and use of AI are done according to European values and principles.

Finally, in January 2024 the EC has published the so-called “AI innovation package”[9], with the aim of supporting European startups and Small and Medium Enterprises (SMEs) in developing trustworthy AI that respects EU values and rules. The package is composed of (i) an amendment to EuroHPC JU regulation to set up AI Factories that adapt or acquire AI-dedicated supercomputers for fast ML and training of general-purpose AI models, (ii) an AI Office for the coordination of AI policy and supervising implementation and enforcement of AI Act, (iii) the GenAI4EU initiative to support the development of novel use cases and emerging applications in 14 industrial ecosystems in Europe.

Substantial investments have complemented this regulatory effort. The AI4EU[10] H2020 project started in January 2019 to develop the first European AI On-Demand Platform (AIoD), i.e., and ecosystem to share resources, tools, knowledge and algorithms related to AI with the broadest possible audience in the research domain and civil society. This action was complemented by a set of projects aimed at building a European network of AI centres of excellence^[11] and reinforcing the service layer of the AIoD platform^[12]. The activity has continued with new coordination and support projects in the new Digital Europe Programme (AI4Europe and DeployAI). The Digital Europe Programme has also included the co-funding of AI Testing and Experimentation Facilities (TEFs)^[13], specialised large-scale reference sites where technology providers can get support to test their latest AI-based technologies in real-world environments.

Given the new Horizon Europe programme, BDVA, euRobotics, EurAI, Ellis, and Claire joined forces and created the AI, Robotics and Data (ADRA)[14] association. This association became the private counterpart of the EC for the new ADR partnership, covering an umbrella of projects funded under the Horizon Europe Pillar II—Cluster 4: Digital, industry, and space.

Another important activity happened in July 2023, which the EC published a strategy and proposed actions for virtual worlds and Web 4.0[15]. The document proposed the 4 pillars that would guide the strategy (skills, business and ecosystem, government and governance, standards) and main technological trends that constitute the basis of virtual worlds (connectivity, computing, data, digital twins (DTs), Distributed Ledge Technology (DLT) / blockchain, Extended Reality (XR), AI, electronics and photonics, Internet of Things (IoT), AIoT, digital identity, standards). As part of this strategy, the EC has proposed the establishment of a new partnership on virtual worlds, starting most likely in 2025.

Several projects funded under different calls in Horizon Europe^[16] address several aspects of virtual worlds. More specifically, the Openverse Project^[17] aims to create a knowledge base on European virtual worlds, establish a community of stakeholders, test user co-creation and extended reality, explore challenges, and generate standards, technology roadmaps and recommendations. Finally, many projects under the Emerging Enabling Technologies^[18] topic in Horizon Europe address relevant aspects and produce exciting results that can contribute to developing virtual worlds.

Although the landscape is evolving at an extraordinary speed, all the previous activities and initiatives are pivoting around some key trends, among which those more relevant for this document are:

GenAI boomed after the appearance (end of 2022) of conversational chatbots that rely on large language models (LLMs) to generate rich and comprehensive content just from prompt text. Additionally, the combination of GenAI with AI foundation models, trained with broad data such that they can be applied across a wide range of use cases, is revolutionising the use and application of this technology.
Relying on these large AI models, we are closer to Artificial General Intelligence, which is supposed to match or surpass human capabilities in a wide range of tasks. For now, general-purpose AI is already a reality, again using AI foundation models trained on a broad set of unlabelled data and used for different tasks with minimal fine-tuning^[19].
Emerging industry-driven foundation models focused on specific sectors and pre-trained models for particular purposes involve collaborative efforts from several companies working on a concrete field, reducing the need for too many customised models for individual needs.
LLMs are also evolving on a fast path. Different subtypes are appearing: multi-modal approaches that use images, audio, etc., to extend the training data, a combination of models with traditional retrieving systems (Retrieval-Augmented Generation) to enhance their accuracy and reliability, LLMs as reasoning engines to be eventually used as part of other products, small language models aimed to more straightforward tasks with the same performance as LLMs but less complex and computational intensive.
Synthetic data, whether generated artificially (GenAI) or via simulation (DT), is experiencing an incredible upsurge (the global synthetic data generation market was valued at USD 316.11 million in 2023 and is anticipated to grow at a rate of 34.8% from 2024 to 2033[20]) as a way to address diversity, privacy protection, quality, and scarcity of data.
Despite this evolution of AI capabilities, keeping the human in the loop is seen as more and more critical, both in the training and the application phases of the models, enabling a loop where humans provide feedback (e.g. reinforcement learning from humans, also considering human objectives) and suggestions to algorithms, thus leading to human-centric AI, where AI technology would complement, support and improve human abilities, but not replace them, ultimately augmenting human intelligence.
The emergence of Edge AI has enabled AI algorithms to reside in devices that can analyse and process data without depending on a server, thus providing an immediate response. Additionally, AI agents benefit from this evolution by performing their tasks more accurately without human intervention. Suppose these agents can collaborate and coordinate. In that case, they can solve problems more efficiently than individual agents towards Distributed AI, which offers many advantages concerning the traditional centralised approach: security and privacy, accuracy, scalability, robustness, and efficiency.
AI is applied in many sectors, with many scenarios and use cases implying tangible benefits for companies, organisations, and society.
The emergence of innovative and transformative technologies that sustain virtual worlds enables a societal evolution at an incredible speed.

[1] P. Xu, X. Zhu and D. A. Clifton, “Multimodal Learning with Transformers: A Survey,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 12113-12132, Oct. 2023, doi: 10.1109/TPAMI.2023.3275156.

[2] K. S. Kaswan, J. S. Dhatterwal, K. Malik and A. Baliyan, “Generative AI: A Review on Models and Applications,” 2023 International Conference on Communication, Security and Artificial Intelligence (ICCSAI), Greater Noida, India, 2023, pp. 699-704, doi: 10.1109/ICCSAI59793.2023.10421601.

[3] https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/next-generation-virtual-worlds-opportunities-challenges-and-policy-implications-2023-07-03_en

[4] AIOTI White Paper „Edge IoT Industrial Immersive Technologies and Spatial Computing Continuum, Release 1“.

[5] https://digital-strategy.ec.europa.eu/en/policies/european-ai-alliance

[6] https://digital-strategy.ec.europa.eu/en/policies/expert-group-ai

[7] https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review

[8] https://ec.europa.eu/commission/presscorner/detail/en/ip_21_1682

[9] https://ec.europa.eu/commission/presscorner/detail/en/ip_24_383

[10] https://cordis.europa.eu/project/id/825619

[11] https://cordis.europa.eu/programme/id/H2020_ICT-48-2020/en

[12] https://cordis.europa.eu/programme/id/H2020_ICT-49-2020/en

[13] https://digital-strategy.ec.europa.eu/en/activities/testing-and-experimentation-facilities

[14] https://adr-association.eu/