University of Limerick’s Dr Mairtín Cunneen examines the opportunities emerging from AI and the potential of immersive and virtual worlds.
The year 2023 will probably be famous as the time when AI became a global and social phenomenon. Mostly, because billions of people around the world from all areas of life, ages and backgrounds are now actively engaging, communicating and talking with AI. For many users, the engagement may feel like a conversation with another human-like intelligent being with agency and personality.
Behind the appearance of sophisticated human-like intelligence, there are countless graphics processing unit (GPU) servers housed across thousands of data centres enabling the functionality of large language models (LLMs) which learn and replicate patterns in human language use mostly from scrapped online data.
The rise of ChatGPT
OpenAI launched ChatGPT to the world on 30 November, 2022 and ChatGPT-4 on 14 March 2023. A part of the global success of ChatGPT relates to the platform model of a free-to-use service available as a webpage and an app.
The sophistication of the LLM architecture was apparent from the beginning in how it generated quick-fire responses and simultaneously answered millions of questions on countless topics. Whether it was general knowledge, science or academic topics, the LLM appeared as a fountain of human knowledge and wisdom.
While some concerns arose regarding the veracity, reliability and accuracy of the responses, the scale of use continued to grow. So much so that the use of LLMs became seemingly normalised in a short time and the use of ChatGPT surpassed all other platforms, even the previous records set by social media platforms such as TikTok. ChatGPT became an AI platform that achieved 100 million users in just two months after launch and continues to be engaged by billions of global users.
With its growing popularity, further questions arose regarding how the model would fabricate answers that were not factual at all, but often had the appearance of reliable information sometimes supported with fake citations and supporting sources. Such responses were called ‘hallucinations’ and this outcome meant that a disclaimer had to be attributed to the outputs stating that they were not always accurate or factual.
From text to images
At the same time as the global attention on LLMs, there was also immense attention on generative AI applications for image and video generation. Platforms such as DALL-E, Stable Diffusion and Midjourney amazed the world by creating photo-realistic images and videos to text prompts. Some of these became viral images capturing much of the media attention on AI in 2023.
Generative AI models are moving toward far more sophisticated models with unimaginable numbers of parameters producing more sophisticated responses. The present phase of development concerns multimodal models, which can combine different modalities of data such as text, images, audio and video to create increasingly sophisticated outputs.
From images to video
In February 2024, OpenAI launched Sora, which is described as “an AI model that can create realistic and imaginative scenes from text instructions”. The architecture of Sora presents the evolution of AI to provide more sophisticated models to process far more complex data as well as potentially combining different modalities.
The videos generated by Sora are impressive, as in a little more than a year, AI has evolved from text, audio and image to complex photo-realistic complex and rich video content. An interesting outcome concerns the challenge of determining what is real and what is AI-generated. While the world continues to embrace and come to terms with such ubiquitous and sophisticated uses of AI, other pressing social and ethical questions need urgent consideration.
The future of AI is immersive realities
Many questions relate to what impacts, opportunities and challenges will continued AI development introduce over the coming months and years. For example, given the dramatic global impact of LLMs, generative AI and photo-realistic video creation, what do the next two years of innovation potentially offer?
Following the trajectory of multimodal AI, one of the most impactful possibilities over the next year or so relates to how AI could create photo-realistic immersive realities and virtual worlds. The potential to prompt AI to create and recreate spaces is something radical, potentially wonderful and also challenging.
Imagine people all over the world asking AI to create an immersive space to relax in, socialise in or learn in. The complex immersive environments will be like the real world, as for example, OpenAI’s Sora states: “The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world”.
My first thought would be to prompt the AI to create an immersive environment to learn Plato’s dialogues by engaging in conversation with Socrates and Plato in Ancient Athens. To achieve this, the model needs the required data and in this scenario, the data is probably available, and the likelihood is that using available data the environment could be created to a sophisticated level of photo realism. Hence, using Plato’s dialogues and countless supporting academic texts, Socrates and Plato appear as engaging participants and could communicate in a fashion (English language version) that one might be familiar with having read Plato.
Socrates and Plato chatting in mixed reality. Image generated by AI.
The key here is that the AI is creating an experience that may look a lot like a real-world experience, but it is one created for my context and may be very different from a more accurate depiction. The amazing potential is that anyone can prompt the AI to create an immersive reality of their choosing and if the AI has the required data, the model will probably do an amazing job of creating a sophisticated experience.
Imagine a recreation of Einstein’s lectures or learning history by being a participant in an immersive recreated event. Until now the creation of such a digital experience would require significant costs and human expertise in programming, design and development. The potential uses are astounding from education and learning to scientific experimentation and social engagement. All aspects of the world could be created in immersive realities by AI.
There are countless other opportunities such as a routine engagement with an AI physiotherapist who could assess and guide your rehabilitation or a personal teacher in an immersive environment where practice-based learning can be developed in new ways.
Risks and ethics
An important opportunity for the potential of AI recreating immersive and virtual worlds for people relates to the need to now focus on anticipating and addressing the potential risks and ethical questions before such technologies become widely available, commercialised and embedded into societies.
A key question is how can we learn from the mistakes of social media and create social-facing platforms less focused on adtech, behavioural analytics, and disinformation and enable more socially positive spaces? In addition, there are important questions and challenges to be addressed regarding access and inclusion.
There are also serious concerns regarding how to address bias embedded in the datasets, especially those that rely on historical and public data. Algorithms and AI applications are known to amplify embedded bias further which presents significant risks of discrimination and harm. Moreover, if such spaces are commercial spaces, there is a need for more transparency regarding the business models that will determine their design and use.
A fundamental question concerns privacy and the challenge of effectively supporting privacy rights. Also, what about the environmental and climate costs of providing AI to billions of people? Should people be made more aware of the energy and resource costs of millions of GPUs, servers and countless data centres? It is important to now address such questions and try to anticipate the opportunities and risks of AI and the next phase of innovation.
The use of AI and the development of immersive realities are big business. The EU Commission highlighted the strategic importance of immersive reality innovation in its Virtual Worlds and Metaverses strategy in July 2023. It expects a potential of 860,000 new jobs by 2025 and estimates the virtual worlds market to grow to more than $800bn by 2030.
The EU is aware of the challenges of addressing such questions and part of its strategy is to support research on anticipating and improving the social value of immersive realities, virtual worlds and metaverses.
Significant technological developments are already occurring with mixed reality (MR) gaining global attention in February 2024 with Apple’s launch of its Vision Pro and supporting visionOS. The VisionPro is an MR device but is referred to as a spatial computer by Apple.
With all if these possible innovation opportunities and ethical questions in mind, University of Limerick (UL) is carrying out research to assess future social and commercial opportunities, as well as the risks of developing immersive realities. The specific focus of the research is on culture and the arts as a social and commercial facing MR innovation.
The research is part of the XTREME project, an EU-funded project by the Horizon Europe programme, led by ITU Copenhagen and consisting of 15 academic and industry partners. The cross-discipline team is responsible for carrying out research to anticipate how to make the next generation of MR and immersive experiences more ethical and trustworthy.
There are lots of opportunities emerging from AI and the potential of immersive and virtual worlds. The UL research will focus on addressing some of the most challenging social and ethical questions relating to the emerging phase of AI innovation and immersive worlds.
Dr Mairtín Cunneen is an associate professor in data analytics and risk governance in the Accounting and Finance Department in the Kemmy Business School, University of Limerick.
Find out how emerging tech trends are transforming tomorrow with our new podcast, Future Human: The Series. Listen now on Spotify, on Apple or wherever you get your podcasts.
