Uncovering the Future of AI: SEA-LION’s Localised Revolution via IBM and AI Singapore

How will the future of artificial intelligence (AI) unfold? It is possible that we will notice little change as the transformation occurs. However, once the changes become more obvious, we may struggle to remember or comprehend how we functioned in the past. 12 years ago, the late entrepreneur Paul Baran advised the Defense Advanced Research Projects Agency (DARPA), the central innovative research unit of the US Department of Defence (DoD). According to his proposal, DARPA should extend its initiative to develop machine reasoning to include the ability to understand and assess metaphors. Baran’s vision foresaw a commander instructing a computer: ‘To capture the opponent’s nearest artillery, marshal our forces as swiftly as leopards on the hunt.’ Despite being an esteemed pioneer in computer networking, Baran’s proposal did not gain much attention at that time. The concept of machines comprehending metaphors seemed as improbable as machines speaking English; few researchers believed it to be achievable any time soon. In fact, at IBM, the AI division made a deliberate decision not to address Baran’s proposal or delve into the question of teaching computer systems to understand human language.

Together, this partnership signals the advent of a new future of a more personalised AI. IBM and AI Singapore (AISG) have embarked on a strategic collaboration to trial and share the Southeast Asian Large Language Model (SEA-LION) – a first of its kind departure in AI development that represents a significant milestone on humanity’s journey towards an open and accessible future for AI.

OPEN: Tailoring AI with a Local Touch

Localizing Generative AI Applications for Southeast Asia

This partnership centres on the idea of localising AI, embedding it with the languages and cultures of Southeast Asia. One of the founding technologies of this endeavour is an open-source software called SEA-LION, developed by AISG. SEA-LION, with its agile, vector-based design, is built to be compact, lightweight, flexible and fast, making it easy for developers and data scientists to build GenAI applications with a local sensibility.

OPEN: Powering Innovation with SEA-LION

IBM's Role in Shaping Bespoke AI Solutions

The AI and data platform provided by IBM, called Watsonx, will be where SEA-LION is being tested. XSEDE plans to add SEA-LION to its Digital Self-Serve Co-Create Experience (DSCE) library, an entry point for developers that provides resources and tools to experiment with AI. Developing and adding entities to the DSCE library enables a kind of ecosystem of innovation: co-creators are free to ‘play’ with the AI models that are relevant to their particular business needs. By incorporating SEA-LION into the DSCE library, IBM isn’t just testing an AI model; it’s testing the ambition to cultivate an open-ended AI system: one where, instead of having to blindly hope that it works, you can choose the right model for the right use.

OPEN: Embracing Multicultural Capabilities

Reflecting Southeast Asia's Diverse Cultural Landscape

The training data corpus at the heart of SEA-LION is appropriately polyphonic and epic, with 981 billion language tokens (spoken and written) in English and in the amazing linguistic diversity of sound systems and scripts that encompass various Southeast Asian languages, Chinese. It is trained to navigate through the corridors of uncertainty of culture and language. As a city-state with a multicultural aura, Singapore’s ambition to foster AI that embodies its multicultural demographics is truly something to marvel at.

OPEN: Crafting Custom-Made AI Foundations

The Drive Towards a Customized AI Model

But the more I read the ambitions to ‘create a custom-made foundation model’ – as Leslie Teo at AISG put it – it feels like the architecture of the AI world is evolving not just to be powerful, but to be astonishingly, intimately human. The collaboration plans not just to offer up SEA-LION but to create use cases that show how it can aid in solving the real-world problems companies face today, all in the hopes of bringing faster, safer, more responsible AI to companies around the world.

OPEN: Governing AI with a Forward-Thinking Approach

Ensuring Compliance and Ethical Use of AI

As AI is accelerating the demand for its governance is increasing Many of the concerns regarding the application of AI can be resolved by a shared commitment. This partnership recognises the imperative of ensuring that, in utilising SEA-LION – an AI-driven decision-support tool for business that aims to equip ESG – the effective governance of AI implementation is achieved, thus facilitating a safe deployment of AI in line with the best industry practices, while ensuring its compliance, appropriateness and ethical implications.

OPEN: The Path to an Open AI Ecosystem

Choosing Flexibility Over Restraint

Catherine Lian of IBM ASEAN put it well: The SEA-LION initiative is one step towards an open AI system that emphasises flexibility, enabling organisations to choose the AI model that meets their own requirements.By reducing barriers to accessing and configuring AI to match local needs, the partnership finally bringing the future of AI closer to fruition in the ASEAN region.

Conclusion: Opening Doors to AI Customization

SEA-LION is more than an IBM-AI Singapore technological partnership; it is a vision for an open AI future with local diversity, customisation and ethical use at its heart, and a beacon for other developers and businesses in Southeast Asia to follow the path towards AI solutions that are not just powerful and efficient, but deeply resonant with the regional culture and languages.

About OPEN

What Lies at the Core of SEA-LION’s Success

From this commitment to openness stems the main advantage of SEA-LION and the novel success of the IBM-AISG partnership: an open AI ecosystem that makes SEA-LION so customisable, flexible and inclusive that it can be moulded to the diverse and often unconventional needs of a heterogeneous region like South East Asia. It is through an open-source approach that we can see how AI technology could (and should) develop, so that it isn’t merely intelligent but is accessible and adaptable to our multifaceted existences in this part of the world. From this perspective, the AI of the future is more than intelligent, it’s open.