Saturday, May 17, 2025

AI roadmaps for Malaysia and Southeast Asia: Sam Majid weighs in

Enterprise IT News met up with NAIO's Sam Majid to gather his thoughts about development of AI roadmaps for Malaysia and Southeast Asia

Enterprise IT News recently met with Shamsul Majid (Sam), CEO of the National AI Office (NAIO), shortly before NAIO convened a meeting of 210 stakeholders to advance Malaysia’s AI roadmap. This 5-year roadmap was initially established by predecessors like MOSTI, and since NAIO’s establishment five months ago, the Office has been tasked with seven deliverables.

The top deliverables that piqued EITN’s interest include:

  • Defining the AI action plan for 2026-2030
  • Developing an AI regulation framework through which NAIO can contribute to national-level regulation alongside policymakers (currently ongoing)
  • Promoting AI adaptation with focus on six sectors of high economic priority

More information about the roadmap is available on the NAIO website.

During our conversation with Sam, he shared insights not mentioned on the website regarding:

  • Regional large language models (LLM)
  • Data sharing
  • Safety in the ecosystem

Regional large language models (LLMs)

“If Malaysia has more scientists and research and development capability, we can produce our own models and have better control of algorithmic bias,” Sam explained. “Otherwise, we become users of popular off-the-shelf models that come from China or the US.”

Given this reality, Sam emphasised the importance of guidance on controlling, understanding, and anticipating model outcomes. Data quality becomes crucial, and practitioners need to monitor training data carefully.

“After data has been loaded, further testing of AI results is necessary to ensure they comply with local guidelines. These steps are essential from a practitioner’s perspective.”

Sam believes local standards are needed for stricter adaptation of LLM models to reflect local culture and values. “A response from an LLM that is acceptable for Western societies may not be appropriate for Asian societies.”

Southeast Asian regional LLM development

Singapore is leading efforts to develop an open-source family of large language models supporting 11 Southeast Asian languages while better understanding regional contexts and cultures. These languages include English, Chinese, Indonesian, Malay, Thai, Vietnamese, Filipino, Tamil, Burmese, Khmer, and Lao. The initiative aims to create locally relevant AI tools instead of relying on Western models.

Sam noted that such initiatives address two key areas: sovereignty and local values.

“Including 11 languages means preserving sovereignty. Otherwise, eleven countries would interact with AI using only English. With proper investment, time, effort, and people, the initiative is populating a model with these other languages.”

“This ensures non-English languages have presence and reality in the digital space during interactions with AI,” Sam added, noting that a NAIO working group is addressing technological aspects of sovereignty.

Without this act, when a data request occurred, the standard response was either ‘No’ or ‘let me consult my supervisor’—leading to a cycle that went nowhere.

Sam Majid

The second area—the local and domestic ecosystem—involves language and cultural nuances. “The language the model produces can still be insensitive to local contexts.” Sam explained that sensitive topics like race, religion, and royalty must be appropriately moderated.

“Local community nuances must be incorporated. That’s where standards and guidelines come in.”

Data Sharing

As of 28th April 2025, the Data Sharing Act 2025 came into force in Malaysia. Sam noted, “Without this act, when a data request occurred, the standard response was either ‘No’ or ‘let me consult my supervisor’—leading to a cycle that went nowhere.”

“Now, public sector entities can request or receive data requests among themselves. This applies not just to the federal government but will extend to state and local authorities.”

Importantly, Sam pointed out that a committee must evaluate requests and implement safeguards for data privacy and cybersecurity, ensuring data sharing requests don’t default to automatic rejection.


AI and Ethics development in Malaysia

The national guidelines on AI governance and ethics are available on NAIO's website. NAIO recognises these guidelines suit national-level discussions but may not fully address industry-level conversations or smaller implementations. For example, the government's Digital Department (JDN) has created specific guidelines for state and local authorities.

"We will conduct a survey to gauge response to these first codes of ethics established by MOSTI," Sam said.

The goal is creating an ecosystem where AI can "advance faster" by having robust safety mechanisms, similar to how well-designed braking systems allow cars to drive at high speeds safely.

"When implemented safely, AI can progress faster. Without safety measures, regulations, and a secure ecosystem, most people resist AI advancement due to uncertainty about its direction."

Sector-specific use cases

Beyond the six sectors identified for AI focus, NAIO works on AI development for specific sectors upon request, including legal, natural resources, and defence. “We collaborate closely with these sectors, participate in their meetings, and provide guidance on ethics, standards, sovereignty, and innovation sandboxes.”

Sam also highlighted two sovereign LLMs: YTL AI Labs’ Ilmu 0.1, a Malaysian LLM representing a major step toward national AI self-reliance, and Mesolitica, a Malaysian startup specialising in LLM training with cloud-hosted generative AI capabilities.

Cat Yong
Cat Yong
Cat Yong is Editor-in-Chief of Enterprise IT News, a regional news website which began in Malaysia circa 2011. A common theme in all of her work - opinions, analysis, features and more - is how technology and innovation drives business and outcomes. A career tech journalist for 22 years, her work has evolved to also encompass narratives of tech powering human potential.

Read more

News