Three ways foundation models promise to put generative AI to work for governments


Generative AI and foundation models are an essential part of the future of Artificial Intelligence (AI). GovInsider speaks to experts from IBM to learn more about how governments can put foundation models to work at the service of the public sector.

Foundation models are set to accelerate digital government endeavours in the coming years. Image: Canva

The Singapore Government is moving decisively to tap on generative AI – the country currently plans to build 100 generative AI solutions in 100 days and has already developed plans to move all public sector chatbots onto large language models (LLMs).


LLMs are just one of many foundation models – machine learning models that are trained on large unlabelled datasets, which can then be adapted to perform more specific tasks – that promise to make AI capabilities easily available to governments and enterprises. Right now, tech companies like IBM are training foundation models on a wide array of data, including code, geospatial data and IT events data, according to a new IBM report.


GovInsider speaks to Cristina Caballe Fuguet, Senior Partner & Vice President, Global Government Leader, and Florian Breger, Vice President Civilian Government, Global Industries, IBM Technology, to understand the three key ways that foundation models can put AI to work for governments.

1. Improving citizen experiences


First and foremost, government leaders should prioritize the consideration of AI and generative AI use cases that improve citizen experiences. When finetuned with relevant domain knowledge, agencies can use generative AI to build applications that can support agencies in providing better, faster information and services to its citizens, Caballe and Breger said.


Take for example conversational AI chatbots powered by generative AI. These tools have the potential to manage large volume of citizen inquiries and interactions, allowing public sector agencies, and its limited resources, to focus on more value-added tasks, says Caballe.


These interactions can then be used to further train the conversational models and continuously improve customer service, bringing more relevant and personalized information to each citizen based on previous interactions.


How does AI and foundational models work on a public facing solution?


Recently, IBM partnered with Wimbledon to develop an AI solution that could:

  • Automatically identify key moments and produce highlight reels from thousands of hours of video
  • Add AI-generated spoken commentary to highlight reels
  • Provide data-driven predictions for players’ performance to fans

This AI platform was built on IBM’s enterprise-grade AI platform, watsonx. First, developers tapped on, a data store which enabled them to pool together a wide range of data sources to inform their machine learning models, while filtering out personally identifiable information, profanity, and hate speech.


Next, the team selected a large language model from, IBM’s studio for building, training and deploying generative AI models. Then, they finetuned the AI model with Wimbledon’s domain expertise, including tennis-specific terms and concepts.


Similarly, agencies can tap on foundation models from to serve their unique citizen communication needs, from chatbots to parliamentary highlight reels.

2. Raising productivity, transforming operations


Second, foundation models can support government organisations in improving productivity and transforming operations in the back end. This generates a tremendous opportunity to exploit the vast amount of data for public use or to improve the internal decision-making process, resulting in increased productivity.


As a matter of fact, according to recent studies, governments that embed generative AI solutions within their everyday operations can expect to see up to a 40 per cent increase in productivity, Caballe points out.


In the United States, IBM has partnered with NASA to build an open source geospatial foundation model that can pore through petabytes of satellite data to track deforestation, predict crop yields, and monitor greenhouse gas emissions, says Breger.


In the public sector, similar generative AI can support agencies in sifting through and summarising vast amounts of documents, applications and files, he explains. Even civil servants with extensive experience in programming can embrace these benefits. With a simple language request can point them to concrete sources within the documents, helping the AI to remain explainable and trustworthy.


Watsonx.governance, one of the tools available later this year in the IBM watsonx studio, is a tool that will help organisations create responsible and transparent AI workflows with security and control throughout the entire AI lifecycle.


“You don't want government services powered by AI to have ‘hallucinated’ responses, or to get different answers for the same question,” he notes, highlighting the importance of having trusted and explainable AI for highly regulated industries like governments. In June, the Monetary Authority of Singapore worked with IBM on a responsible AI toolkit to help financial institutions assess fairness and transparency when using AI.


Caballe adds that IBM offers a consulting practice that can support organisations in developing and deploying AI tools responsibly, and establishing a culture around the safe handling of AI, with full consideration of the potential risks and unintended effects. This practice can also help leaders identify the best use cases for AI within their teams.

3. Modernising applications


Finally, generative AI can support government agencies in modernising their technology applications, such as upgrading legacy applications to run in a cloud-native way. This is a key goal that Singapore’s GovTech has set for itself this year, as GovInsider previously reported.


“In government, there is currently a high priority to run large digital transformation programmes. AI can offer productivity gains of around 30 per cent when it comes to application modernisation,” says Caballe.


This can range from helping organisations automatically generate code when migrating legacy applications to the cloud or building playbooks for automation code.  For example, the Watson Code Assistant can help developers generate secure and workable automation code for the Red Hat Ansible Automation platform, as GovInsider reported previously.


Once such code is generated, developers can review and modify the code if necessary. This can radically simplify the workload of government IT teams that oversee complex multi-cloud and hybrid cloud environments.


This year, Australia’s Digital Transformation Agency announced a partnership with IBM to drive the modernisation of government services with IBM’s tech offerings, including generative AI.


But how can agencies begin to embrace the full benefits of foundation models? Caballe and Breger suggest that government leaders start leading their organizations in identifying initial use cases and pilots that can provide tangible value and go into production quickly as soon as possible.


Once such quick gains have been made, agencies can learn from the process to scale their AI strategy across their organisation with trusted tools like watsonx.


In addition, IBM is currently conducting workshops with government agencies around the globe to help them discover next steps and potential pathways forward in implementing generative AI. It also offers pilots of its products, including watsonx.governance, for these organizations to experiment with the benefits of IBM’s open, trusted, targeted and empowering approach to generative AI.