Putting trustworthy AI at the forefront and in practice

By Jill Arul

Despite the benefits of artificial intelligence (AI), it’s common for users to feel hesitant when entrusting advanced technologies with data. Dubbed the ‘uncanny valley of the mind’, having an unnerving feeling about technology is understandable—after all, AI is something almost human but not quite.

Despite any underlying fears, AI has made its way into almost every aspect of our lives. Face recognition biometrics allow us to quickly access our phones, smart appliances remind us when our fridge is low on food and online virtual nurses are on hand to quickly answer our health questions. This is especially true in Singapore where AI has taken centre-stage. In 2019, the nation topped charts by investing US$68 per capita in AI development and implementation, overtaking China by US$47 per capita.

If AI can instil deeper confidence in consumers and businesses, it has the potential to transform Singapore’s digital landscape and ensure growth in all sectors through automation and prediction. At the ATxAI Conference on 14 July 2021, speakers from leading AI companies from around the world shared their use of trustworthy AI at the Industry Showcase: From Principles To Practice.

Keeping AI accountable

To trust a system, users need to know that it works and will generally not cause harm to users. At the panel discussion, speakers agreed that AI should be an inclusive space to ensure that businesses can be held accountable through explainable and user-friendly systems.

Ms Diana Paredes, Co-Founder and CEO of British regtech firm Suade Labs, believes that people from all walks of life should be educated in the basics of AI. Armed with a fundamental understanding, users should be equipped to harness the power of AI in any industry as well as demonstrate the knowledge needed to explain the technology and keep companies accountable.

Also working to bring AI to the masses is Dr Sinuhe Arroyo. As the Founder and CEO of Singapore-headquartered AI solutions provider Taiger, his team is developing ‘no code tools’ like the Omnitive Studio that will allow anyone to develop beneficial AI without coding expertise. With these tools, Taiger hopes to democratise AI and drive mainstream adoption, allowing everyone fair access to the power of technology.

Having been featured in IMDA/PDPC’s Volume 2: Compendium of Use Cases, Taiger makes use of explainable models when implementing Omnitive and swiftly deletes all data when it is no longer necessary, noted Dr Arroyo.

Joining forces for trustworthy AI

In the spirit of encouraging trustworthy AI practices, speakers agreed that regulation and the pursuit of trust is ultimately a multi-stakeholder effort. From governments to businesses to providers, each player in the AI ecosystem has a role to ensure that the technology is practical, beneficial, safe and secure to use.

For instance, Head of Innovation at local AI solutions provider UCARE.AI Dr Tan Kuan Pern, explained that while tech suppliers should be responsible for basic security, the government or an independent panel can step in to develop robust regulations that protect both users and businesses. As AI makes its way into various industries, each sector should be involved in the regulation of AI models and data protection.

Ms Arunima Sarkar, AI Lead, Centre for the Fourth Industrial Revolution, World Economic Forum, observed the need for more international collaboration. She added that it would be important to have multi-stakeholder discussion to aggregate the best available tools, and bring together the private and public sectors to build real-time learning opportunities and feedback loops.

An example of this is the suite of AI governance initiatives that IMDA and PDPC have released to guide organisations to deploy AI responsibly. This includes the Model AI Governance Framework, a guidance document that translates high level principles into practical measures that companies looking to deploy ethical AI solutions can adopt voluntarily. Like Taiger, UCARE.AI and VISA are included in the IMDA/PDPC’s Compendium of Use Cases that demonstrates how companies of different sizes in different sectors have implemented the various practices outlined in the Model Framework.

Dr Tan also emphasised that while government effort is important, companies also need to do their part to protect personal data. As part of their human-centric AI approach and commitment to ensuring data security, UCARE.AI is dedicated to their client’s privacy.

From principles to action

The panellists agreed that any new developments have to be easy to use and centred on consumers in order to drive growth. Dr Nuri Purswani, Senior Data Scientist from VISA highlighted that even harmless models can have unintended consequences. This is why we must care about ethics and put human affected by the AI output upfront. VISA builds AI ethics principles into the model as part of its global risk management. For example, VISA aims to design model explanations at the right level for stakeholders consuming the recommendations.

Dr Norberto de Andrade, Global Policy Lead for Digital and AI ethics at Facebook, shared his efforts to contribute to policy implementation through prototyping programmes. By testing what legal text might look like as well as evaluating guidelines and playbooks, Dr de Andrade looks into how these guiding documents can be scaled up to benefit multiple industries.

With an eye on the future, Dr de Andrade describes defining principles and writing guidelines as “the easy part.” Now, organisations must look at possible tensions between stakeholders and commercial execution and how these necessary principles can be implemented in practice.

Ultimately, such digitalisation efforts are integral to the Republic’s growth as a tech powerhouse. As responsibility takes a front seat in innovation, trustworthy AI will drive Singapore into a digital age that will benefit all members of the community.