The Face Value Of Artificial Intelligence

With Singapore doubling down on its Smart Nation goals, artificial  intelligence (AI) has taken centre-stage of socio-economical development, especially with these uncertain times ahead of us.

Portrait of Tammy Strobel


Understanding what AI can or cannot do for us is a critical step towards uplifting our economy and its people. Is AI our future, or is it just a buzzword thrown around to get people and businesses excited? Is there more value to AI technologies we are familiar with, such as facial recognition? Will AI ever be abused in Singapore?

We spoke to Dr Fanglin Wang, the Head of Computer Vision at ADVANCE.AI, on his thoughts around Singapore’s AI journey, AI itself in its various manifestations, and the possible misuse of such technologies.

Is facial recognition underutilised in Singapore?

Dr. Fanglin Wang:  Facial recognition technology has already been part of daily life for almost a decade. It’s widely used in the algorithms in our smartphone cameras. In Singapore, cameras have been installed at public housing, hawker centres, and major train and bus stops since 2012 for security purposes.

Social networks can be considered to be the world’s largest facial recognition databases today when you consider the amount of photos uploaded to these platforms. Mobile banking applications use facial identification to verify customer log-ins, and you also see this technology at airports and border checkpoints.

That said, our broad use of facial recognition in Singapore still cannot compare to China, where facial recognition technology is a way of life. In China, facial recognition technology scans people in public. It monitors errant behaviour, which is then used to build a national “social credit system”. Disobeying traffic rules, for example, can impact loan scores or ability to rent apartments. Ultimately, every country needs to find the right balance between convenience and privacy, and that scale will move as consumers get more familiar with the technology.

What are the barriers and enablers for Singaporeans to embrace AI-based technologies?

Dr. Fanglin Wang:  Privacy concerns are always top of mind. Am I constantly monitored? How is my data being stored, collected, and used? How secure is the data? These are all valid concerns, and we need to move cautiously.

As mentioned earlier, facial recognition in China is a way of life, and people are accustomed to the technology in every facet of their lives. In contrast, when Google launched the Glass in the US and it was being used to record people in restaurant settings, people got very upset; the public pushed back, and the product failed to take off.

Having been closely involved with three Smart Nation bidding projects before, I know the Singapore government is very strict about rigorous and extensive testing before deployment. This extends to strict data privacy and security regulations.

How do you envision public and private sector organisations working together to prevent abuse of AI?

Dr. Fanglin Wang:  Right and ethical use of AI is a constant learning process and needs to be debated openly. Singapore proposed updates to its AI Governance Framework at Davos 2020 in January, which says human involvement must be paired with AI technology for accountable decisionmaking. It also commits to AI decision-making always being explainable, transparent, and fair.

Every country needs its regulation and framework, which is enforced by the government to prevent bad actors and abuse. Regulatory-approved sandboxes can be used to encourage public and private innovation. Development and collaboration with schools and universities to develop future AI research scientists, helping Singapore become a centre of AI innovation and excellence, is the right path forward. This is critical for the Singapore government’s Smart Nation ambitions and will also benefit private enterprises by grooming future local talent.

Broader consumer education to help them understand the possibilities and limitations of AI is key. This speaks to the process of continuous learning, unlearning, and re-learning to keep up with the technology and make people comfortable with it. We must also ensure no one feels left behind as technology advances.

Are there any new developments in the AI industry to help overcome the limitations of facial recognition technology?

Dr. Fanglin Wang:  Facial recognition tech is only as accurate as the data you’re feeding it. If you want your facial recognition tech to work in Singapore or Southeast Asia, you’ve got to train the underlying algorithm with relevant and high-quality local data. That means training it with Singaporean or Southeast Asian faces, whose facial structures and skin tone vary compared to Caucasians.

Extensive and rigorous testing needs to be conducted to understand the limitations, as well as privacy and security requirements, of consumer data. Poor image quality, caused by bad lighting conditions or quick motion, can result in blurred images which are not conducive to facial recognition tech. Nowadays, this can be resolved with better technology around sensors and depth control on cameras.

Wouldn’t combining multiple biometric-enabled AI for identity verification be more secure?

Dr. Fanglin Wang:  That’s a very interesting question. Intuitively, you would think that combining multiple AI technologies should be better than using only one. However, personally, I do not recommend doing so. Firstly, the cost is higher to use multiple AI technologies together, as you need multiple sets of hardware and software systems. Both the R&D and the maintenance effort required will be much greater. Secondly, it can create a poor user experience if the customer has to go through multiple biometric checks.

The third consideration is that combining multiple technologies properly is hard in practice. They may not end up not complementing but undermining each other. If multiple techniques do not agree with each other in some cases, which one should you trust more? You need to figure out which use cases you can trust with which technology. Then you need to build a more complicated system to combine those calculations and judgements automatically.

Besides facial recognition, what do you think are some AIpowered technologies that will likely see greater use?

Dr. Fanglin Wang:  Facial recognition is a subset of artificial intelligence, which essentially is intelligence demonstrated by machines able to perceive environments and take action. Such “intelligent” behaviour is called AI. This can be classified into two types. The first is “strong AI”, a machine intelligent enough to be able to complete various types of tasks. The second is “weak AI”, which is machine intelligence trained to complete a very specific set of tasks.

Broadly, AI excels in three areas: automating processes to ensure faster and more accurate task completion 24/7; gaining critical customer and competitive insights by crunching mountains of data; and dramatically improving customer and employee engagement, as well as their experience, by working in the background (e.g. facial authentication to make a payment).

In my area of work and expertise, which is using AI in sectors such as banking, financial services, payments, retail, and e-commerce, we focus on a few key areas. Our enterprise AI solutions include know-yourcustomer (eKYC), intelligent process automation, and chatbot capabilities. Our risk management solutions include alternative credit scoring, and fraud detection and prevention. Our digital lending solutions include digital on-boarding, smart decision engines, and smart collection systems.

What all this means for enterprises is that precious resources - such as human capital and money - can be redeployed across the organisation for greater efficiency.

What kind of AI project will likely re-invent the way Singaporeans’ live today?

Dr. Fanglin Wang:  We are really at the start of the AI journey, and the use case for consumer applications are limitless. Separate from my work at ADVANCE.AI, I often wonder how AI can improve the way we communicate. For example, how can we apply speech recognition and machine translation in natural language processing (NLP) and multi-language translation?

I know companies are developing more accurate text-to-speech and speechto-speech translations across languages and dialects. Singapore being a tourist destination, this could prove to be a very useful application of AI technology.

My Reading Room