A*STAR Explains: The future of AI in Singapore
By Yun Xuan Poon
Singapore's public sector R&D agency A*STAR reveals what’s brewing in Singapore’s AI labs.
Today’s AI still can’t quite achieve this, but it has certainly become an integral part of our everyday lives. Singapore launched its national AI strategy in 2019, beginning with test cases for five projects across education, health and customs. As Singapore's public sector R&D agency, A*STAR continues to develop innovative AI tools and solutions in areas such as telehealth, smart living and manufacturing, among others.
The agency shares how AI is already being used in Singapore, the challenges that researchers have to overcome, and opportunities for the future.
Genomics and Covid-19
AI is useful when processing large volumes of information quickly – this is especially so when analysing genomic data.
Genomics form the basis of many diagnostic tests, including determining the risk of Down syndrome in pregnancy, and can predict one’s likelihood of suffering from a disease, an A*STAR spokesperson told GovInsider. But genomic data is massive and complex, and requires advanced analysis that would be tricky without AI.
Researchers in Singapore use algorithms to understand how significant a given genetic variation is, group samples that have similar molecular profiles, and predict how genetic variation can lead to disease. Machine learning can also link genomic data to other types of data, such as data from wearable sensors and blood tests, to guide doctors’ treatment decisions.
Machine learning isn’t just useful for unpacking genomic data. A*STAR has developed a way to more accurately predict patient admission from their electronic medical records. Another algorithm can identify the most effective drug combinations for individual cancer patients.
A*STAR has contributed to the fight against Covid-19 as well. It worked with Tan Tock Seng Hospital, one of Singapore’s largest hospitals, to develop RadiLogic. This tool uses AI to speed up chest X-Ray screenings for suspected Covid-19 patients. This allows them to be diagnosed and treated more quickly, A*STAR explains.
The agency also collaborated with SingHealth to develop the “Doctor COVID” Telegram chatbot, which allows healthcare workers to communicate with patients remotely.
“The current explosion in the volumes of omics data and electronic health records will open up new frontiers for biomedicine,” A*STAR notes. Researchers and developers will need a good way of exchanging and analysing data to create cutting-edge tools.
That’s where the A*STAR Data Analytics Exchange Platform (A*DAX) comes in. It connects multiple data sources from various separate systems and organises them so that data can “be shared with proper control”, A*STAR says. This also makes analytics algorithms easier to apply, so healthcare can make better use of all its data.
Outside of healthcare, A*STAR has lent its expertise to public sector agencies. Its AI tools are being used for transcribing in the State Courts and translation on Gov.sg’s Telegram channel for national Covid-19 updates.
How to build fair algorithms?
Unfortunately, AI has acquired a slightly unsavoury reputation for carrying inherent biases. A quick Google search for “healthy skin” returns mostly images of fair-skinned ladies, and algorithms used to detect skin cancer do a lot worse for darker skin types, MIT Tech Review has noted.
Developing fair and unbiased algorithms is an ongoing area of research interest globally, shares A*STAR. “One school of thought is that it is not algorithms that are biased, but rather the data that they are trained on. If the training data is historically biased, then algorithms trained on that data would reflect such biases and perpetuate them,” the A*STAR spokesperson explains.
One way to combat this is to keep humans in the decision-making process. This allows humans to bring in socio-cultural, ethical and legal context, detect biases that the AI may not be aware of, and intervene as necessary, says A*STAR.
But AI is more than its algorithm. “Indeed, algorithms are only a fraction of all useful IT systems driven by AI,” notes A*STAR. Every AI project needs expert knowledge about the specific industry, a well-defined problem statement, and a well-designed plan for collecting data.
Useful and useless AI
Another common frustration with AI is that it doesn’t always understand our commands. Machines learn from patterns without really understanding the human need behind a task. This can lead to “catastrophic failures” when AI is used for something other than what it was designed for, warns A*STAR.
“We need to carefully consider the risks of such failures in selecting use cases for application of AI technologies,” the agency adds. It works closely with government agencies to identify suitable use cases, says Dr Lim Keng Hui, Executive Director, Institute of High Performance Computing and Executive Director, AI, Analytics and Informatics (AI3) Horizontal Technology Programme.
One of A*STAR’s goals in AI research is to develop AI that understands human intentions better. Robots have the potential to become much better assistants, whether in manufacturing or rehabilitation. It can even conduct personalised skills training for workers.
A*STAR has begun this work with robot assistants in hospitals. Researchers are training these robots to understand nurses’ intentions and anticipate their movements, so they won’t get in their way during emergencies.
The agency is also working with higher learning institutes to improve manufacturing and engineering robots. The goal is for these robots to understand humans through natural interaction such as speech and gestures, and learn from instruction and demonstration. Besides saving money and time on programming each robot for each task, this project could “fundamentally change how robots work with people”, a spokesperson shares.
We may not yet have developed robots that are indistinguishable from humans, but perhaps helpfulness is more critical than human-likeness. A*STAR has developed various AI tools that have improved healthcare and sped up research. It will continue to explore exciting new opportunities while ensuring AI remains unbiased and safe.