Have you ever asked Siri or Alexa to ‘Text mum “I love you”’, only to realise that the message instead reads ‘rainbow blue’? Voice assistants can be immensely helpful, but our messages sometimes get lost in translation.

Such is the fallibility of speech recognition. When an individual is speaking in multiple languages, such programmes may not be able to understand them at all. This is why AI Singapore developed a speech recognition programme that can recognise the colloquial English spoken in the country.

Laurence Liew, Director of AI Singapore (AISG), shares more about how the programme helps emergency services, and delves into how governments can use AI for good.

AI for emergency services

Singapore set up AISG to groom local AI talents and spearhead AI innovation to create tools for good. The organisation helped the Singapore Civil Defence Force (SCDF) dispatch emergency medical resources more quickly with a transcription programme for emergency calls.

This programme can reduce the time needed for dispatchers to pen down information from the call, such as the severity of any injury, reported the Straits Times. This will improve how SCDF dispatches its emergency medical resources, said Assistant Commissioner Daniel Seet, the Force’s Director of Operations in the same report.

While existing transcription programmes exist, many of them only work for those speaking standard English, shares Liew. In Singapore, however, the norm is to speak a mixture of languages including Singlish – a form of colloquial English that integrates multiple languages like English, Mandarin and Malay.

A lot of these programmes will not work when multiple languages or Singlish is taken into account, Liew says. This programme by AISG is able to transcribe emergency calls in Singapore with an accuracy rate of about 90 per cent, wrote the Straits Times.

It is also able to recognise Hokkien terms such as ‘jiak ba bueh’, meaning ‘have you eaten?’, and even the names of local dishes like ‘char kuey teow’, the article stated.

This programme also allows SCDF to better plan out their emergency service resources in the longer term, shares Liew. The Force can perform text and data analysis on the transcriptions to identify times and areas when emergency services are most needed.

For instance, the SCDF stationed ambulances at local community centres during morning and evening peak hours after analysing data from their past emergency responses, according to a Facebook post on their page.

Such data analysis would be difficult without accurate transcriptions, adds Liew.

AI in government 

Beyond emergency services, AI has immense potential to help governments provide better citizen services. Countries around the world are using it to tackle climate change, improve security, or even diagnose diseases in their plants.

AI will automate, streamline, improve productivity, and lower the cost of delivery of citizen services. But it often comes with fears that it will take away millions of jobs.

This isn’t true. Instead, AI often automates tasks which workers don’t want to do because they are “boring, mundane, or dangerous”, shares Liew. This frees up the workforce to focus on other, more important tasks, he explains.

For example, when emails first became popular, people often had to spend the first 15 minutes of their work day clearing their inboxes of spam. Today, emails are automated such that spam mail is automatically detected, labelled, and removed from the inbox, he shares.

Rethinking hiring and talent retention in AI 

With AI being such a competitive industry, government agencies need to think of novel ways to hire and retain AI talent if they wish to build up their AI capabilities, Liew suggests.

This is because experienced AI talents are often headhunted by big tech firms. It is difficult for governments to compete with these firms as the pay scale in governments is often fixed, he explains.

AISG solves this problem through their apprenticeship programme, which brings in AI graduates and trains them for nine months. Subsequently, many of their apprentices stay on to fulfil a two-year contract before leaving. At this stage, a new wave of apprentices will join the team.

This is a new model that HR may not be familiar with, but is inevitable in the AI and machine learning talent war, Liew shares. With this model, AISG has managed to grow from a team of just four engineers, to nearly 40 people today, he adds.

With AI, governments can divert their attention from menial tasks like emails, and instead focus on providing better services both in emergencies and in citizens’ daily lives. All it takes is a spark of innovation, both in tech and hiring.

Want to hear more about how governments can use AI? Laurence Liew will be speaking at AI x GOV as a panellist for ‘Future of AI in government’, happening on 6 April, 4:45 – 5:45 pm GMT+8. Register here!