Interview with Laurence Liew, Director @ AI Singapore
Welcome to the second round of our new interview series. Given AI and machine learning is pervading many industries and professions, we thought we’d reach out to a few experts in various industries to find out more about their experience to date. We were fortunate enough to get Laurence Liew, Director of AI Industry Innovation and Makerspace in AI Singapore, to share his thoughts and experiences with AI and Machine Learning.
Verge Labs: Tell us a little about yourself.
Laurence Liew: I am currently the Director of AI Industry Innovation and Makerspace in AI Singapore. The mission here is to enable both companies and people to learn, do and use AI. I have a Mechanical Engineering degree from the National University of Singapore (NUS) and a Masters in Technology (Knowledge Engineering) also from NUS.
Prior to AI Singapore, the last 2 decades I have been in the industry building start-ups and companies in the open source, high performance computing, cloud and analytics space. My last company, was Revolution Analytics (enterprise R) where I was General Manager for Asia. We were acquired by Microsoft in 2015.
VL: What are some of the more exciting AI applications you’ve seen both from AI Singapore and globally?
LL: AI have been around for a long time, and researchers and companies have been trying to do machine vision and language for a while. It was only the recent breakthrough with Deep Learning coupled with large datasets and cheap computing power, we now have very exciting real-world applications of machine vision and language, including speech to text.
What excites me is that with AI, we are getting closer to the “Universal translator” which will break down barriers in communications across the world. In larger countries, with their larger market and data corpus, the big AI and cloud vendors like Microsoft, AWS and Google have very good mainstream languages supported and achieving reasonable accuracy.
For smaller countries like Singapore, where dataset is much smaller, and in addition with local slang (Singlish), this is where the big boys speech to text system becomes less usable. One of AI Singapore popular projects is our AI Speech Lab, where we worked with two leading researchers and professors who have developed a Singlish speech to text engine. This Singlish engine is being enhanced with 3 to 4 organizations at the moment for use cases in call-centers, meetings and interviews.
VL: AI Singapore looks to have some healthy working relationships between industry and research. How do you foster that collaboration?
LL: The Singapore government has many programmes to support such collaborations. It can be as simple as undergraduate internships with industry partners, to full-fledged joint research labs between the university and industry where the Singapore government co-fund part of these collaborations.
In AI Singapore, our flagship industry collaboration programme is the 100 Experiments (100E), where we assist industry partners to co-develop an AI/ML solution or product. In the 100E, the problem statement would typically be such that there is no existing solution in the market or the industry partner wants to create a new competitive product, and hence would like our help to develop it. We then put our researchers and engineers in a shared cost model (50-50) to co-create the AI/ML solution. Besides cash, the industry partner will also contribute in-kind in terms of manpower, so that they can learn, maintain and enhance the solution after we hand over the project to them..
VL: One of the really interesting focus AI Singapore is on education and apprenticeships. Why is it important for you to ‘grow your own timber’, and what are some of the outcomes of that program so far?
LL: There is generally a shortage of AI/ML talent globally. Singapore is no different. In fact when we started AI Singapore and launched the 100E, one of the problems I had was recruiting AI/ML engineers to work on the 100E projects.
So to solve our dilemma, we created the AI Apprentice Programme (AIAP), which brings in Singaporeans who have taught themselves AI/ML on their own through books, MOOC and classes. Our recruitment process includes a technical assessment which consists of 5 - 6 questions in the areas of software engineering, AI/ML modelling, mathematics and statistics, and also a final face to face interview.
We typically get 120-160 applicants per in-take, and only accept around 18 - 26 per batch.
These apprentices are then provided the opportunity to deep-skill themselves for a further 2-months with guidance from our AI Mentors, after which they work on the 100E project for 7-months and deliver a Minimal Viable AI Model (MVM) to the industry partner. So far, we have trained about 62 Singaporeans and the next batch of 18 starts in the middle of September. Our goal is to train 500 AI engineers through the AIAP over the next few years.
In terms of 100E, we have approved 40 projects with more than 30 already started, of which 6 or 7 of them will finish in October 2019 after 9-months..
VL: Your group open sourced TagUI, a tool for Robotic Process Automation and continually maintains it as a project. What are some of the success stories of RPA?
LL: Yes, TagUI is a relatively popular project with 247 forks, and more than 3159 stars from all over the world. We have very active users and collaborators in Europe!
In Singapore TagUi is used by several government agencies, accounting firms and banks. AI Singapore also uses it internally for our projects and administrative tasks.
One use case is student information submission to a government website for student claims. The website was originally developed for small class sizes of 10 - 30 students, hence the use case was student information was typed in by administrative staff one at a time. However in AI Singapore, some of class size is 300, hence we automated the process with TagUI and reduce the time taken from 3 days to 2 hours, where the 2 hours were to correct errors encountered during the RPA process..
VL: Related to RPA, the word ‘automation’ is a fairly loaded term and can be worrying for a lot of people. What’s your take on automation vs augmentation? Are people’s fears justified?
LL: Technology is here and will continue to advance. Some tasks can be fully automated, while other tasks will be augmented by RPA (or AI).
For example, today, we no longer have to manually clear our INBOX of spam. This is an example of full automation. It has relieved us of this boring and mundane tasks and given us back 15 - 30 minutes of our time daily.
Augmentation will play a bigger role with RPA and AI, and it is therefore important that workers of the future, learn how to use these tools. RPA/AI is just a tool. Learn to use it!
You will not lose your job to RPA/AI, but you will lose your job to another person who uses RPA/AI to be more productive and faster than you!.
VL: What skills outside of AI/ML and coding that you believe will be important as AI continues to become a part of people’s everyday lives?
LL: Not everyone will code. AI/ML is also becoming easier. What is valued now, and increasing in the future will be problem solving skills including the ability to understand the business domain, know where the data comes from, how to use the data, and use an automated AI/ML tool to do the heavy lifting. The person will then need to know how to interpret correctly the output of the AI/ML model. “Garbage in, garbage out” holds true here!
The above is actually for the business analyst. In fact today, a lot of applications are already powered by AI, and we are users of such AI systems whether we know it or not, for example, your Spotify or Netflix recommendations, what you see on your Amazon page, when you take a loan or swipe your credit card for a purchase - AI is running at the backend..
VL: Data privacy is obviously a key concern for many people working with data. How have you seen the landscape change in recent years, and how has it affected you?
LL: To do machine learning and deep learning, lots of data is required. Recent news of Apple, Microsoft, Google etc having contractors listening to recorded audio from their applications sends a strong reminder of the privacy issues around AI and ML.
In AI Singapore, our 100E requires the industry partner to provide the dataset, else we cannot approve the project and execute it. So we have less of an issue getting data since the 100E adds real value to the industry partners’ use case and they often will bend backwards to get the data for 100E to proceed.
What I see happening in the next few years is other forms of training which will require less data, or training with encrypted data, or federated learning approaches, all of which will help to address the data privacy issues..
VL: How can companies ensure that their algorithms are acting ethically and unbiased?
LL: In reviewing and approving over 40 100E projects, we have only encountered one case where our team was unsure of the ethical and biases of the use case. The project did not proceed.
So at the moment, we use human judgement - our AI Engineers and approving committee - to spot potential ethical concerns.
We will continue to use human judgement and exploring newer techniques to determine if dataset is unbiased.
The Singapore government recently announced an AI Ethics and Governance Framework model - and as part of best practice, my AI engineering team will incorporate the framework as part of our 100E review and development process.
VL: Any final words?
LL: I like to quote AI Singapore’s Chief Scientist - Prof Chen Tsuhan - he said this during a BCC interview a while back: “AI will make us more human.”
AI ( or any advanced automation) - will give us back our time, remove the boring and mundane, and allow us to do more of what we love, and allow us to be more human!
VL: Lots of really exciting things happening at AI Singapore, thanks for taking the time to share your thoughts with us!
To find out more about Laurence Liew and AI Singapore, check out aisingapore.org.
And if you enjoyed this interview you’re sure to like our previous interview with Creel Price from Investible.