Copy
View this email in your browser

Singapore AI Newsletter

Fortnightly updates about AI policy and the wider ecosystem in Singapore. The field moves fast, subscribe to keep up. Forward this to your friends to make their day.

This week's issue includes an update from our neighbour up north, written by Julian Theseira. If you think other updates from SEA are helpful, let us know. 

Speech recognition, hard mode

The Singapore Civil Defence Force (SCDF) and AI Singapore’s Speech Lab are developing a speech-to-text system to transcribe emergency calls in real time. Automatic Speech Recognition 995 (ASR995) is trained to recognise English, Mandarin, Malay and Tamil as well as Singlish, an english-based informal language spoken in Singapore. SCDF receives ~550 calls every day.

According to its developers, ASR995 achieves a 90% accuracy rate, but more technical details have not been released yet. Based on previous papers on Mandain-English code-switching from Speech Lab researchers, my best guess is that the system employs an End-to-End attention-based architecture (as opposed to DNN-HNN) and multitask approach (where the language identification task is explicitly learned in addition to the speech recognition task).

Beyond this narrow use case, speech recognition in Singapore is very difficult because people use English, Malay, Tamil, Mandrin, Hokkein, Teochew, and Cantonese, oftentimes in the same utterance. In other words: less data, uncommon sentence structures, and many possible sounds. “Wah chope table damn sian, dapao la” (I’m not in the mood to wait for a table, so let’s get this to go). This project fits into the larger challenge of adapting ML for different cultures and languages, which matters if we care about who benefits from (or get surveilled by) these systems.

‘Accountability’ obligation for private sector data protection practices

The Personal Data Protection Commission (PDPC) announced that the 9th obligation of the Personal Data Protection Act (PDPA) has been changed from ‘openness’ to ‘accountability’. The PDPA is a legally binding document that governs the collection, use, and disclosure of personal data by private companies. Other obligations include consent, purpose limitations, notification, and data retention.

Jo: Note that ‘accountability’ was previously part of the ‘openness’ obligation, but is now being brought to the forefront. ‘Openness’ seemed to be about ensuring the public availability of organisations’ data protection policies—important but performative in its mandate. This change reflects an emphasis on holding organisations’ data practices to a higher standard. It’s good to see PDPC make the principle explicit and detail how it can be operationalised with the accompanying advisory guidelines.

Public service introduces new data protection measures

Context: the past year has seen a few high-profile cyber attacks in Singapore. In June 2018, the medical information of 1.5 million patients, including the prime minister, was stolen. In January 2019, the national HIV database was leaked. In response, the prime minister convened the Public Sector Data Security Review Committee in April 2019. 

Update: so far, the committee has investigated five government agencies. It found that public sector organisations have insufficient policies governing data sharing with third parties, and that human errors are a common failure node. The committee has introduced 13 technical measures which aim to detect unusual data flows, limit the number of people with access rights, and increase encryption standards for sensitive information. The full report from the committee is due in November 2019.

Bonus: Malaysia’s AI ecosystem

The Beyond Paradigm Summit was recently held in Kuala Lumpur (the capital of Malaysia) on July 17–18 and Kuching (the capital of Malaysia’s largest state Sarawak) on July 20–21. The summit is a milestone that signals Malaysia’s ambitions for AI. The 2019 AI Readiness Index, which tracks how prepared governments are to adopt AI in the delivery of public services, ranked Malaysia at 22nd.

In light of the summit, our friend Julian Theseira wrote an accessible introduction to recent policy updates and potential challenges for Malaysia. His take on the challenge of coordinating between various government departments and the federated structure of Malaysia was informative for me. Countries are not monolithic, and outcomes are affected by infighting and internal comprise. Julian is currently based in KL and is exploring a career in AI governance and policy research.

Others

DeepNude comes to Singapore (Straits Times)
A reminder that code has transboundary consequences, and that women and other typically disempowered groups tend to suffer most when technology is misused.

Strides made, but some way to go for Government to meet data needs (Channel News Asia)
GovTech runs an open data portal, and previously said that “open data sharing is one of the priority areas for Singapore’s Smart Nation vision”. This article suggests that these efforts have been insufficient. For further reading, here’s a paper by NTU’s Hallam Stevens which looks at the data available on the portal and the resulting applications.

Certification scheme for cross-border data sharing (IMDA)
This move is in line with ongoing efforts to harmonise data protection standards across APEC member countries in order to encourage cross border information flows and trust in those processes. 

PDP Digest (PDPC)
An annual document that collates key decisions made by the commission, as well as articles on digitalisation and data protection. It includes “Artificial Intelligence and the PDPA” (p.74) that touches on how the 9 obligations of the PDPA might be applied to data protection challenges related to AI.

Thanks for reading! Sign up here.

This newsletter was written by Jia Yuan (jyloke@smu.edu.sg) and Josephine (shseah@smu.edu.sg). We’re research associates at SMU’s Centre for AI and Data Governance.

Opinions expressed reflect the beliefs of individual authors and not CAIDG as an institution.

Photo by Adrian Jakob on Unsplash.

Twitter
Copyright © 2019 Singapore AI Newsletter, All rights reserved.






This email was sent to <<Email Address>>
why did I get this?    unsubscribe from this list    update subscription preferences
Singapore AI Newsletter · 55 Armenian Street · Singapore 179943 · Singapore

Email Marketing Powered by Mailchimp