ST Engineering’s deepfake detector among early wave of tools to battle AI-generated misinformation
Sign up now: Get ST's newsletters delivered to your inbox
ST Engineering’s head of data science and AI Tan Boon Leong said its tool detects inconsistencies in an uploaded video that are likely to be generated by computers.
PHOTO: LIANHE ZAOBAO
Follow topic:
SINGAPORE - An early wave of deepfake detection software built in Singapore has been released to the market as the nation builds its arsenal to deal with risks of artificial intelligence-generated misinformation ahead of the general election.
The latest of the detection programmes comes from Singapore’s ST Engineering, which launched its Einstein.AI deepfake detection tool for enterprises at its annual InnoTech Conference on Sept 3 at Marina Bay Sands.
When deployed by banks, investment platforms and media companies or other clients, the tool detects inconsistencies in an uploaded video, like odd eyebrow or lip movements, and audio frequencies that are likely to be computer-generated, said ST Engineering’s head of data science and AI Tan Boon Leong.
“These are increasingly becoming tougher for people to spot with their own eyes,” said Mr Tan, who worked on Einstein.AI.
Once scanned, the AI generates a report that includes a deepfake likelihood score, a transcript and summary of the video’s audio feed and a list of related news content found online to allow social media users and investors to assess whether the content is reliable.
Senior Minister of State for the Prime Minister’s Office Desmond Tan said in an opening address at the conference that tools like Einstein.AI are a significant step to combating deepfakes, which surged by some 500 per cent in 2023 compared with 2022.
Deepfakes also pose a threat for enterprises, said Mr Tan, citing an incident of executives in Hong Kong who transferred some HK$200 million (S$34 million) to fraudsters who used AI to mimic the likeness of their chief financial officer to approve of a transaction.
“Scammers are now using advanced AI tools to create deepfake voice recordings and videos that can fool even the closest of relatives and friends,” he said.
“With just a few photos from social media and a short voice clip – sometimes as brief as seconds – scammers can create frighteningly realistic deepfakes.”
The authorities worldwide are scrambling to rein in deepfakes amid concerns that fraudsters will use it to disrupt elections in many countries in 2024, including in Singapore.
India faced a scourge of political deepfake content on social media during its elections earlier in 2024, continue to proliferate on social media
Fake videos in 2023 of Senior Minister Lee Hsien Loong Prime Minister Lawrence Wong promoting investment policies
Concerns over deepfakes have prompted heavy investment in detection capabilities here.
The Ministry of Digital Development and Information in May announced a $50 million investment over five years into the new Centre for Advanced Technologies in Online Safety, which will focus on building tools to spot misinformation and deepfakes.
Separately, developers at Singapore’s Agency for Science, Technology and Research announced in late 2023 it is developing a deepfake detection model called The Sleuth, which can spot synthetic content on media platforms, messaging apps and web browsers.
The Home Team Science and Technology Agency in April showcased a similar detection tool called AlchemiX, which can compare recordings of a suspected deepfake video with a recording of a speaker’s actual voice.
