AI Update: Deepfake Tech Generates Unease

Welcome to the latest edition of FindBiometrics’ AI update — a look at what’s changing in the financial, political, and technological landscape of artificial intelligence.

Major institutional investors are starting to pressure technology companies to take responsibility for the ethical and human rights implications of their products, with AI-powered technologies like facial recognition being a top concern. Many are doing so through the “Collective Impact Coalition for Digital Inclusion”, which comprises 32 financial institutions managing $6.9 trillion in assets. A manager with one member of the Coalition, Aviva Investors, tells the Financial Times that meetings to discuss these concerns with tech companies have “accelerated in pace and tone”.

A pair of well-known Silicon Valley startup investors are trying to get deals with founders by offering access to servers equipped with cutting edge Nvidia chips that are in short supply. Former GitHub CEO Nat Friedman and former Y Combinator investment partner Daniel Gross call their chip service, which comprises over 2,500 H100 GPUS, the “Andromeda Cluster“, and they say it can train a 65-billion parameter AI model in about a week and a half. The pair have also raised an investment fund with over $1 billion in assets.

Stability AI has unveiled a new image generation model that offers more photorealistic images than the previous version. The company attributes the superior performance of SDXL 0.9 (an acronym for “Stable Diffusion XL”) to its use of a higher number of parameters in AI training, and says it will be released as open-source software in July. Speaking at this week’s Bloomberg Technology Summit, CEO Emad Mostaque addressed concerns about the use of such image generation technology for deepfakes. “You had to build in things like watermarking,” he said, “Part of what we want is to build standards around that, so you can track it and so you can use it appropriately.”

Miami-based Sumsub has launched an enhanced version of its liveness detection solution as part of its effort to counter the threat of AI-driven deepfake technology. Its liveness detection system creates a “3D FaceMap” of a given subject, which, once established, can be used for authentication. The company says it has been tracking deepfake data, and that the number of deepfakes detected in the first quarter of this year was 10 percent higher than the total number of deepfakes detected in all of 2022.

The Cyber Administration of China (CAC) has published a “white list” of approved developers of deepfake technology – or ‘deep synthesis’ technology, in its bureaucratic parlance – that includes Alibaba, Baidu, and Tencent. The companies registered at the CAC’s encouragement under a new regulation introduced at the start of this year, called the “Administrative Provisions on Deep Synthesis for Internet Information Service”. The Chinese government has sought to gain regulatory oversight in the rapidly evolving generative AI space in an effort to protect the public interest and promote “core socialist values”.

OpenAI is planning to launch an app store that would offer custom AI models built on its core technology, which powers the popular chatbot ChatGPT. CEO Sam Altman revealed the potential plan to developers last month. The apps offered through the marketplace would be based on the customized models designed by OpenAI’s enterprise customers, a couple of which have reportedly already expressed interest in participating in the program.

American VCs have poured almost $17 billion into defense and aerospace startups in the first five months of this year, eclipsing the $16 billion in investment in the space in 2019. As the FT reports, the investment boom has mirrored that of the AI sector, fueled in part by geopolitical tensions and the war in Ukraine, where sophisticated AI tools such as mobile-based facial recognition have been deployed to help Ukraine resist Russia’s invasion.

ESWIN Computer Technology, a China-based developer of chips for Internet of Things devices, has raised 3 billion yuan, or about $421 million, in a Series D investment round. The round was led by state-owned investment platforms Beijing Financial Street Capital Operation Group and Guoxin Venture Capital. The round reportedly saw the participation of a number of state-affiliated investors, and, unusually, another tech company — Shanghai-listed facial recognition software developer Cloudwalk Technology. Facial recognition played a prominent role in the “Security China” trade show earlier this month.

A new, quadrupedal robot that Wired describes as ‘a budget version of Boston Dynamics’ Spot’ features support for facial recognition and emotion detection. Luwu Dynamics’ “XGO-Mini2” is primarily meant to serve as a learning tool for developers who are interested in playing with robotics and computer vision. It is highly programmable; and upon startup offers a few demo capabilities including gender estimation, emotion recognition based on facial expressions, and a facial recognition tool that allows the user to register different people’s face biometrics.

Stay tuned to FindBiometrics for more updates on the AI landscape as momentum continues to pick up!

–

June 22, 2023 – by Alex Perala