🤖 Are deepfakes smart enough to deceive voters?

Preparing for AI

Your weekly CogX newsletter on AI Safety and Ethics

The week's developments on AI and regulation, explained | 26.01.24

If you’re enjoying this briefing, sign up here — and share it with a friend. Reading time: ~3 minutes

This week deepfakes are taking center stage: are they 2024’s greatest threat to democracy?

In the UK a significant majority of MPs are raising the alarm to stop AI-generated misinformation from disrupting the democratic process. This concern is not unfounded, as seen in the recent incident in the US where a robocall, eerily imitating President Biden, urged New Hampshire residents not to vote — a suspected tactic of voter suppression.

Think deepfakes aren’t a threat? Test your identification skills in our AI dilemmas section.

Meanwhile, NYC’s anti-bias in hiring law is being ignored by 95.4% of companies, and parents want AI off social media: can robots promote unrealistic beauty standards?

We also cover the experts’ take on urgent regulation, the potentially devastating effects of AI on software development, and the simple LLM solution to XAI

- Charlie and the Research and Intelligence Team

P.S. You can now register your interest for CogX Festival LA 6-9 May, where we’ll be talking all things AI, Impact, and Transformational Tech!

Ethics and Governance

🏛️MPs call for urgent action against deepfake threats in elections. Following several targeted attacks against high-profile politicians, 70% of MPs are concerned about AI-generated misinformation. Regulations for the 2024 elections are being called for.

🗳️Deepfake President Biden? A recent robocall imitating Biden, urging New Hampshire residents not to vote, is likely an AI-generated deepfake. This effort is a suspected voter suppression tactic ahead of the state's Democratic presidential primary.

⚖️NYC’s AI-hiring discrimination law is being largely ignored, with only 18 out of 391 companies complying. The law requires companies to disclose information on their AI-hiring practices, yet most are noncompliant due to a lack of clarity on scope, and relevance.

🌐America's controls on AI tech sales to China are undermined by enforcement challenges and loopholes. This has led to ineffective restrictions, and prompted China to develop its own technological capabilities, to bypass its reliance on US tech.

AI Dilemmas

❓Do you know which face is AI? AI-powered tools can generate increasingly realistic deepfakes, can you tell the difference? This quiz tests your ability to identify which of these images are real people, and which are AI-generated.

🤳Parents are increasingly concerned about AI-generated influencers on social media platforms, fearing they promote unrealistic beauty standards to children. Over 12,000 parents have petitioned TikTok to better label the content, to prevent further harm.

Insights & Research

💻How is AI affecting the future of software development? A study has shown an increase in "code churn" and "copy/pasted code," suggesting that while AI tools like GitHub Copilot can speed up coding, they may lead to lower quality and increased technical debt.

🚨First regulate deepfakes. Connor Leahy is urging for the immediate regulation of AI and deepfakes, to prevent existential risk. He is calling for legal accountability for creators of harmful AI, and proposes a compute cap to slow AI’s rapid progression, particularly in AGI.

🖥️The law must recognise that computers make mistakes. Research shows transparency in AI-related legal cases, especially through the disclosure of relevant data and code, is crucial to avoid injustices as AI increasingly influences critical decision-making areas.

📚Can LLM’s simply XAI? This paper proposes "x-[plAIn]", a custom LLM designed to make XAI more accessible to non-experts. This model creates clear, audience-tailored summaries of XAI methods, and bridges the gap between complex AI concepts and practical applications.

7 OCTOBER | LONDON 2024

SEPTEMBER 12TH - 14TH
The O2, LONDON