7 OCTOBER | LONDON 2024
SEPTEMBER 12TH - 14TH
The O2, LONDON
Preparing for AI
Your weekly CogX newsletter on AI Safety and Ethics
White House endorses AI hackers
Thousands of hackers gathered at Def Con 31 in Las Vegas to try and break LLMs at an event endorsed by the White House. All the large AI labs, including OpenAI and Anthropic, opened up their models for testing to identify vulnerabilities in a safe environment. But should more testing be done before models are actually released?
Meanwhile, momentum behind new regulations and monitoring agencies is increasing in DC, while the UN hosted its first ever Security Council meeting on AI risks.
New research also found that participants are unable to distinguish between human-written and AI-generated tweets, while a Harvard Business School professor outlined the real risk from AI on jobs is to those who don’t adopt it.
Explore these topics and more - from AI images of your future kids to AI surveillance at the Paris Olympics - in the CogX Must Reads.
Top Stories
White House encourages hackers
Over two and a half days, 3,000 people will each be given 50 minutes to find flaws in eight LLMs. Completing challenges wins points, with prizes for the winners. The event has the support of the White House, which believes it will provide critical information to researchers and the public about the impact of these models.
Want to know what your kids will look like?
A new app, Remini, is taking social media by storm as it uses AI to generate photos of what your future children will look like after you upload a photo of you and your partner. But is it safe? Cyber security experts caution the risks of AI firms using your likeness.
Politics and Regulation
US policymakers’ consider AI rules
AI regulations are being vigorously debated in DC as the US develops its approach to AI safety. Actions being considered include new regulations on AI models, new institutions to enforce standards, additional funding for AI safety research and a people and skills strategy to attract AI talent and ensure the workforce can safely use it.
UN Security Council debates AI
The UN has been using AI to identify patterns of violence and monitor ceasefires. However, Secretary-General Guterres highlighted that the technology also poses risks for international peace, security and human rights as he hosted the Security Council’s first formal meeting to address these issues.
Latest Research
Humans can’t distinguish between AI tweets
University of Zurich researchers asked 697 participants to identify human-written tweets from and AI-generated ones and found they couldn’t tell the difference. The results highlight the potential risks from misinformation, but the authors also argue that it can be used positively with information campaigns to improve global health, for example.
How can we improve resilience to AI risk
A briefing paper from the Centre for Long Term Resilience et. al provides a clear framework for how the UK Government should urgently understand and respond to AI risks. Delaying tackling AI threats risks one of two undesirable outcomes: either AI starts to directly harm society and we are unprepared for it, or the widespread fear of risk impinges on adoption and reduces benefits.
AI Thinkers
The real replacement risk
Harvard Business School Professor Karim Lakhani explains that managers and leaders need to embrace AI and start using ChatGPT. He has advised leaders that AI is not going to replace humans, but people who don’t use it will be replaced by those that do.
AI regulation needs to be international and inclusive
Dorothy Chou from Google Deepmind argues that if AI labs want to avoid superficial ethics washing, marginalised communities must be brought into the conversation. All those involved in building AI should adopt and adhere to principles that prioritise safety while pushing frontiers of innovation.
Use Cases
Surveillance at the Olympics
Real time cameras will use AI to detect suspicious activity and find unattended bags, crowd rushes and fights at the Paris 2024 Olympics. While facial recognition technology won’t be used, campaigners have pushed back, calling the technology a threat to civil liberties. They worry that the Olympics will be used as an excuse for permanent changes to surveillance.
How to create a fake memory
Researchers found they could incept false memories by showing participants deepfakes of made up movies. Participants were tricked into believing the films were real, and some even ranked the fictional movies to be better than the real originals. While this memory distortion can also be created by text descriptions, the ease of creating deepfakes could increase its prominence.
CogX Must Reads
In case you missed it
CogX Festival 2023 speaker, Jaan Taalin, explains the threats behind AI and how to develop it safely.
We'd love to hear your thoughts on this week’s Issue and what you’d like to see more of.