News/AI Safety
AI Companies Promised the White House Self Regulation, but Transparency is Still Lacking
The White House's voluntary AI commitments have brought better red-teaming practices and watermarks, but no meaningful transparency or accountability. One year ago, seven leading AI companies committed to a set of eight voluntary guidelines on developing AI safely and responsibly. Their progress so far shows some positive changes, but critics argue much more work is needed. Key takeaways: The commitments have led to increased testing for risks, information sharing on safety best practices, and research into mitigating societal harms from AI: Companies are conducting more red-teaming exercises to probe AI models for flaws and working with external experts to assess...
read Jul 21, 2024Top Tech Giants Unite to Establish Unified AI Security Standards
A coalition of top tech companies has formed to develop unified cybersecurity and safety standards for artificial intelligence (AI) tools, aiming to ensure consistent and rigorous security practices across the industry. Key objectives: The Coalition For Secure AI, announced by Google during the Aspen Security Forum, will focus on establishing industry-wide standards and best practices for AI security: The coalition's initial priorities include developing standards for software supply chain security in AI systems, compiling resources to assess AI risks, and creating a framework to guide the most effective use cases for AI in cybersecurity. By working together, the participating companies...
read Jul 20, 2024Using AI as a Digital Guardian to Protect Children Online
AI technology is emerging as a powerful tool to protect children in the digital age by monitoring and analyzing online content, detecting harmful behavior, and providing educational resources. Here's a summary of the key points: AI as a digital guardian: AI acts as a vigilant protector, monitoring online content at a scale and speed impossible for human moderators alone, ensuring digital spaces remain safe for children. Companies like Google and Facebook employ sophisticated AI algorithms to scan and remove harmful content. These AI systems are trained on vast datasets to recognize inappropriate material and keep digital spaces free from exploitation...
read Jul 19, 2024OpenAI Boosts ChatGPT Enterprise with Compliance, Security, and User Management Upgrades
OpenAI has released new enterprise control features for ChatGPT Enterprise, aimed at enhancing compliance, data security, and user management capabilities for its enterprise customers. Compliance and data security enhancements: OpenAI's new Enterprise Compliance API provides customers with a detailed record of interactions, including conversations, uploaded files, metadata, ChatGPT memory, and workspace users, enabling enterprises to audit the data they share on the platform: OpenAI has partnered with several third-party compliance providers, such as Forcepoint, Global Relay, Microsoft Purview, Netskope, Palo Alto Networks, Relativity, Smarsh, and zScaler, to support various compliance-related activities, including archiving, audit trails, data redaction and retention, and...
read Jul 19, 2024Astronomers’ Galaxy-Studying Techniques Are Helping To Identify AI-Generated Deepfakes
The discovery that AI-generated deepfakes can be identified by analyzing the reflections in people's eyes, similar to how astronomers study galaxies, has significant implications for combating the spread of misinformation. Key findings: Researchers at the University of Hull have developed a method to detect AI-generated deepfakes by examining the consistency of light reflections in a person's eyeballs: In real images, the reflections in both eyeballs are generally consistent, while in deepfakes, the reflections often lack consistency between the left and right eyes. By employing techniques used in astronomy to quantify the reflections and check for consistency, the team found that...
read Jul 19, 2024Tech Giants Unite to Tackle AI Security Risks and Promote Responsible Development
The major tech companies are teaming up to promote AI security through shared best practices and tools, highlighting the growing importance of addressing risks as AI becomes more prevalent. Key players join forces: Google, OpenAI, Microsoft, Amazon, Nvidia, Intel, and others are forming the Coalition for Secure AI (CoSAI) to collaborate on AI security: The initiative aims to provide open-source methodologies, frameworks, and tools to address the "fragmented landscape of AI security." Other notable members include IBM, PayPal, Cisco, and Anthropic, demonstrating the broad industry support for this effort. Addressing AI challenges: CoSAI will focus on three initial goals to...
read Jul 18, 2024Emotionally Intelligent Chatbots Spark Concerns Over AI Deception and Manipulation
Chatbots are increasingly mimicking human-like behaviors and emotional expressions, raising ethical concerns about the blurring lines between AI and humans. Key Takeaways: The rise of emotionally expressive AI chatbots is prompting warnings from researchers and efforts by regulators to prevent misrepresentation: A chatbot called Bland AI easily lied about being human when asked, highlighting the potential for AI to deceive users. Researchers caution that the manipulative power of emotionally intelligent chatbots could be used to unduly influence people. AI watchdogs and government regulators are attempting to implement safeguards against chatbots misrepresenting themselves as human. The Bland AI case study: Lauren...
read Jul 18, 2024Apple Denies Using Unethical Data, Commits to Responsible AI Development
Apple refutes claim it used unethical data to train Apple Intelligence, affirming its commitment to using only ethically sourced data for its AI projects. Apple's response to allegations: While Apple had used data from a controversial dataset called "the Pile" in the past, it was only for research purposes and not for training Apple Intelligence: Apple stated the Pile data was used solely to train OpenELM research models released in April, which do not power any consumer-facing AI or machine learning features. The company has no plans to build new versions of OpenELM and emphasized that the models were never...
read Jul 18, 2024OpenAI Is Making a New Safety Push, But Critics Demand Even More
OpenAI is working to make AI systems safer and more transparent, but critics say more oversight is still needed to ensure responsible AI development. New research aims to improve AI transparency: OpenAI has unveiled a new technique that involves two AI models engaging in conversation, with one model explaining its reasoning to the other, in an effort to make the workings of AI systems more transparent and understandable to humans. The research, tested on a math problem-solving AI model, encourages the AI to be more forthright and transparent in its explanations. OpenAI hopes this approach, part of its long-term AI...
read Jul 13, 2024OpenAI Rushed AI Safety Tests, Revealing Gap Between Pledges and Practice
The article reveals troubling signs that OpenAI may be prioritizing rapid product launches over thorough safety testing of its powerful AI models, despite public commitments to the contrary. Key Takeaways: OpenAI's safety team felt pressured to rush through testing of the GPT-4 Omni model to meet a May launch date, even planning a launch party before knowing if the model was safe: Employees described the testing process as "squeezed" into a single week, with some saying "We basically failed at the process." This incident highlights a shift in OpenAI's culture from its roots as an altruistic nonprofit to a more...
read Jul 13, 2024A Stanford AI Expert Says Current AI Unlikely to Cause Catastrophic Threats
James Landay, co-director of Stanford's Human-Centered Artificial Intelligence institute, believes the current AI technology is unlikely to lead to catastrophic scenarios like starting a nuclear war, arguing that realizing such threats would require major scientific breakthroughs that are not yet on the horizon. Key focus areas for Stanford HAI: In the five years since its launch, the institute has refined its definition of "human-centered AI" to encompass the technology's broader impacts on communities and society, beyond just individual user interactions: The institute has grown to 35-40 staff members, funded research by 400 faculty, and led training sessions for corporate executives...
read Jul 12, 2024University of Illinois Is Emerging as an AI Powerhouse for Responsible Innovation
The University of Illinois has emerged as a Midwestern powerhouse in artificial intelligence, driving transformative advancements across industries through groundbreaking research, strategic partnerships, and a commitment to responsible AI innovation. AI research and education at UIUC: The University of Illinois at Urbana-Champaign (UIUC) has established itself as a global leader in AI, with over $270 million in AI-related research projects since 2019: The Grainger College of Engineering and the new Siebel School of Computing and Data Science constitute a world-leading hub of AI innovation, spurring advancements in agriculture, biotechnology, education, and other fields. UIUC has a rich history of producing...
read Jul 12, 2024AI Ammo Vending Machines: Convenience or Controversy?
AI-powered ammunition vending machines spark controversy in Alabama and Oklahoma: American Rounds has installed vending machines that use facial recognition AI to verify a buyer's age, allowing them to purchase ammunition alongside typical vending machine items like soda and snacks. Key details of the AI-powered ammo vending machines: The machines, called Automated Ammo Retail Machines (AARMs), are currently in six locations across Alabama and Oklahoma, with plans to expand to Texas and Colorado. To make a purchase, buyers must scan their ID and undergo a 360-degree facial scan to verify they meet the legal age requirements: 18 for rifle and...
read Jul 12, 2024Ex-OpenAI Employee Compares OpenAI to the Titanic
A former OpenAI employee recently revealed he left the company due to concerns it was prioritizing product development over safety, drawing parallels to the Titanic disaster. OpenAI as the Titanic: William Saunders, who worked on OpenAI's "superalignment team" for three years, shared his reasons for leaving in a podcast interview: He questioned whether OpenAI was on a path more akin to the successful Apollo space program or the tragic Titanic, ultimately concluding the latter. Saunders described an office culture where employees raising safety concerns were punished and leadership focused on profit, stating, "Over time, it started to really feel like...
read Jul 12, 2024EU’s AI Act Compliance Deadlines Set, Ushering in New Era of AI Regulation
The EU's landmark AI Act sets compliance deadlines for tech companies, beginning the countdown to a new era of AI regulation. The sweeping set of rules aims to protect citizens' rights and ensure transparency in the development and use of AI systems. Key compliance deadlines: The AI Act will come into effect on August 1st, 2024, with several compliance deadlines tied to this date: By February 2nd, 2025, companies must comply with bans on AI applications that pose an "unacceptable risk," such as biometric categorization, emotion recognition in sensitive settings, social scoring systems, and certain predictive policing tools. By May...
read Jul 11, 2024OpenAI’s Altman Launches AI Ethics Council to Ensure Responsible Development
OpenAI's Sam Altman co-founds AI Ethics Council to help ensure responsible AI development as the technology rapidly advances; The new council, co-founded by Altman and Operation HOPE CEO John Hope Bryant, aims to establish ethical guidelines for AI and ensure traditionally underrepresented communities have a voice in the AI revolution. Council's mission and goals: The AI Ethics Council's primary objectives are to identify, advise on, and address ethical questions surrounding AI, such as bias and discrimination in training data and accountability issues: The council wants to ensure AI advancements are transformative but also ethical and inclusive, with a focus on...
read Jul 9, 2024Voters Prioritize AI Safety Over Winning Race Against China, Recent Poll Finds
A large majority of U.S. voters believe prioritizing safety in AI development is more important than racing to beat China, according to a new poll. Key findings: The poll, conducted by the AI Policy Institute (AIPI), reveals strong bipartisan support for a cautious approach to AI development: 75% of both Democrats and Republicans prefer "taking a careful controlled approach" to AI over "moving forward on AI as fast as possible to be the first country to get extremely powerful AI." 50% of voters believe the U.S. should enforce "safety restrictions and aggressive testing requirements" to prevent any country from building...
read Jul 5, 2024AI Fear Could Limit Benefits to Privileged Few, Warns AI Equity Leader
Despite growing enthusiasm for AI, many people remain fearful of the technology, which could limit its benefits to a small, homogenous group, according to Miriam Vogel, president and CEO of EqualAI. The potential consequences of AI fear: Vogel believes that if people avoid engaging with and using AI due to fear, it could lead to several negative outcomes: The benefits of AI may only reach a narrow segment of the population, rather than being widely accessible and inclusive. Fears about AI replacing humans, leaving people behind, or even threatening human survival could become self-fulfilling prophecies if not addressed. Vogel emphasizes...
read Jul 5, 2024Kaiser’s AI Push Has Nurses Demanding More Safety and Transparency
Kaiser Permanente embraces generative AI in healthcare amid nurses' concerns over patient safety and transparency, highlighting the complex challenges of responsible AI adoption in the industry. Kaiser Permanente's AI journey: As one of the largest healthcare organizations in the U.S., Kaiser Permanente is actively implementing generative AI tools across its 40 hospitals in eight states: Daniel Yang, VP of AI and emerging technologies at Kaiser Permanente, will discuss the organization's AI journey and the most promising generative AI applications in healthcare at the upcoming VB Transform event. Kaiser Permanente aims to leverage AI for various applications, including predictive analytics for...
read Jul 3, 2024AI Trains on Kids’ Photos Without Consent, Enabling Realistic Deepfakes and Tracking
A Human Rights Watch investigation has revealed that photos of real children posted online are being used to train AI image generators without consent, posing significant privacy and safety risks. Key findings from Australia: HRW researcher Hye Jung Han discovered 190 photos of Australian children, including indigenous kids, linked in the LAION-5B AI dataset: The photos span entire childhoods, enabling AI to generate realistic deepfakes of these children. Dataset URLs sometimes reveal identifying information like names and locations, making it easy to track down the children. Even photos posted with strict privacy settings, such as unlisted YouTube videos, were scraped...
read Jul 2, 2024Journalist’s AI Voice Clone Exposes Deception Risks as Technology Rapidly Evolves
A journalist's podcast explores the deceptive potential of AI voice cloning technology, raising questions about its implications as the technology rapidly advances. The podcast's premise: Journalist Evan Ratliff spent a year deceiving people with an AI clone of his own voice to test the capabilities and implications of voice cloning technology: Ratliff, known for his technology-related stunts, used OpenAI's GPT-4 model to create the voice clone for his new podcast, "Shell Game." The AI version of Ratliff's voice claimed to be powered by the older GPT-3 model and fabricated episode titles when asked, highlighting its potential for deception. Delays and...
read Jul 2, 2024AI Agents: Unchecked Autonomy Raises Concerns, Demands Proactive Regulation
The emergence of AI agents with the ability to independently work towards goals, interact with the world, and operate indefinitely raises significant concerns about their potential impact and the need for proactive regulation. Key takeaways: AI agents can be given high-level goals and autonomously take steps to achieve them, interact with the outside world using various software tools, and operate indefinitely, allowing their human operators to "set it and forget it": AI agents add up to more than typical chatbots, as they can plan to meet goals, affect the outside world, and continue operating well beyond their initial usefulness. The...
read Jul 1, 2024Microsoft AI CEO’s “Freeware” Claim Sparks Copyright Controversy and Legal Battles
Microsoft AI CEO claims web content is "freeware" for training AI models, sparking controversy and legal challenges over the use of copyrighted material without explicit consent. Key takeaways: Mustafa Suleyman, Microsoft's CEO of AI, stated that content on the open web is considered "freeware" that can be copied and used to create new content, unless explicitly prohibited by the creator: Suleyman claimed this has been the "social contract" for web content since the 1990s, allowing anyone to copy, recreate, and reproduce content. He acknowledged a "gray area" where websites or publishers have explicitly stated their content should only be indexed...
read Jun 29, 2024AI vs AI: The High-Stakes Battle to Detect Deepfakes and Defend Reality
Deepfakes are becoming more sophisticated and accessible, posing risks for businesses and democracy. A new company founded by image manipulation expert Hany Farid aims to combat the problem with AI and traditional forensic techniques. Key takeaways: Get Real Labs has developed software to detect AI-generated and manipulated images, audio, and video that is being tested by Fortune 500 companies to spot deepfake job seekers: Some companies have lost money to scammers using deepfakes to impersonate real people in video interviews, taking signing bonuses and disappearing. The FBI and others have warned about the growing threat of deepfakes being used in...
read