The Artificial Intelligence Podcast

New York City's AI-powered chatbot intended to assist business owners in understanding government regulations has been found to provide misleading information that may encourage illegal activities. After five months of its launch, the chatbot has been observed to offer incomplete and dangerously inaccurate guidance on housing policies, worker rights, and rules for entrepreneurs. An example is its incorrect information that landlords are not obliged to accept tenants with Section 8 vouchers or rental assistance, whereas it is illegal to discriminate based on income source in the city. This finding raises concerns about the chatbot's reliability and underscores the importance of testing and verification before implementing AI-powered systems.

Microsoft's Copilot AI service will soon be able to run locally on PCs, thanks to built-in neural processing units (NPUs) with over 40 trillion operations per second (TOPS) of power. By running more elements of Copilot locally, lag time will be reduced and performance and privacy may be improved. Currently, Copilot runs primarily in the cloud, causing delays for smaller tasks. Intel's Lunar Lake chips, shipping in 2025, will have triple the NPU speeds of its current chips. Microsoft is also expanding Copilot's capabilities in Teams, allowing it to pull insights from both meeting chat and call transcripts, help rewrite messages, and generate new messages based on chat context. Microsoft is also introducing features to improve hybrid meetings, such as individual video feeds for each attendee and automatic camera switching for the best view. These updates will be rolled out in the coming months.

Artificial intelligence (AI) is revolutionizing eye exams, particularly in the detection of diabetic retinopathy, the leading cause of blindness among working-age adults. An AI algorithm implemented in clinics is able to perform eye exams using retinal images, providing quick diagnoses without the need for a doctor to be present. This saves time, reduces costs, and increases accessibility for patients. Several companies have received FDA approval for their AI eye exams, and the Centers for Medicare & Medicaid Services (CMS) has provided a reimbursement code to facilitate adoption. The technology is evolving, with ongoing studies exploring new applications and potential advancements in screening for other eye diseases.

Tech company Meta is set to release a major update for its Ray-Ban Meta camera glasses, incorporating artificial intelligence (A.I.) software. Priced at $300 and up, the glasses can now scan landmarks, translate languages, and identify objects such as animal breeds and exotic fruits, among other tasks. Activated by saying "Hey, Meta," followed by a request, the A.I. delivers responses through the glasses' speakers using computer-generated voice. Early testers found the glasses to be both amusing and impressive in their abilities, although there may still be some limitations and room for incorrect information. The incorporation of A.I. software provides users with a unique and interactive way to engage with the world.

Microsoft has implemented new security measures to safeguard its AI chatbots from malicious attacks. The tools, integrated into Azure AI, aim to prevent prompt injection attacks that manipulate the AI system into generating harmful content or extracting sensitive data. Microsoft is also addressing concerns relating to the AI system's quality and reliability, with prompt shields to detect and block injection attacks, groundedness detection to identify AI "hallucinations," and safety system messages to guide model behavior. The collaboration between Microsoft and OpenAI has been crucial in training AI models using diverse datasets and propelling generative AI forward. These measures underscore Microsoft's commitment to responsible AI usage.

The Writers Guild of America (WGA) and other unions representing film and TV writers and journalists have sent a letter to Senate Majority Leader Chuck Schumer, requesting protections for their industries in relation to Artificial Intelligence (AI) legislation. The letter raises concerns over the use of AI-generated content to replace journalists' work, unauthorized use of writers' work by AI developers, and the need for safeguards to protect the rights and compensation of creative professionals. The unions call for immediate action to prevent the abuse of AI technology and ensure the preservation of writers' work in the face of advancing technology.

AI Ph.D. graduates are increasingly choosing to work for large technology companies, which could have negative implications for open innovation. The debate between open and closed AI models is overshadowing a comprehensive understanding of openness in AI. Open science, transparency, and equity are crucial for developing AI that serves the public interest. The Partnership on AI exemplifies open innovation by bringing together academia, civil society, industry partners, and policy-makers to collaborate on AI challenges. Public funding and open publication of research foster an open ecosystem. However, recent data shows a decline in AI Ph.D.s pursuing academia, limiting open models and research availability. Transparency and accountability are important for responsible and safe AI deployment. Creating an open ecosystem requires investment in skills and education while promoting diversity of perspectives. Establishing research and open innovation systems that foster scientific integrity and resilience is essential. Incorporating diverse voices is imperative for AI development.

The NYPD plans to test metal detectors equipped with AI technology in subway stations to address the presence of guns in the transit system. The scanners, manufactured by Evolv, are being deployed in response to an increase in subway-related killings. The decision to implement the scanners follows the deployment of National Guard soldiers and increased police presence in the subway system. The scanners, which will not use facial recognition technology and will be mobile within the stations, will not be mandatory for riders. The exact number and locations of the scanners have not been disclosed, and the decision to expand their use will depend on the results of the testing phase. Evolv is currently under investigation for overstating the capabilities of its technology.

The rise of AI-driven deepfakes is creating challenges for identity verification, forcing CIOs and IT leaders to understand the impact on identity management processes. The process typically involves submitting a photo ID and taking a selfie for biometric comparison. However, deepfakes pose a risk by allowing attackers to present fraudulent identities. To combat this, organizations should ensure their identity verification vendor uses robust liveness detection during the selfie step, actively prompting users or passively evaluating micro movements. Additionally, leveraging AI can enhance the process by training algorithms to detect deepfake attacks more accurately and addressing issues of demographic bias. Collaboration and vigilance through partnerships and bounty programs are also crucial for staying ahead of potential deepfake adversaries.

The growing popularity of artificial intelligence (AI) is causing concern in schools as instances of its misuse emerge. One high school student in Illinois used AI to create and circulate nude photos of classmates, sparking outrage among parents and prompting calls for stricter regulation. Similar incidents have been reported across the country, leading to a debate about policing AI use in schools. Some states have implemented laws against non-consensual creation of explicit images using AI, but legal experts argue that charging young individuals with felonies is excessive. Experts recommend that schools establish clear rules to prevent future misuse of AI technology.