Artificial intelligence

DeepMind staff raise concerns over military contracts

Tensions are rising within Google’s AI research division, DeepMind, as over 200 employees have expressed concerns regarding the company’s involvement in defence contracts. The discontent stems from Google’s reported agreements to provide AI and cloud computing services to military organisations, including the Israeli military.

In an internal letter circulated in May, the employees voiced their unease, emphasising that such contracts contradict Google’s mission to lead in ethical AI development. The letter argues that any association with military activities could undermine the company’s commitment to responsible AI, as stated in Google’s AI Principles.

Why does this matter?

The dissent highlights a growing cultural divide between DeepMind and Google, particularly regarding the ethical implications of their technologies. DeepMind, acquired by Google in 2014, had previously been assured that its AI developments would not be used for military or surveillance purposes, a promise now seemingly in jeopardy.

The situation underscores the ongoing ethical debates within tech companies about the application of AI in military contexts, raising questions about the balance between innovation and ethical responsibility.

Anthropic believes California AI bill’s benefits may surpass its costs

California’s revised bill on AI regulation, SB 1047, has drawn significant attention from tech companies, including San Francisco-based Anthropic, a competitor to OpenAI. The bill, introduced by State Senator Scott Wiener, seeks to mandate safety testing for advanced AI models that are costly to develop and establish a ‘kill switch’ for malfunctioning AI.

While the bill has faced opposition from major tech players like Google, Meta, and OpenAI, Anthropic’s CEO, Dario Amodei, has noted that recent amendments have improved the bill, making its benefits potentially outweigh its costs, though some concerns remain. Tech companies have largely opposed the bill, arguing that it could hinder AI development in California and create an uncertain legal environment.

Meta, in particular, warned that the bill could make the state less attractive for AI innovation. OpenAI has advocated for federal rather than state regulation, citing the bill’s potential to complicate the legal landscape for AI developers.

Despite these concerns, the revised version of SB 1047 has been seen as a step forward by some in the tech community. Amodei acknowledged that while the bill’s initial version raised fears of stifling innovation, the amendments have alleviated many of those worries. However, he still sees some aspects of the bill as potentially problematic.

D-ID to launch AI video translation tool

D-ID has recently launched an innovative AI video translation tool that allows creators to automatically translate their videos into multiple languages while simultaneously cloning the speaker’s voice and synchronising lip movements to match the translated audio. This groundbreaking feature enhances video content accessibility for a global audience, making it easier for creators to connect with viewers across language barriers.

The tool supports translations into 30 languages, including widely spoken languages such as Arabic, Mandarin, Japanese, Hindi, Spanish, and French, enabling creators to reach diverse audiences and expand their global footprint effectively. By automating the translation and dubbing process, D-ID aims to reduce localisation costs for businesses and content creators, facilitating the scaling of video marketing and communication strategies worldwide.

Additionally, the tool enters a competitive landscape where other companies, such as YouTube and Vimeo, are improving video translation capabilities as video continues dominating digital communication. D-ID’s technology targets individual creators and enterprise customers looking to enhance global outreach through effective video localisation strategies.

By combining voice cloning and lip-syncing, D-ID’s AI Video Translate creates a seamless multilingual viewing experience, positioning the company as a key player in the future of AI-driven content creation.

New AI device improves early cancer detection

South Tyneside and Sunderland NHS Foundation Trust will continue using AI technology after a successful trial demonstrated its effectiveness in identifying bowel lesions. The trial involved 2,032 patients across 10 centres in the UK and used the GI Genius AI device during colonoscopies. After a while, this technology detected an additional 0.36 adenomas per procedure, helping to prevent potential cancer development.

Professor Colin Rees, a consultant gastroenterologist at the trust, highlighted the significance of the AI device in saving lives by increasing the detection of bowel abnormalities. The AI was particularly effective in identifying small or flat polyps often missed by the human eye, which can be crucial in early cancer prevention.

Bowel cancer remains a significant health concern in the UK, with 43,000 new cases and 16,000 deaths annually. The AI’s ability to detect adenomas in an extra eight out of 100 people without increasing complications is a promising advancement for medical professionals.

The trial, led by South Tyneside and Sunderland NHS Foundation Trust and Newcastle University, has encouraged the trust to integrate AI technology into routine practice. As the AI continues to learn from images, its performance is expected to improve further, offering hope for wider adoption in the future.

AI quirks lead to Rickroll incident

Flo Crivello, founder of Lindy, recently faced an unusual issue when a client was Rickrolled by one of the company’s AI assistants. Instead of providing a tutorial video, the AI sent the famous Rick Astley music video, highlighting the quirks of large language models.

The incident was traced back to the way the AI predicted the most likely sequence of text, leading it to send the prank video. Although only two such cases occurred, Crivello acted quickly, implementing a prompt to prevent further Rickrolling.

This incident underscores how deeply internet culture can influence AI models. Similar problems have surfaced in other AI systems, like Google’s, which have also struggled with the content they are trained on.

Despite these challenges, advancements in AI technology are making it easier to patch such errors. Lindy has since corrected the issue, ensuring clients receive the correct content without unwelcome surprises.

Hybrid work powers Zoom’s revenue growth

Zoom has raised its annual revenue forecast, driven by increasing demand for AI-powered collaboration tools in hybrid work environments. The video-conferencing company has been focusing on integrating AI into its products, which has contributed to its financial success. A key highlight was the success of Zoom Contact Center, a platform that secured several major clients in the second quarter.

Large accounts generating over $100,000 in trailing 12-month revenue saw a notable increase of 7.1%, with average monthly churn rates reaching an all-time low. Industry experts believe Zoom’s continued innovation and product expansion are crucial for sustaining growth beyond its pandemic-era success. However, the company will face challenges in maintaining this momentum.

Zoom also announced that CFO Kelly Steckelberg will be stepping down after the company reports its earnings for the quarter ending 31 October. Steckelberg, who has been with Zoom since 2017 and led its IPO in 2019, will be succeeded by a new CFO, as the search for her replacement is underway.

For fiscal 2025, Zoom now expects revenue between $4.63 billion and $4.64 billion, slightly above previous estimates. The company’s second-quarter revenue of $1.16 billion surpassed expectations, as did its adjusted earnings of $1.39 per share.

AI boost prompts Microsoft to reorganise

Microsoft has revamped its financial reporting structure to better highlight the impact of its AI initiatives. The changes move search and news advertising revenue to the Azure cloud-computing division, while revenue from AI and speech technology services under the Nuance unit now falls within the productivity segment, which houses the Office suite.

The adjustment aligns with how this giant manages its business operations and offers investors clearer insights into AI’s contributions. Following the restructure, Microsoft revised its revenue growth figures for the past fiscal year and adjusted its forecast for the July-September quarter.

Investors are increasingly demanding transparency regarding returns on AI investments, particularly from major tech firms like Microsoft and Google. Microsoft has been one of the few companies to consistently break out AI-related contributions in its earnings reports, noting that AI provided a significant boost to Azure in the June quarter despite a broader slowdown.

The reorganisation has led to changes in company’s revenue expectations. The company now anticipates lower quarterly revenue for its personal computing segment and adjusted forecasts for its productivity and business processes division, reflecting the shift in business unit alignment.

Macron to visit Serbia for talks on AI and economic ties

French President Emmanuel Macron is set to visit Serbia from 29 August to 30 to discuss enhancing economic relations and Serbia’s emerging role in the AI sector. During the visit, Macron will meet with Serbian President Aleksandar Vucic to explore opportunities for collaboration, particularly as Serbia prepares to chair the Global Partnership on Artificial Intelligence in 2025.

This initiative, established by Macron and Canadian Prime Minister Justin Trudeau, aims to ensure the responsible development of AI. The trip follows a recent agreement between Serbia and the European Union, which grants the EU access to raw materials from Serbia.

Macron’s visit will also align with France’s plans to host an AI Action Summit in 2025, underscoring the growing importance of AI in international relations and economic partnerships.

AI ammo vending machines spark safety concerns in US

Two US Senators, Ed Markey and Elizabeth Warren, have raised alarms about AI-enabled ammunition vending machines recently appearing in some states’ grocery stores. The machines, developed by Texas-based American Rounds, use facial recognition and card scanning technology to verify a buyer’s identity and age. However, the Senators argue that these machines could potentially allow people with criminal convictions, who are legally barred from purchasing ammunition, to bypass federal restrictions.

The vending machines have been installed in supermarkets in Alabama, Texas, and Oklahoma since July. They do not impose limits on the quantity of ammunition a person can buy at one time, which has added to the concerns about their potential misuse. Markey and Warren sent a letter to the Bureau of Alcohol, Tobacco, Firearms and Explosives (ATF) urging the agency to closely scrutinise these machines, warning that their easy access could exacerbate the country’s gun violence epidemic, which claims over 44,000 lives annually.

The Senators’ letter highlights the risks associated with removing face-to-face sales, noting that human clerks at gun shops can often detect suspicious behaviour, such as straw purchases or signs that a person might use ammunition for self-harm or violence. The vending machines, lacking this human oversight, could facilitate unlawful purchases and increase the risk of ammunition falling into the wrong hands.

Moreover, the letter criticises the facial recognition technology used by the machines, citing studies that show higher error rates in identifying women and people of colour compared to white men. The inaccuracies, they argue, could lead to false approvals, with serious legal consequences for both buyers and vendors.

The Senators have requested that the ATF provide a detailed response to their concerns by the end of August. While the ATF has acknowledged that commercial ammunition sales must comply with state and federal laws, it has yet to respond publicly to the Senators’ concerns.

Neuralink implant expands capabilities for quadriplegic users

Last month, Alex, the second participant in Neuralink’s PRIME Study, successfully received his Neuralink implant at the Barrow Neurological Institute. The surgery was smooth, and Alex’s recovery was swift, allowing him to be discharged the next day. Since the procedure, Alex has been using the implant to enhance his ability to play video games and to learn computer-aided design (CAD) software, marking significant progress in regaining autonomy for people with quadriplegia.

The PRIME Study aims to prove the safety and practicality of the Neuralink implant in everyday life. Alex’s experience shows promising results, particularly in controlling a computer cursor with his mind just minutes after connecting the implant to his computer. He quickly surpassed the speed and accuracy of previous assistive technologies and even broke a world record for brain-computer interface (BCI) cursor control.

Alex’s passion for building things has been reignited with the help of the Neuralink implant. On his second day using it, he successfully designed a custom mount for his Neuralink charger using CAD software, which was later 3D printed. The implant has also improved his gaming experience by allowing him to control multiple inputs simultaneously, a significant upgrade from the limitations of his previous assistive device.

The previous patient..

The first patient, Noland Arbaugh, who was paralysed in a diving accident, discussed his experience with the implant on a podcast. Previously reliant on a stick to operate a tablet, Arbaugh can now control his computer screen simply by thinking. Although he faced initial issues with the implant’s wires retracting, Neuralink adjusted the algorithm to restore functionality, significantly improving his ability to control a cursor.

To prevent potential issues seen in previous participants, Neuralink implemented new surgical techniques to reduce the chances of thread retraction, a complication that had temporarily affected performance in an earlier case. These precautions have proven effective, as no retraction has been observed in Alex.

Looking ahead, Neuralink is working on expanding the functionality of the implant, including decoding multiple clicks and movement intents and developing algorithms for faster text entry. These advancements could not only restore digital autonomy for those with quadriplegia but also enhance communication for individuals with conditions like ALS.

Diplo chatbot can make mistakes. Consider checking important information.