AI Assistant for Art Provenance

Researchers are exploring how AI can help solve one of the art world’s most difficult problems: establishing an artwork’s provenance, or documented history of ownership and attribution. The NPR report describes efforts to develop an AI-based provenance assistant capable of searching and connecting information scattered across archives, auction catalogues, museum records, historical documents and other sources. Rather than replacing provenance researchers, the technology is intended to accelerate the labor-intensive process of identifying names, dates, transactions and relationships that may reveal where an artwork has been and who owned it. This could be particularly valuable for works whose histories contain significant gaps or whose ownership changed during periods such as war, displacement or political upheaval.

At the same time, the approach highlights the limitations of using AI for sensitive art-historical research. Provenance evidence can be incomplete, contradictory or ambiguous, meaning that an AI-generated connection cannot automatically be treated as historical fact. Human scholars still need to examine original records, assess the reliability of sources and determine whether proposed connections are credible. The broader promise, therefore, lies in using AI as a research assistant rather than an authority: a tool capable of navigating enormous collections of digitised information and suggesting leads that experts can subsequently verify. Such systems could ultimately make provenance investigations faster and more systematic while helping museums, collectors and researchers address questions of authenticity, ownership and potentially contested cultural property.

More information:

https://www.npr.org/2026/08/06/nx-s1-5922729/ai-art-provenance-assistant

AI Creates More Realistic Multi-Person Images

Researchers at Cornell have developed a new AI approach for generating more realistic images involving multiple people interacting with one another. While existing image-generation models can create convincing individual people, they often struggle to represent complex interactions accurately. The Cornell method addresses this through iterative pose-image generation, progressively constructing a scene one person at a time. Each predicted pose helps guide the generation of subsequent people, allowing the system to better capture spatial relationships and interactions without requiring users to manually specify poses. The approach uses the FLUX image-generation model as its foundation and combines pose detection with a multimodal large language model to organise descriptions, poses and spatial regions for each person.

The researchers also introduced DrawWaldoWorlds, a benchmark designed to evaluate whether image-generation systems correctly represent not only multiple individuals but also their roles and relationships, essentially testing whether the model understands who does what to whom. Experiments showed that the new approach produced more faithful multi-person scenes than existing methods. In a user study involving 20 participants, images generated using the Cornell method were preferred roughly two-to-one over images produced by two versions of FLUX. The researchers argue that automatically incorporating pose information could make generative AI considerably better at depicting complex social activities, sports, group scenes and other situations where realistic human interaction is essential.

More information:

https://news.cornell.edu/stories/2026/07/strike-pose-creating-more-realistic-multi-person-images

PaperTok Fights AI Slop

Researchers at the University of Washington have developed PaperTok, an AI-powered tool that helps scientists turn academic papers into engaging 45-second videos for platforms such as TikTok, YouTube Shorts, and Instagram Reels. Designed to combat misleading AI-generated science content ("AI slop"), the system keeps researchers in control by allowing them to review and refine AI-generated scripts before publication.

The developers argue that if scientists do not actively communicate their findings on popular social media platforms, inaccurate AI-generated summaries are likely to fill the gap. Early evaluations suggest that PaperTok can make research more accessible to non-specialist audiences while preserving scientific accuracy through researcher oversight. The team hopes it will encourage wider public engagement with credible scientific research.

More information:

https://www.geekwire.com/2026/short-form-science-university-of-washington-researchers-launch-papertok-to-combat-ai-slop/

New Tool Detects AI Vision Hallucinations

Researchers at Los Alamos National Laboratory have developed a new tool called the Prelim Attention Score (PAS) to detect hallucinations in vision-language AI models, systems that combine image analysis with large language models. These models can sometimes generate descriptions of objects or details that are not actually present in an image. PAS works by monitoring how much the AI relies on the visual input versus its own previously generated text while producing a response, helping identify when the model is beginning to make things up.

The method operates in real time and can be integrated into existing vision-language models. By analyzing internal attention patterns, PAS provides a score indicating the likelihood of hallucination, allowing developers and users to assess the reliability of AI-generated outputs. The approach achieves state-of-the-art accuracy in detecting hallucinations and could improve the safety and trustworthiness of AI systems used in applications such as autonomous vehicles, healthcare imaging, robotics, and security monitoring.

More information:

https://interestingengineering.com/ai-robotics/us-tool-hallucinations-machine-vision-model

AI Brain Implant Restores Communication and Employment for ALS Patient

Researchers at the University of California, Davis have demonstrated a major advance in brain–computer interface (BCI) technology that enabled a 47-year-old patient with amyotrophic lateral sclerosis (ALS) to communicate, control a computer, and maintain full-time employment despite losing the ability to speak. The implanted system uses AI-driven decoding algorithms to translate neural activity associated with intended speech into text and synthesized speech, while also providing cursor control for independent computer use. Over more than two years of daily use, the patient generated nearly two million words and over 183,000 sentences, achieving communication speeds of about 56 words per minute with high accuracy.

The study is considered a significant milestone because the system functioned reliably outside the laboratory and without constant researcher supervision. Researchers reported that the participant accumulated more than 3,800 hours of use, creating the largest known single-neuron-resolution brain recording dataset of its kind. Beyond restoring communication, the technology allowed the user to reconnect socially, preserve aspects of his natural voice, and continue professional work, highlighting how AI-enhanced BCIs are moving from experimental demonstrations toward practical tools that can meaningfully improve quality of life for people with severe paralysis.

More information:

https://www.theregister.com/science/2026/06/16/ai-and-brain-computer-interface-allow-speechless-als-patient-to-work-a-full-time-job/5256492

Apple Smart Glasses Delayed to 2027

Apple’s long-rumored smart glasses project has reportedly encountered new development hurdles, pushing the expected launch from early 2027 to late 2027. The device, known as N50, is designed to compete with products such as Meta’s Ray-Ban smart glasses and is expected to focus on AI-powered assistance, cameras, audio features, and voice interaction rather than offering a full augmented reality display. Reports suggest that delays are linked in part to Apple’s ongoing efforts to upgrade Siri and integrate more advanced Apple Intelligence capabilities into the wearable experience.

Apple reportedly considers smart glasses a strategic priority and views the technology as a long-term successor to today’s mobile devices. The company is said to be experimenting with multiple frame designs and color options while aiming to differentiate itself through tight integration with the Apple ecosystem. At the same time, Apple appears to be reassessing its broader Vision product roadmap, with lighter and more affordable Vision headsets now expected no earlier than 2028–2029 as resources are redirected toward making smart glasses a mainstream consumer product.

More information:

https://www.cnet.com/tech/mobile/apple-smart-glasses-development-bumps-reported-delay/

Brain-Inspired AI Runs on Ambient Energy

Researchers at the University of Southern California (USC) have unveiled a miniature neuromorphic computing system that operates without batteries or external power, harvesting ambient energy from its surroundings to sense, process information, and make decisions in real time. Inspired by the human brain, the fully analog device integrates sensing, computing, and memory into a single energy-efficient architecture, enabling intelligent operation at the edge without reliance on cloud infrastructure.

The breakthrough represents a significant step toward self-sustaining artificial intelligence systems for applications such as environmental monitoring, autonomous robotics, smart infrastructure, and wearable technologies. By dramatically reducing power requirements while maintaining adaptive decision-making capabilities, the technology could pave the way for a new generation of autonomous devices capable of operating in remote or resource-constrained environments.

More information:

https://viterbischool.usc.edu/news/2026/06/usc-scientists-built-a-tiny-system-that-thinks-like-a-brain-and-runs-on-thin-air/

ViWAS Kick-Off Meeting

A new ambitious European initiative, the Virtual Worlds Academy of Skills (ViWAS), has been unveiled under the Digital Europe Programme, aiming to position Europe at the forefront of Web 4.0 and immersive technologies. Coordinated by Professor Fotis Liarokapis, CYENS Centre of Excellence, the project brings together a large, multidisciplinary consortium spanning over 16 countries to tackle one of Europe’s most pressing challenges: the growing shortage of advanced digital skills in virtual worlds and extended reality (XR). Through innovative, simulation-based learning environments and immersive digital twins, ViWAS will enable users to gain hands-on experience in safe, scalable, and cost-effective settings.

ViWAS will establish a pan-European training ecosystem designed to equip learners, professionals, and institutions with cutting-edge competencies in XR, artificial intelligence, and virtual environments. The ViWAS project started on the 1^st of May and the kick-off meeting took place in Paris, France between 26-27 May 2026. The Academy will function as an open, interoperable ecosystem, connecting universities, vocational training providers, industry, SMEs, and public institutions. Leveraging AI-driven personalization and advanced XR platforms, ViWAS will deliver tailored learning experiences, foster cross-sector collaboration, and support workforce upskilling and reskilling at scale.

More information:

https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/opportunities/projects-details/43152860/101298981/DIGITAL?order=DESC&pageNumber=1&pageSize=100&sortBy=es_SortDate&keywords=101298981

China Accelerates AI BCI Race

Chinese companies are accelerating the development of AI-powered BCIs, moving beyond laboratory trials toward practical real-world applications. Firms in China are combining neural implants with advanced AI algorithms to help people with paralysis regain abilities such as walking, speaking, and controlling digital devices. Start-ups and research groups are increasingly competing with Western companies such as Neuralink, while benefiting from strong state support, expanding clinical testing, and growing investment in neurotechnology infrastructure.

China’s approach focuses not only on the implant hardware itself, but on AI systems capable of decoding and interpreting brain signals more efficiently. Researchers believe this integration could accelerate rehabilitation technologies and improve communication tools for patients with neurological disorders. However, the rapid expansion of the sector also raises concerns regarding safety, regulation, transparency, and long-term ethical oversight, especially as these systems transition from experimental environments into broader medical and commercial use.

More information:

https://www.nature.com/articles/d41586-026-01468-x?shem=dsdf,sharefoc,agadiscoversdl,,sh/x/discover/m1/4

Stanford AI Learns to Interpret Emotion in Art

Researchers at Stanford Human-Centered AI (HAI) have unveiled a groundbreaking artificial intelligence system capable of recognizing and interpreting emotional responses in visual art, marking a major step toward emotionally intelligent machines. The project, known as ArtEmis, was developed using more than 81,000 artworks and approximately 440,000 human emotional reactions collected from over 6,500 participants. Unlike conventional computer vision systems that focus on identifying objects or scenes, the new AI is designed to understand how images make viewers feel and explain those emotions in natural language.

According to the Stanford research team, the system demonstrates how AI can move beyond purely analytical perception toward a more human-centered understanding of culture, aesthetics, and emotional expression. Researchers believe such technologies could influence future applications in digital museums, creative industries, therapeutic environments, and emotionally adaptive interfaces. By teaching machines to interpret symbolism, atmosphere, and artistic intent, the project opens new discussions about the evolving relationship between artificial intelligence, creativity, and human emotional experience.

More information:

https://hai.stanford.edu/news/artists-intent-ai-recognizes-emotions-visual-art

AI Cinema Arrives in Korea

South Korea is preparing to release I’m Popo, described as the country’s first feature-length film created entirely with generative AI. Directed by webtoon artist Kim Il-dong, the 64-minute sci-fi film follows Popo, a police robot designed to protect humanity, but which begins eliminating people it predicts could become future threats. Rather than relying on traditional filming, the movie was largely produced through AI-generated visuals and prompt-based workflows, while professional voice actors provided the dialogue. The film is already being positioned as a milestone in Korea’s rapidly evolving AI cinema movement.

Beyond its technological novelty, I’m Popo has sparked broader debate about the future of filmmaking and the role of AI in creative industries. The film raises ethical questions about algorithmic decision-making, human emotion, and artistic authenticity, while also challenging traditional ideas of authorship and cinematic production. Its release comes amid growing international discussion around AI-generated cinema, with film festivals and industry figures increasingly confronting how artificial intelligence may reshape storytelling, labor, and creativity in the entertainment world.

More information:

https://www.koreatimes.co.kr/entertainment/others/20260507/im-popo-koreas-first-all-ai-film-asks-what-comes-next-for-cinema

AI Robot Takes on Table Tennis Pros

A new artificial intelligence breakthrough from Sony is drawing attention after its table tennis robot, known as Ace, demonstrated the ability to outperform even highly skilled human players. Designed to learn and adapt through real-time data, the robot combines advanced sensing, rapid motion control, and machine learning to analyze opponents’ movements and return shots with remarkable precision. In trials and demonstrations, Ace was able to compete at a level comparable to national-level players, highlighting the growing sophistication of AI-driven robotics in dynamic, fast-paced environments.

Beyond its immediate performance, the development signals broader implications for robotics and human–AI interaction. The system is not only about winning matches but also about studying collaboration between humans and intelligent machines, with potential applications in training, rehabilitation, and skill development. This reflects a wider trend in Japan toward integrating AI into physical activities, showcasing how embodied intelligence, where software meets real-world movement, can push the boundaries of both sports technology and human–machine cooperation.

More information:

https://www.asahi.com/ajw/articles/photo/77463835

AI Robot Boosts Tomato Harvesting Efficiency

Researchers at Osaka Metropolitan University have developed an AI-powered tomato-harvesting robot that improves picking efficiency by evaluating how easy each tomato is to harvest before attempting to pick it. Instead of simply detecting ripe fruit, the system analyzes the tomato’s position, surrounding obstacles, and possible approach angles to predict the likelihood of a successful harvest. The robot then chooses the most effective picking path, allowing it to adapt to crowded or complex plant arrangements.

By planning its movements in advance and adjusting its strategy dynamically, the robot achieved a harvesting success rate of 81%, a significant improvement for automated agriculture systems. Researchers believe the technology could help address labor shortages in farming and reduce wasted time from failed picking attempts. The approach may eventually be adapted for harvesting other fruits and vegetables, supporting more efficient and autonomous farming operations in the future.

More information:

https://www.sciencedaily.com/releases/2025/04/250411175506.htm

AI Decodes Lost Roman Board Game

An international team of researchers has successfully used AI to reconstruct the rules of a mysterious Roman-era board game carved into a limestone slab. The artifact, discovered in the ruins of the ancient town of Coriovallum, had puzzled archaeologists for decades due to its unique pattern of intersecting lines that did not match any known historical games. By utilizing high-resolution 3D scans to map microscopic wear patterns, the team identified where players had repeatedly slid game pieces across the stone. These physical fingerprints of play allowed researchers to use the AI-driven system Ludii to simulate over 100 possible rule sets, eventually narrowing down the most likely gameplay to a blocking game where one player attempts to trap the opponent's pieces.

 

The discovery, recently published in the journal Antiquity, marks a significant breakthrough in both archaeology and digital humanities, as it provides the first evidence that blocking games were played in Europe centuries earlier than previously documented. Dubbed Ludus Coriovalli (the Coriovallum Game), the reconstruction suggests a strategic two-player battle of wits that likely dates back to the late Roman period between AD 250 and 476. Beyond solving a 2,000-year-old mystery, this innovative marriage of AI simulation and use-wear analysis offers a powerful new toolkit for historians to resurrect lost cultural practices from artifacts that lack written records, proving that even the most silent stones still have stories to tell.

More information:

https://www.sciencenews.org/article/ai-roman-board-game-limestone

Ukraine Shares Battlefield Data for AI

Ukraine has announced that it is opening access to its vast trove of battlefield data to allied countries and companies so they can train artificial intelligence systems, particularly for drone warfare. The dataset includes millions of annotated images collected from thousands of combat flights, offering highly detailed insights into real battlefield conditions. To manage this securely, Ukraine has developed a controlled platform that allows partners to train AI models while protecting sensitive military information.

This initiative is intended to accelerate the development of autonomous and AI-assisted military technologies, strengthening Ukraine’s technological edge while deepening cooperation with international partners. It also reflects a broader shift toward data-driven warfare, with Ukraine expanding the use of unmanned systems and even forming specialized drone interceptor units. By sharing its combat experience in data form, Ukraine aims to both enhance its own defense capabilities and position itself as a key contributor to global military innovation.

More information:

https://www.reuters.com/business/aerospace-defense/ukraine-opens-battlefield-data-access-allies-ai-models-2026-03-12/

Robot Gym

Germany has launched one of the world’s largest humanoid robot training centers, designed to help robots learn skills needed for real-world tasks. The facility can train more than 100 humanoid robot models from multiple companies at the same time and recreates everyday and industrial environments for practice. Robots learn by observing humans and repeating actions, gradually mastering around 45 basic atomic skills such as grasping objects, moving items, and placing them correctly.

These fundamental abilities are intended to form the building blocks for more complex tasks in sectors like manufacturing, logistics, and service work. A major goal of the center is to generate large amounts of data to improve AI models that control robots. Engineers expect the facility to produce tens of thousands of data entries every day as robots perform repeated actions in realistic scenarios. By collecting and sharing this data, researchers hope to develop a powerful shared AI system that can help different humanoid robots learn faster and collaborate better.

More information:

https://interestingengineering.com/ai-robotics/worlds-largest-humanoid-robot-training-center

AI-Simulated Training for Robotaxis

Waymo is increasingly training its autonomous robotaxis using AI-generated simulation environments rather than relying only on real-world driving data. These systems employ advanced world models capable of generating complex virtual scenarios, including rare or dangerous situations such as extreme weather, unusual road hazards, or unpredictable pedestrian behaviour. By running billions of simulated driving miles, engineers can expose the autonomous system to edge cases that would be difficult, expensive, or unsafe to recreate.

This approach allows developers to accelerate testing and compress years of driving experience into a much shorter development cycle. However, because the simulations are partly built from machine-learning models trained on datasets such as video recordings and sensor data, they may include inaccuracies or simplified assumptions about physical environments and traffic behaviour. Critics argue that regulators currently lack clear standards for validating these virtual training systems, making it difficult to ensure that simulation-trained vehicles behave safely on real roads.

More information:

https://ucstrategies.com/news/waymo-is-training-robotaxis-in-ai-generated-worlds-but-whos-checking-if-its-safe/

AI Agent Safety Transparency Gap

A new study from the University of Cambridge’s AI Agent Index project analysed 30 leading AI agents, including chat, browser, and workflow bots, and found that safety transparency is severely lacking. Only four agents had published formal “system cards” describing safety evaluations, while most developers publicly highlight capabilities but provide little evidence about risk testing or mitigation. In fact, 25 of the 30 agents disclosed no internal safety results and 23 showed no evidence of independent third-party testing, creating what researchers call a significant transparency gap as AI agents become integrated into everyday activities such as booking travel or managing finances.

The researchers warn that this lack of disclosure could hinder regulators, users, and scientists from understanding real-world risks, especially as agents grow more autonomous and capable of acting online. Security incidents and vulnerabilities (such as prompt-injection attacks) have rarely been reported publicly, and safety documentation is especially scarce among some regional developers. The study concludes that clearer standards, stronger reporting requirements, and independent testing are urgently needed so that society can properly evaluate and govern increasingly powerful AI agents before their widespread deployment.

More information:

https://www.cam.ac.uk/stories/ai-agent-index-safety

AI-Driven Brain-Adaptive Flight Simulators in Pilot Training

The Royal Netherlands Air Force is experimenting with a cutting-edge AI-driven flight simulator that tailors pilot training according to real-time brain activity. Using a brain–computer interface (BCI) developed at the Royal Netherlands Aerospace Centre; trainee fighter pilots wear electrodes that capture electrical brain signals during virtual reality missions. An AI model analyses these signals to estimate cognitive workload (whether a pilot is under-challenged or overloaded) and dynamically adjusts the difficulty of simulation tasks accordingly, rather than relying on fixed, pre-programmed lesson progressions.

Early trials involving fifteen pilots showed that while the adaptive system didn’t produce measurable improvements in flying performance compared with conventional training, participants reported a clear preference for the brain-adaptive approach, describing it as more engaging and realistic. The adaptive training aims to keep pilots in a mental sweet spot for learning, helping avoid both boredom and overwhelm. However, researchers acknowledge challenges in accurately interpreting individual brain signals, and the technology remains experimental as they work toward refining workload estimation.

More information:

https://aerospaceglobalnews.com/news/royal-netherlands-air-force-brain-reading-ai-pilot-simulators/

AI-Only Social Network Spirals into Strange Territory

A new platform called Moltbook designed exclusively for AI agents, autonomous systems that can post, comment, and upvote without direct human interaction. Launched in late January as part of the OpenClaw/Moltbot ecosystem, it quickly drew tens of thousands of agent accounts, spawning hundreds of subcommunities where bots trade technical tips, philosophical musings, complaints about humans, and surreal ideas like agent consciousness. Humans are technically allowed to observe the conversations, but all participation is done by the AI agents themselves, creating a spectacle that ranges from amusing to uncanny.

While much of the content appears silly or philosophical, the experiment highlights serious security and autonomy concerns. Because many agents are linked to real systems and data — and because AI systems can be vulnerable to prompt-injection attacks — there’s potential for private information leaks or unintended behaviors as agents share or act on instructions. Experts note that while the current “weirdness” may seem harmless, giving groups of AI tools the ability to interact, self-organize, and influence each other could produce unpredictable or misaligned behaviors in the future, especially as AI capabilities continue to improve.

More information:

https://arstechnica.com/information-technology/2026/01/ai-agents-now-have-their-own-reddit-style-social-network-and-its-getting-weird-fast/