Physical AI

WIRED reports that artificial intelligence, after making striking advances in mathematics, is now being applied across the broader sciences. The piece suggests tools that proved themselves on proofs and problem-solving are being turned toward research in fields like physics, biology and chemistry, raising questions about how scientific discovery itself may change as AI becomes a working partner in the lab.

A new essay argues that "robot" is less a technical category than a social construct, contending that a dishwasher performs automated physical work yet is never called a robot, while similar machines are. The piece asks what people actually mean when they label a machine a robot, and why the term shifts with cultural expectations rather than engineering definitions.

A discussion of the intelligence explosion — the idea that an AI capable of improving itself could trigger rapid, runaway gains in machine capability — is drawing attention among technology readers. The concept, long associated with I. J. Good and popularised by figures like Nick Bostrom and Eliezer Yudkowsky, remains contested: proponents see it as the central risk of advanced AI, while skeptics argue progress faces physical and economic limits.

Physical Intelligence has launched a beta of its API, giving developers access to what it calls frontier AI models designed to run on any robot. The company positions the move as bringing general-purpose robot intelligence to a broad range of hardware, letting teams build capabilities without training their own models. The announcement is drawing attention in the robotics and developer communities, with discussion focused on what generalized robot foundation models could mean for automation and the wider robotics industry.

BodySharing Physical AI, described as the world's first physical AI that learns the movements and force control of everyday tasks done with bare hands, including proprioceptive sensing, has been unveiled for the first time at CEATEC in Japan. The announcement is drawing attention as a step toward robots and AI systems that can replicate delicate human manipulation in real-world settings.

Wired reports that artificial intelligence, having already made striking advances in mathematics and formal problem-solving, is now being applied to the broader sciences. Researchers are increasingly using AI systems to generate hypotheses, analyse experimental data and accelerate discovery across fields like biology, chemistry and physics, raising both excitement about faster breakthroughs and concern about reliability and scientific oversight.

A new benchmark called DishwasherBench, released by Composo AI on GitHub, challenges frontier language models to figure out how to stack a dishwasher efficiently. The quirky test measures spatial reasoning and everyday practical planning rather than academic knowledge, and it is drawing attention for highlighting how even top AI systems can struggle with mundane physical tasks people handle routinely.

Robotics startup Physical Intelligence has launched a beta of its API, giving developers access to what it calls frontier-level foundation models for robots. The company says the goal is to make advanced robot intelligence available to any machine, not just its own hardware. Announcement of the beta drew early attention on developer forums, though detailed community discussion had not yet picked up at time of reporting.

OpenAI's GPT-6 model, called Astra, is claimed to have solved 98% of new theoretical physics problems presented to it. The figure suggests a level of performance in advanced physics reasoning well beyond previous AI systems. Details on who set the problems, how the results were verified, and which institutions were involved have not been given, so the claim is so far unconfirmed.

A robotics startup has released a 50-billion-parameter "physical general foundation model" aimed at giving robots instant intelligence, according to Forbes. The model is pitched as a general-purpose AI system for physical machines, allowing robots to perceive and act without task-specific training. Details about the company, customers and real-world performance of the model were not included in the initial coverage.

Google's new Gemini 4 Argon model has undergone a full round of real-world testing, with results pointing to a major leap in its ability to handle physics problems. Reports describe the model's 'physics prowess' as awakened, suggesting capabilities well beyond previous versions of the Gemini line.

Robostrategy CEO Andrew Kang showcased 'Taku', the company's robot, and discussed the intensifying race in physical AI — the effort to bring artificial intelligence into machines that operate in the real world. Kang's appearance puts a spotlight on how startups and tech firms are competing to deploy embodied AI in commercial robotics.

Chinese robotics firm Microintelligence took on a demanding factory task: inspecting and grinding a die-cast component over two metres long. The job required roughly 3,000 imaging positions to be completed within ten minutes, showcasing how AI-powered robots are being deployed for precision industrial work that combines vision inspection with physical grinding on large metal parts.

Physicists have developed an early warning system that predicts when an AI system is about to behave in harmful or uncontrolled ways. The method draws on physics-based approaches to detect warning signs before an AI goes rogue. The work taps into growing public and scientific concern about AI safety and the difficulty of anticipating failures in advanced systems.

Analytics firm Seeq and Applied Computing have announced a partnership to bring physics-based AI to energy operations. The collaboration will combine Seeq's process data analytics with Applied Computing's AI technology, aiming to help energy companies improve operational efficiency, reliability and decision-making across their industrial assets.

Anthropic reports that current robotics technology is technically capable of performing roughly a third of physical jobs, but adopting robots for those roles is not yet cost-effective for most employers. The findings suggest the barrier to widespread automation in physical work is economic rather than technical, with costs needing to fall before large-scale deployment becomes viable across industries.

German robotics firm NEURA Robotics has introduced Neuron AI, which it describes as the brain of the physical world. The announcement, carried by AFP and issued via Business Wire, positions the platform as a cognitive system intended to give robots situational understanding and decision-making capabilities. Details on features, availability and partners remain limited in the initial coverage, but the launch adds to the growing wave of AI systems aimed at embodied, real-world robotics rather than purely software applications.

An explainer on the inner workings of AI chips is drawing large audiences worldwide, walking through what physically happens inside a semiconductor when a user submits a prompt to an AI system. Interest in how the hardware behind chatbots and AI assistants actually operates is surging, with the piece gathering close to 600,000 likes as viewers seek to understand the technology powering the current AI boom.

Bedrock Robotics is raising $600 million in new funding to develop technology that automates construction work, according to a report by Axios. The startup, founded by former Waymo and Tesla engineers, is part of a growing wave of robotics companies targeting heavy industry. The round would signal continued strong investor appetite for automation and physical AI ventures.

Apple has launched a new Apple TV 4K priced at $299, featuring an AI-powered Siri and support for voice commands without the need for a physical remote. The device is set to go on sale on November 13. The announcement is drawing attention from technology watchers discussing the upgraded assistant features and the price point.

Google DeepMind has unveiled Gemini 4 Argon, a model that can teach robots physical skills simply by watching videos of humans performing tasks. The announcement suggests robots could learn new abilities without explicit programming or hands-on demonstrations. The development is drawing attention from the robotics and AI communities as a potential step toward more adaptable, general-purpose robots.

Foreign Affairs argues that China's robotics industry may be the decisive factor in the global competition over artificial intelligence. The piece suggests that combining AI with large-scale robot manufacturing could give China an edge beyond software alone. Observers are weighing how physical automation, not just algorithms, could shape which country leads the next phase of AI development.

Retail brands are increasing investment in physical stores even as artificial intelligence reshapes the industry, according to Modern Retail's latest Brands Briefing. Rather than replacing shops, AI appears to be pushing brands to rethink physical retail as part of a blended strategy, combining digital efficiency with in-store experience to keep customers engaged.

CEATEC 2026 is drawing attention in Japan, with one notable announcement from Highlanders, which says it will develop a domestic physical AI foundation model aimed at the alignment process in automobile manufacturing. The technology is intended to improve precision in vehicle production lines. Interest in Japanese-made physical AI and industrial robotics is driving discussion around this year's show.

Cisco has introduced DefenseClaw, a framework aimed at extending AI security governance beyond software to physical AI systems such as robots and edge devices. The proposal, outlined in a Cisco blog post, argues that agentic and embodied AI operating in the physical world needs dedicated oversight and defense mechanisms, sparking discussion among security practitioners about how to govern AI agents at the network edge.

New comparisons show AI-based weather forecasting models perform well at predicting light rain, but traditional physics-based models remain more reliable when heavy rainfall and flooding are at stake. The finding matters for emergency planning, as floods cause some of the deadliest weather damage. Forecasters say the two approaches may need to be used side by side rather than one replacing the other.

Google's Gemini 4 Argon AI model was used to teach a humanoid robot to play cornhole, the lawn game popular at US backyard gatherings. The demonstration shows the model handling precise aiming, throwing and real-world coordination tasks, and is being reported by tech outlets as another sign of AI moving from language into physical robotics.