AI robots want office jobs

A desk robot with blinking eyes and a projector arm wants a seat in your office.For years, generative AI lived mostly inside chat windows, browser tabs and smartphone apps. Now the industry is racing to push those systems into the physical world, turning AI from something people chat with onscreen into something that can see, listen, respond and interact in real time.Two recent announcements underscored the push. Earlier this week, startup Thinking Machines Lab, founded by former OpenAI CTO Mira Murati,unveiledresearch on AI systems that process audio, video and text simultaneously. Meanwhile, Lenovo and Hugging Face are testing devices that see, project and talk back at your desk.“It’s genuinely a fundamental change,”Daniel Wigdor, a professor of computer science at the University of Toronto, toldIBM Thinkin an interview. “For decades, the dominant model in computing has been what we call WIMP: windows, icons, menus and a pointer.”Now, he argued, AI systems are beginning to move beyond screens entirely, turning offices, desks and physical environments into the next computing interface.

Lenovo’sprototypeis a desktop assistant with a small screen that serves as a face, cameras mounted on an articulating arm and a built-in projector that displays slides or documents on nearby surfaces. The company says the system could connect to office calendars, messages and files, allowing workers to ask questions about projects without opening another laptop window.A broader movement away from traditional screen-based computing interfaces may be driving interest in systems like Lenovo’s AI Workmate, according to Wigdor.“When a device like Lenovo’s AI Workmate can see your desk, hear what’s happening in the room and project information directly into your physical space, the interface stops being something you visit and starts being something you’re inside of.”Wigdor and other researchers use the term “embodied AI” to describe systems that interact with the physical world through sensors, microphones, cameras, robotics or other hardware. Wigdor said those systems could change not only how people use AI, but also how they relate to it psychologically.After several years of investment in generative AI workplace tools for writing, coding and productivity tasks, companies including Lenovo and Hugging Face are beginning to test AI systems designed for physical environments.The push also extends beyond hardware. Thinking Machines Lab’s Muratiintroducedresearch on what it calls “interaction models”: AI systems designed to process audio, video and text simultaneously, and respond in real time. The company said current chatbot interfaces still rely on turn-based exchanges where users type or speak, wait for a response and then repeat the process. By contrast, the company envisions AI systems that can continuously interpret what users are seeing, saying and doing without requiring a stop-and-start chatbot exchange.Hugging Face, the open-source AI platform best known for hostingmachine learning models, recently expanded into robotics with a desktop companion calledReachy Mini. Designed for AI experimentation and conversational interaction, the device includes cameras, microphones, speakers and a motorized head that can respond to users in real time.Even as companies pitch AI devices for workplaces, office environments may prove especially difficult terrain for robotic systems according toGabe Goodhart, Chief Architect for AI Open Innovation at IBM.“The biggest challenge with embodied AI will be designing a platform that has a real reason to be embodied,” Goodhart toldIBM Thinkin an interview. “Successful embodied AI will need to figure out how to manifest this in the real world.”Rather than changing what work gets done, Wigdor suggested embodied AI could change how people interact with information inside physical spaces.“The reason this matters is that people are actually very good at working in physical space,” he said. “We read context from our environment, and we coordinate with objects and other people through gesture and proximity.”

The same hardware that makes these systems useful could also makes them intrusive.Wigdor sees surveillance and trust as the next pressure points.“The privacy question is the one that will make or break this category,” he said. “An embodied system with cameras, microphones and environmental sensors knows what you’re doing, who you’re talking to, how long you spent at your desk and what you were looking at before you picked up the phone.”The privacy concerns extend beyond the individual user sitting at the desk.“The concern I’d put at the top is what I’d call the problem of involuntary participation,” Wigdor said. “When you share an office with an AI system that is persistently sensing its environment, your colleagues, your visitors and anyone who walks through that space are also being observed.”Longstanding debates around robotics and surveillance have shaped how some researchers think about embodied AI systems.Heather Knight, a robotics researcher and former professor of robotics and computer science at Oregon State University, said debates around robotics often center more on human misuse than on the machines themselves.“It’s not the robots we have to worry about, it’s the people who misuse them,” Knight toldIBM Thinkin an interview.Alternatives to traditional cameras and microphones could potentially reduce privacy risks, Knight said. “Tracking a warm body with pyroelectrics does not identify you in the same way as voice or image,” she said, referring to motion-sensing technology often used in automatic outdoor lights.Companies alreadyuse softwarethat tracks keystrokes, browser activity, messaging patterns or meeting participation. Wigdor stressed that embodied AI could intensify those concerns by turning physical space into another source of workplace analytics.“If workers suspect the data their desk companion collects is flowing to management dashboards—and often they’ll be right—the psychological dynamic in that workspace changes permanently,” Wigdor said. “People become performative.”As embodied AI systems become more prevalent inside workplaces,governanceissues may also become more urgent. “Organizations need to decide what kind of authority these systems have, and write it down,” Wigdor said. “An embodied AI that can interrupt a conversation, surface a document or route a task is taking actions with real consequences.”Not everyone is eager to share the office with a conversational robot. “I can’t see anyone wanting a googly-eyed chatty robot pal at work,” Goodhart said. “Though I could definitely see this for education or kids’ fun.”The technology may be advancing quickly. Convincing workers to live alongside it could take longer.