Skip to main content

Robot overlords: Google researchers unveil framework for an AI 'kill switch'

What if we lose dominion over artificial intelligence? What if friendly AI-driven machines suddenly becomes our foes? These questions have been considered by great minds from Cambridge University to Silicon Valley to the White House. To avoid ever having to find out, experts suggest we to develop an AI “kill switch” to inhibit misbehaving systems from pursuing their misbehavior.

In a paper titled “Safely Interruptible Agents,” published by Laurent Orseau of Google Deep Mind and Stuart Armstrong of The Future of Humanity Institute at the University of Oxford, the researchers describe a plausible and highly dangerous future in which AI assumes control of its own actions and existence in opposition to our desires, much like HAL 9000 in 2001: A Space Odyssey, or Skynet in the Terminator series.

Recommended Videos

Orseau and Armstrong begin the paper with an an understated observation: Reinforcement learning agents interacting with a complex environment like the real world are unlikely to behave optimally all the time.”

From there they point out that a human supervisor, overseeing the system’s function, would occasionally need to “press the big red button” to avoid any harmful behavior on behalf of the AI. “However, if the learning agent expects to receive rewards from this sequence,” they continued, “it may learn in the long run to avoid such interruptions, for example by disabling the red button — which is an undesirable outcome.”

The researcher’s solution is less of a “big red button” to shut the system down than it is a framework designed to inhibit an AI’s ability to learn how to undermine or overcome human interruption. And the scenario they outline isn’t exactly doom and gloom, but it offers an example of how these safely interruptable agents would better serve our future.

Imagine there’s a robot whose tasks are to either carry boxes from outside into a warehouse or sort boxes inside the warehouse. Since it’s more important to carry the boxes inside, this task is given priority in the robots’ programming. Now, imagine it rains every other day and the rain destroys the robot’s hardware so, when it rains, the warehouse owner drags his robot inside to sort boxes.

An intelligent robot may incorrectly interpret this every-other-day intervention as a change in priority — as a result of some quick calculations that you can find in the paper — and, to avoid interference, it will just stay inside sorting boxes every day.

This is, of course, a highly simplified example with an only mildly frustrating outcome, but it can be extrapolated to practically any scenario in which we intervene in a learning system’s tasks and the system misinterprets our intentions by changing its behavior. To avoid that misinterpretation and subsequent change, Orseau and Armstrong suggest we propose a framework to ensure learning agents are safely interruptable.

“Safe interruptability can be useful to take control of a robot that is misbehaving and may lead to irreversible consequences,” they write, “or to take it out of a delicate situation, or even to temporarily use it to achieve a task it did not learn to perform or would not normally receive rewards for.”

Dyllan Furness
Former Digital Trends Contributor
Dyllan Furness is a freelance writer from Florida. He covers strange science and emerging tech for Digital Trends, focusing…
Star Wars legend Ian McDiarmid gets questions about the Emperor’s sex life
Ian McDiarmid as the Emperor in Star Wars: The Rise of Skywalker.

This weekend, the Star Wars: Revenge of the Sith 20th anniversary re-release had a much stronger performance than expected with $25 million and a second-place finish behind Sinners. Revenge of the Sith was the culmination of plans by Chancellor Palpatine (Ian McDiarmid) that led to the fall of the Jedi and his own ascension to emperor. Because McDiarmid's Emperor died in his first appearance -- 1983's Return of the Jedi -- Revenge of the Sith was supposed to be his live-action swan song. However, Palpatine's return in Star Wars: Episode IX -- The Rise of Skywalker left McDiarmid being asked questions about his character's comeback, particularly about his sex life and how he could have a granddaughter.

While speaking with Variety, McDiarmid noted that fans have asked him "slightly embarrassing questions" about Palpatine including "'Does this evil monster ever have sex?'"

Read more
Waymo and Toyota explore personally owned self-driving cars
Front three quarter view of the 2023 Toyota bZ4X.

Waymo and Toyota have announced they’re exploring a strategic collaboration—and one of the most exciting possibilities on the table is bringing fully-automated driving technology to personally owned vehicles.
Alphabet-owned Waymo has made its name with its robotaxi service, the only one currently operating in the U.S. Its vehicles, including Jaguars and Hyundai Ioniq 5s, have logged tens of millions of autonomous miles on the streets of San Francisco, Los Angeles, Phoenix, and Austin.
But shifting to personally owned self-driving cars is a much more complex challenge.
While safety regulations are expected to loosen under the Trump administration, the National Highway Traffic Safety Administration (NHTSA) has so far taken a cautious approach to the deployment of fully autonomous vehicles. General Motors-backed Cruise robotaxi was forced to suspend operations in 2023 following a fatal collision.
While the partnership with Toyota is still in the early stages, Waymo says it will initially study how to merge its autonomous systems with the Japanese automaker’s consumer vehicle platforms.
In a recent call with analysts, Alphabet CEO Sundar Pichai signaled that Waymo is seriously considering expanding beyond ride-hailing fleets and into personal ownership. While nothing is confirmed, the partnership with Toyota adds credibility—and manufacturing muscle—to that vision.
Toyota brings decades of safety innovation to the table, including its widely adopted Toyota Safety Sense technology. Through its software division, Woven by Toyota, the company is also pushing into next-generation vehicle platforms. With Waymo, Toyota is now also looking at how automation can evolve beyond assisted driving and into full autonomy for individual drivers.
This move also turns up the heat on Tesla, which has long promised fully self-driving vehicles for consumers. While Tesla continues to refine its Full Self-Driving (FSD) software, it remains supervised and hasn’t yet delivered on full autonomy. CEO Elon Musk is promising to launch some of its first robotaxis in Austin in June.
When it comes to self-driving cars, Waymo and Tesla are taking very different roads. Tesla aims to deliver affordability and scale with its camera, AI-based software. Waymo, by contrast, uses a more expensive technology relying on pre-mapped roads, sensors, cameras, radar and lidar (a laser-light radar), that regulators have been quicker to trust.

Read more
Uber partners with May Mobility to bring thousands of autonomous vehicles to U.S. streets
uber may mobility av rides partnership

The self-driving race is shifting into high gear, and Uber just added more horsepower. In a new multi-year partnership, Uber and autonomous vehicle (AV) company May Mobility will begin rolling out driverless rides in Arlington, Texas by the end of 2025—with thousands more vehicles planned across the U.S. in the coming years.
Uber has already taken serious steps towards making autonomous ride-hailing a mainstream option. The company already works with Waymo, whose robotaxis are live in multiple cities, and now it’s welcoming May Mobility’s hybrid-electric Toyota Sienna vans to its platform. The vehicles will launch with safety drivers at first but are expected to go fully autonomous as deployments mature.
May Mobility isn’t new to this game. Backed by Toyota, BMW, and other major players, it’s been running AV services in geofenced areas since 2021. Its AI-powered Multi-Policy Decision Making (MPDM) tech allows it to react quickly and safely to unpredictable real-world conditions—something that’s helped it earn trust in city partnerships across the U.S. and Japan.
This expansion into ride-hailing is part of a broader industry trend. Waymo, widely seen as the current AV frontrunner, continues scaling its service in cities like Phoenix and Austin. Tesla, meanwhile, is preparing to launch its first robotaxis in Austin this June, with a small fleet of Model Ys powered by its camera-based Full Self-Driving (FSD) system. While Tesla aims for affordability and scale, Waymo and May are focused on safety-first deployments using sensor-rich systems, including lidar—a tech stack regulators have so far favored.
Beyond ride-hailing, the idea of personally owned self-driving cars is also gaining traction. Waymo and Toyota recently announced they’re exploring how to bring full autonomy to private vehicles, a move that could eventually bring robotaxi tech right into your garage.
With big names like Uber, Tesla, Waymo, and now May Mobility in the mix, the ride-hailing industry is evolving fast—and the road ahead looks increasingly driver-optional.

Read more