Top scientist wants to prevent AI from going rogue
Concerned about the rapid spread of generative AI, a pioneer researcher is developing software to keep tabs on a technology that is increasingly taking over human tasks. – AFP photo
NEW YORK (June 8): Concerned about the rapid spread of generative AI, a pioneer researcher is developing software to keep tabs on a technology that is increasingly taking over human tasks.
Canadian computer science professor Yoshua Bengio is considered one of the godfathers of the artificial intelligence revolution and recently announced the launch of LawZero, a non-profit organisation intended to mitigate the technology's inherent risks.
The winner of the Turing Award, also known as the Nobel Prize for computer science, has been warning for several years of the risks of AI, whether through its malicious use or the software itself going awry.
Those risks are increasing with the development of so-called AI agents, a use of the technology that tasks computers with making decisions that were once made by human workers.
The goal of these agents is to build virtual employees that can do practically any job a human can, at a fraction of the cost.
'Currently, AI is developed to maximise profit,' Bengio said, adding it was being deployed even as it persists to show flaws.
Moreover, for Bengio, giving AI human-like agency will easily be used for malicious purposes such as disinformation, bioweapons, and cyberattacks.
'If we lose control of rogue super-intelligent AIs, they could greatly harm humanity,' he said.
One of the first objectives at LawZero will be to develop Scientist AI, a form of specially trained AI that can be used as a guardrail to ensure other AIs are behaving properly, the company said.
The organisation already has over 15 researchers and has received funding from Schmidt Sciences, a charity set up by former Google boss Eric Schmidt and his wife Wendy.
The project comes as powerful large language models (or LLMs) from OpenAI, Google and Anthropic are deployed across all sectors of the digital economy, while still showing significant problems.
These include AI models that show a capability to deceive and fabricate false information even as they increase productivity.
In a recent example, AI company Anthropic said that during safety testing, its latest AI model tried to blackmail an engineer to avoid being replaced by another system. artificial intelligence LawZero Yoshua Bengio
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Sinar Daily
6 hours ago
- Sinar Daily
Time machine: How carbon dating brings the past back to life
It uses a technique called carbon dating, which has "revolutionised archaeology", winning its discoverer a Nobel Prize in 1960, French scientist Lucile Beck said. 08 Jun 2025 04:04pm A laboratory technician presses graphite samples in an aluminum cathode to prevent contamination before the measurement process at the LMC14 Laboratory (Laboratoire de Mesure du Carbone 14) at the French Alternative Energies and Atomic Energy Commission (CEA: Commissariat a l'Energie Atomique et aux Energies Alternatives) in Saclay, southwest Paris, on May 22, 2025. (Photo by Thomas SAMSON / AFP) SACLAY - From unmasking art forgery to uncovering the secrets of the Notre-Dame cathedral, an imposing machine outside Paris can turn back the clock to reveal the truth. It uses a technique called carbon dating, which has "revolutionised archaeology", winning its discoverer a Nobel Prize in 1960, French scientist Lucile Beck said. Carbon dating can turn back the clock to reveal the truth. By comparing the number of carbon particles separated by the particle accelerator, scientists can get an estimate of how old something is. This all makes it possible to spot a forged painting by demonstrating that the linen used in the canvas was harvested well after when the purported painter died. Carbon dating can be used to date materials that do not directly derive only from living organisms. (Photo by Thomas SAMSON / AFP) She spoke to AFP in front of the huge particle accelerator, which takes up an entire room in the carbon dating lab of France's Atomic Energy Commission in Saclay, outside the capital. Beck described the "surprise and disbelief" among prehistorians in the 1990s when the machine revealed that cave art in the Chauvet Cave in France's southeast was 36,000 years old. The laboratory uses carbon dating, also called carbon-14, to figure out the timeline of more than 3,000 samples a year. So how does it work? First, each sample is examined for any trace of contamination. "Typically, they are fibres from a jumper" of the archaeologist who first handled the object, Beck said. The sample is then cleaned in an acid bath and heated to 800 degrees Celsius (1,472 Fahrenheit) to recover its carbon dioxide. This gas is then reduced to graphite and inserted into tiny capsules. Next, these capsules are put into the particle accelerator, which separates their carbon isotopes. Isotopes are variants of the same chemical element which have different numbers of neutrons. Some isotopes are stable, such as carbon-12. Others -- such as carbon-14 -- are radioactive and decay over time. Carbon-14 is constantly being created in Earth's upper atmosphere as cosmic rays and solar radiation bombard the chemical nitrogen. In the atmosphere, this creates carbon dioxide, which is absorbed by plants during photosynthesis. Then animals such as ourselves get in on the act by eating those plants. So all living organisms contain carbon-14, and when they die, it starts decaying. Only half of it remains after 5,730 years. After 50,000 years, nothing is left -- making this the limit on how far back carbon dating can probe. By comparing the number of carbon-12 and carbon-14 particles separated by the particle accelerator, scientists can get an estimate of how old something is. Cosmic radiation is not constant, nor is the intensity of the magnetic field around Earth protecting us from it, Beck said. That means scientists have to make estimations based on calculations using samples whose ages are definitively known. This all makes it possible to spot a forged painting, for example, by demonstrating that the linen used in the canvas was harvested well after when the purported painter died. The technique can also establish the changes in our planet's climate over the millennia by analysing the skeletons of plankton found at the bottom of the ocean. Notre-Dame revealed Carbon dating can be used on bones, wood and more, but the French lab has developed new methods allowing them to date materials that do not directly derive from living organisms. For example, they can date the carbon that was trapped in iron from when its ore was first heated by charcoal. After Paris's famous Notre-Dame cathedral almost burned to the ground in 2019, this method revealed that its big iron staples dated back to when it was first built -- and not to a later restoration, as had been thought. The technique can also analyse the pigment lead white, which has been painted on buildings and used in artworks across the world since the fourth century BC. To make this pigment, "lead was corroded with vinegar and horse poo, which produces carbon dioxide through fermentation," Beck explained. She said she always tells archaeologists: "don't clean traces of corrosion, they also tell about the past!" Another trick made it possible to date the tombs of a medieval abbey in which only small lead bottles had been found. As the bodies in the tombs decomposed, they released carbon dioxide, corroding the bottles and giving scientists the clue they needed. "This corrosion was ultimately the only remaining evidence of the spirit of the monks," Beck mused. - AFP More Like This
The Star
11 hours ago
- The Star
Is China's Tiandu-1 first to enter resonant Earth-moon orbit? US raises doubts
China's Tiandu-1 satellite has entered a critical fuel-efficient orbit that synchronises with the motion of the Earth and moon, but claims that it is the first spacecraft to achieve the feat have been challenged by US and Canadian experts. The 3:1 resonant orbit – in which Tiandu-1 completes three loops of the Earth for every circuit made by the moon – is seen as a breakthrough for lunar infrastructure, including a BeiDou-like guidance system that will be needed for interplanetary exploration. Developer Deep Space Exploration Laboratory said that 'the Tiandu-1 navigation test satellite successfully carried out a key manoeuvre at perigee [the point in orbit when it is closest to Earth] on May 22 and precisely transitioned into a 3:1 Earth-moon resonant orbit'. 'This made Tiandu-1 the first spacecraft to enter a round-trip 3:1 Earth-moon resonant trajectory,' according to a statement on the website of the laboratory based in Hefei, Anhui province in central China. 'Its flight data will provide support for advancing technologies such as orbit maintenance, control, and autonomous navigation in complex gravitational environments.' Experts in the US and Canada immediately challenged the claim, pointing to Nasa's 15-year-old Interstellar Boundary Explorer (IBEX) probe that entered a near-identical orbit in 2011, where it continues to operate. Jonathan McDowell, a Harvard astronomer and space historian, and Canada-based amateur stargazer Scott Tilley both said it was debatable whether Tiandu-1 could lay claim to the title, with the IBEX craft's achievement of near 3:1 resonance. 'Yes, indeed Tiandu-1 has entered a 3:1 resonance orbit. Whether it's the first is launched in 2008 uses a similar orbit,' Tilley said. According to McDowell, the advantage of resonant orbits is that they allow a spacecraft to operate far from Earth while avoiding the unpredictable, chaotic motion caused by frequent lunar fly-bys. 'The resonance provides stability,' he said. In a 3:1 resonance, the 61kg (135lb) Tiandu-1 – which launched alongside the Queqiao-2 lunar relay satellite last year – completes each petal of its three-lobed orbit around the Earth roughly every nine days – the same amount of time it takes the moon to complete one. In April, state news agency Xinhua reported that China's DRO-B satellite had departed lunar orbit, after helping to establish the 'world's first three-satellite constellation in cislunar space', and entered an Earth-moon resonant orbit. DRO-B, a 277kg Chinese satellite that was salvaged after being stranded in the wrong orbit due to a launch mishap last year, is currently in a 3:2 Earth-moon resonant orbit, according to Tilley, who is best known for helping Nasa to find its long-lost IMAGE satellite in 2018. The Canadian amateur astronomer, who tracks satellites in his spare time, said he noticed that DRO-B had vanished from its known lunar orbit more than a month ago and began searching for it. Using graphics and research papers, Tilley modelled a 3:2 resonance orbit – one that tours the gravitational balance points between Earth and the moon known as L3, L4, and L5. After an exhaustive search, he spotted DRO-B in such an orbit. DRO-B completes three circles around Earth in the time it takes the moon to complete two. The satellite traces a broad, triangular loop with lobes near the L3, L4, and L5 points, which it swings past roughly every 18 days. 'DRO-B is not transmitting all the time like it used to, and it's likely [to be] having minor power issues as its solar panel supports were damaged during the launch mishap,' Tilley said. He added that DRO-B's orbit was especially notable because this specific class of 3:2 resonance orbit had never been used before. 'Japan's Hiten lunar probe did visit some of these points, but didn't use a 3:2 resonance orbit.' According to Tilley, China's use of these orbits seemed to be all about testing navigation-related technologies. 'Having a system that covers the entire Earth-moon system for orbital determination and positioning would be very helpful,' he said. - SOUTH CHINA MORNING POST
Malaysian Reserve
14 hours ago
- Malaysian Reserve
Everything Apple plans to show at its iOS 26-focused WWDC 2025 event
APPLE Inc.'s upcoming Worldwide Developers Conference will do little to assuage fears that the iPhone maker is a laggard in AI. Instead, the event will focus on design and productivity enhancements for its long-established operating system franchises. The company's keynote address, which begins at 10 a.m. Pacific time on Monday, will introduce redesigned software interfaces for the iPhone, iPad, Mac, Apple TV and Apple Watch, in addition to more minor tweaks to the Vision Pro headset. As part of the end-to-end overhaul, the company is also making a sweeping change to its software branding, which will shift from version numbers to a year-based system. That means Apple will introduce iOS 26, iPadOS 26, tvOS 26, visionOS 26, macOS 26 and watchOS 26 — named for 2026. Internally, the operating systems are known as Luck, Charisma, Discovery, Cheer and Nepali, respectively. The Mac operating system will still retain its California landmark theme, with this year's release known as macOS Tahoe. While the design changes will make up a notable portion of the keynote, the company will also discuss its Apple Intelligence AI strategy. On that front, Apple will let third-party developers begin tapping into its large language models — the underpinnings of generative artificial intelligence. The company also is introducing iPad enhancements that will make the device better suited for office work and unveiling significant new features for the Mac. The AI changes will be surprisingly minor and are unlikely to impress industry watchers, especially considering the rapid pace of innovation by Alphabet Inc.'s Google, Meta Platforms Inc., Microsoft Corp. and OpenAI. Apple was already slow to launch its AI platform last year and indefinitely delayed a revamp of the Siri voice assistant. In a symptom of its struggles, it recently shook up management of its AI operation by removing the Siri and robotics teams from the command of its artificial intelligence chief. Google last month, meanwhile, introduced more powerful models and additional AI search features at its developers conference last month. It also unveiled new AI subscription services and a way to make realistic videos with simple prompts. Apple is facing potential competition from another source as well. Jony Ive, the company's former design chief who helped create the original iMac, iPod, iPhone, iPad and Apple Watch, has teamed up with OpenAI's Sam Altman to develop new hardware devices. Samsung Electronics Co. is also moving deeper into AI, planning an extensive partnership with the startup Perplexity. Here is everything to expect from Apple's WWDC event, including details that haven't been reported until now: Operating system design and features: The standout announcement will be a brand-new interface for all of Apple's operating systems, including CarPlay. The new look — code-named Solarium internally — is based on visionOS, the software on the Vision Pro headset. The main interface element will be digital glass. In a nod to the code name, which won't be used externally, there will be more use of light and transparency throughout the operating systems. Tool and tab bars will look different, and there will be redesigned app icons and other buttons. There's also a strong focus on the use of pop-out menus, meaning users can click a button to get a quick list of additional options. On the Mac, the menu bar and window buttons will also get fresh designs. While there has been speculation that the app icons will be round to match the style on the Apple Watch and Vision Pro, the shape is staying largely the same on the iPhone and iPad. Widgets for the iPhone, iPad and Mac — customizable views on the home screen for weather data, stock tickers and appointments — have been redesigned to match the new interface. But they largely function the same way as they do today and aren't a bigger piece of the user experience. Though most of the core apps on iOS and iPadOS won't get overhauls beyond the design changes, three of them are due for significant revamps: the Phone, Safari and Camera apps. The Phone app hasn't changed much since the original version of what was then called iPhone OS in 2007. For iOS 26, Apple is introducing a new view that combines favorite contacts, recent calls and voicemails into a single, scrollable window. This modernized approach is optional, and users can toggle to the legacy interface with a button in the Phone app itself. Safari is getting a significantly updated look, highlighted by a more transparent and glassy address bar. The Camera app will be revamped with a focus on simplicity. Apple has added several new photo and video-taking options in recent years — including spatial video, panorama and slow-motion recording — and that's made today's interface a bit clunky. In iOS 26 and iPadOS 26, Apple is rethinking the approach. Messages is getting a notable upgrade as well, with a focus on going after Meta's WhatsApp and other modern messaging apps. The two main changes are the ability to create polls and set a background image. The backgrounds will sync between devices, including those of other users, meaning that you and the people you are chatting with have the same look. Apple is bringing its longstanding Preview app from macOS to iPadOS and iOS for the first time, giving iPad and iPhone users an in-house solution to PDF management, annotation and editing. The software looks similar to the Mac version, and its launch screen is in the same style as apps like Pages and Keynote. It includes a big logo on the top portion of the screen advertising the Preview name and a gallery of document options below it. The app will be preinstalled rather than launch via the App Store. Another preinstalled app is Games. It puts game downloading and access to Apple's Arcade platform in one place, looking like a games-centric version of the App Store. The app has five tabs: Home, Arcade, Play Together, Library and Search. On the heels of the Nintendo Switch 2 launch, Apple is hopeful that the new app can make its mobile devices a bigger part of the gaming industry. But this new app is unlikely to do the trick and is fairly underwhelming. The Apple Vision Pro will get support for eye scrolling — a new approach for moving up and down through documents and web pages, as well as within homegrown and third-party apps. Apple is also preparing to give so-called magic wand support to the Vision Pro. That means customers will be able to use third-party hand controllers, such as the ones sold by Sony Group Corp. for the PlayStation VR console. These hand controllers are distinct from standard game controllers and can be used to mimic specific hand movements. They're ideal for gaming and creating a virtual wand for navigating visionOS. Apple is completely revamping multitasking features on the iPad to be more Mac-like. The new functionality may require a user to be hooked up to a keyboard and trackpad, such as the company's Magic Keyboard. The Apple Pencil is getting an upgrade to include a digital reed calligraphy pen. The keyboard on iOS and iPadOS, meanwhile, will include a bidirectional mode to move between Arabic and English. The company is planning a feature to sync captive wireless network details across devices. This means if you log in at a gym, hotel or office building on one device, it will automatically log you in on other devices. New AI-related features: The most notable AI upgrade — that consumers can see — will be a systemwide push into translation. The company launched its Translate app a few years ago, and now the functionality is getting integrated across its operating systems as an Apple Intelligence feature. The main use will be live translation of phone calls and text messages. At the same time, the company is also preparing translations of live conversations for AirPods wearers. Google has offered many of these features on Android for several years. The biggest AI-related announcement for developers will be the company opening up its foundation models, the large language models that power Apple Intelligence, for the first time. This will let outside app creators build their own AI features using the same technology Apple relied on to make Writing Tools, Genmoji custom emoji and its summarization technology. Genmoji, one of the more popular Apple Intelligence features, is getting a small upgrade. For the first time, users will be able to create a Genmoji by combining a pair of existing standard emoji (for instance, turning a basketball and a trash can into a basketball going into the can). The company will introduce an upgraded Shortcuts app, its software that allows users to create quick shortcuts for actions across its operating systems. It will use the Apple Intelligence models. Apple has also developed an upgraded version of its foundation models that it plans to announce for both on-device and cloud use. Developers will have access to the on-device version. Apple is preparing a battery optimization mode that uses AI to save power on its iPhones. This functionality may come later, though, as it was developed in concert with the company's slimmer iPhone 17 model, which isn't slated for release until later this year. The technology is well-suited to that device since the phone is going to have a smaller battery. Apple has been working to add Google's Gemini software as an alternative to OpenAI's ChatGPT, which works with Siri and the Writing Tools. Though Alphabet Chief Executive Officer Sundar Pichai hinted that an accord was imminent, Apple has no current plan to announce such integration at WWDC (there likely won't be any public movement on this front until the US Justice Department makes its ruling on Google's search deal with Apple). The company's delayed Siri upgrade will eventually offer the ability to tap into personal data and on-device content, as well as more precisely control applications and in-app features. It's still far out, and the company isn't planning to showcase any significant new Siri features at this year's event. A further revamped voice assistant, dubbed LLM Siri internally, is still probably a year or two away — at minimum — from being introduced. Apple is working on a revamped Calendar app for release across its platforms. It originally planned to introduce the software this year, but it's been delayed and is now slated for the subsequent set of operating systems. Likely to be dubbed iOS 27 and macOS 27, next year's software is already in development. Those upgrades are internally dubbed Buttercup and Honeycrisp, respectively. Apple has also been working on an end-to-end revamp of its Health app tied to an AI doctor-based service code-named Mulberry. Neither will be shown at WWDC and, due to delays, likely won't be released at full scale until the end of next year at the earliest, as part of Buttercup. There will be smaller changes this year. Last year, the company announced Swift Assist, a feature for Xcode that could use Apple Intelligence to complete lines of code. It never launched because of hallucinations — a problem where AI makes up information — and other snags. The solution: a new version of Xcode that taps into third-party LLMs, either remotely or stored locally on the Mac. Apple is already using this internally with Claude from startup Anthropic. –BLOOMBERG
