‘Godfather of AI' believes it's unsafe - but here's how he plans to fix the tech
This week the US Federal Bureau of Investigation revealed two men suspected of bombing a fertility clinic in California last month allegedly used artificial intelligence (AI) to obtain bomb-making instructions.
The FBI did not disclose the name of the AI program in question.
This brings into sharp focus the urgent need to make AI safer. Currently we are living in the 'wild west' era of AI, where companies are fiercely competing to develop the fastest and most entertaining AI systems.
Each company wants to outdo competitors and claim the top spot. This intense competition often leads to intentional or unintentional shortcuts – especially when it comes to safety.
Coincidentally, at around the same time of the FBI's revelation, one of the godfathers of modern AI, Canadian computer science professor Yoshua Bengio, launched a new nonprofit organisation dedicated to developing a new AI model specifically designed to be safer than other AI models – and target those that cause social harm.
So what is Bengio's new AI model? And will it actually protect the world from AI-faciliated harm?
In 2018, Bengio, alongside his colleagues Yann LeCun and Geoffrey Hinton, won the Turing Award for groundbreaking research they had published three years earlier on deep learning. A branch of machine learning, deep learning attempts to mimic the processes of the human brain by using artificial neural networks to learn from computational data and make predictions.
Bengio's new nonprofit organisation, LawZero, is developing 'Scientist AI'. Bengio has said this model will be 'honest and not deceptive', and incorporate safety-by-design principles.
According to a preprint paper released online earlier this year, Scientist AI will differ from current AI systems in two key ways.
First, it can assess and communicate its confidence level in its answers, helping to reduce the problem of AI giving overly confident and incorrect responses.
Second, it can explain its reasoning to humans, allowing its conclusions to be evaluated and tested for accuracy.
Interestingly, older AI systems had this feature. But in the rush for speed and new approaches, many modern AI models can't explain their decisions. Their developers have sacrificed explainability for speed.
Bengio also intends 'Scientist AI' to act as a guardrail against unsafe AI. It could monitor other, less reliable and harmful AI systems — essentially fighting fire with fire.
This may be the only viable solution to improve AI safety. Humans cannot properly monitor systems such as ChatGPT, which handle over a billion queries daily. Only another AI can manage this scale.
Using an AI system against other AI systems is not just a sci-fi concept – it's a common practice in research to compare and test different level of intelligence in AI systems.
Large language models and machine learning are just small parts of today's AI landscape.
Another key addition Bengio's team are adding to Scientist AI is the 'world model' which brings certainty and explainability. Just as humans make decisions based on their understanding of the world, AI needs a similar model to function effectively.
The absence of a world model in current AI models is clear.
One well-known example is the 'hand problem': most of today's AI models can imitate the appearance of hands but cannot replicate natural hand movements, because they lack an understanding of the physics — a world model — behind them.
Another example is how models such as ChatGPT struggle with chess, failing to win and even making illegal moves.
This is despite simpler AI systems, which do contain a model of the 'world' of chess, beating even the best human players.
These issues stem from the lack of a foundational world model in these systems, which are not inherently designed to model the dynamics of the real world.
Bengio is on the right track, aiming to build safer, more trustworthy AI by combining large language models with other AI technologies.
However, his journey isn't going to be easy. LawZero's US$30 million in funding is small compared to efforts such as the US$500 billion project announced by US President Donald Trump earlier this year to accelerate the development of AI.
Making LawZero's task harder is the fact that Scientist AI – like any other AI project – needs huge amounts of data to be powerful, and most data are controlled by major tech companies.
There's also an outstanding question. Even if Bengio can build an AI system that does everything he says it can, how is it going to be able to control other systems that might be causing harm?
Still, this project, with talented researchers behind it, could spark a movement toward a future where AI truly helps humans thrive. If successful, it could set new expectations for safe AI, motivating researchers, developers, and policymakers to prioritise safety.
Perhaps if we had taken similar action when social media first emerged, we would have a safer online environment for young people's mental health. And maybe, if Scientist AI had already been in place, it could have prevented people with harmful intentions from accessing dangerous information with the help of AI systems.
Armin Chitizadeh is a Lecturer in the School of Computer Science at the University of Sydney. This article is republished from The Conversation under a Creative Commons license. Read the original article
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
WIRED
33 minutes ago
- WIRED
How to Convert an Analog Bike to an Electric Bike
Michael Venutolo-Mantovani You can save yourself thousands of dollars on an electric bike by upgrading your current ride. Let us walk (or cycle!) you through it. Courtesy of Science Photo Library via Getty Images All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links. Say you want to get a bicycle up and down hills with a minimal amount of pedal power. What do you do? The first option is simple: buy an electric bike. However, ebikes aren't cheap. These days, the least expensive but still reliable ebike you can buy is Aventon Soltera 2.5, which costs around $1,200. You can rent an ebike, or find a city bike program that uses them, or, if you're really lucky, find an ebike in a lending library. Or you can just make your regular bike into an electric bike. That is, you can slap an aftermarket unit on your traditional bike, creating a motor-assisted ebike with little more than your bicycle and any of a variety of aftermarket power sources. What used to be a very niche field with only a small handful of offerings has become a cottage industry full of contenders and pretenders, and any number of startups offering a readymade conversion for your bike. So how do you turn your traditional bike into an ebike? And what are some of the best offerings out there? We'll walk you through it. If you don't see anything you like, make sure you check out our guides to the Best Electric Bikes or the Best Electric Scooters. Hot to Throt Well, first, you need to decide what kind of conversion you want. There are two basic kinds of ebikes, pedal-assist and throttle, with many newer models boasting some combination of both. Pedal assist is a mechanism that provides an added boost of electric power when you're pedaling. Via the use of cadence and/or torque sensors (how fast your pedals are rotating and the power that's being applied to them, respectively), pedal-assist units kick on at a certain point, making pedaling easier. Sometimes pedal assist can be so minimal, it's difficult to tell just how much work you're doing versus how much of the load your bike is handling. Throttle units act more like a motorcycle or moped, allowing the rider to either twist a tube mounted near the handgrip or use a thumb-based knob to deliver immediate power to the wheels. Throttle-based ebikes can be ridden without pedaling. When it comes to ebike conversions, where your power comes from and how it's delivered can look several different ways. Some mechanisms deliver power to either your bicycle's rear or front hub. Others are situated in the bottom bracket (that is, where the cranks meet the bike frame). And some newer conversion kits act something like a treadmill mounted to your bike, using rollers to help propel your wheel forward. Roll the Clip Photograph: Stephanie Pearson Having said all that, it's important to consider your level of bicycle know-how when it comes to what kind of ebike conversion you want to undertake. If you're a wrencher, the conversion process is generally straightforward, regardless of the type of unit. If you can change a tire or swap out your cassette, you'll probably be fine with a bit of trial and error and the help of YouTube tutorials. If you're asking yourself, 'What's a cassette?' it's probably best to visit your local bike shop and have a pro handle the work. Those 'treadmill' style conversion kits—such as the Rubbee X, Livall PikaBoost 2 rear-wheel units, or the Clip front-wheel unit—are typically the easiest to install, as you simply need to fix the machine to either your bike's seatpost frame or its fork, make sure its rollers are contacting your tire, and off you go. However, that 'contacting your tire' bit is a bit of a rub (pun intended), as the added contact with your rubber might lead to some extra wear. As disc brakes are becoming ever more the norm (as opposed to a traditional rim brake), Skarper offers a conversion unit that is equally as simple to install to your bike's disc-brake mounts and can be done by almost anyone who knows how to use an Allen wrench. The cool thing about each of these aforementioned units is the ease with which they can be attached and removed, meaning your ebike conversion isn't permanent. With them, you can use your road bike as a commuter during the week, yet keep on with your weekend group rides on the weekend. Hub It Out If you're looking for something more permanent (and way more powerful), units such as the Bafang M-Series replace your traditional bottom bracket with a motor that can deliver up to 1,000 watts of added power to the crank arms of your bike. However, as replacing a bottom bracket is one of the more involved things you can do on a bike, this isn't something you'll want to swap in and out. Somewhere in between the clip-on ease of units like the PikaBoost and the more permanent solution the M-Series offers, are the aforementioned hub-based power units, which tuck motors into the hubs of your wheels. Many of these units also offer battery packs that are meant to be attached to your bike's existing bottle cage mounts. Among the many options for hub-based power units, it's hard to beat the Zehus, which requires nothing more than replacing your existing rear wheel with a Zehus-outfitted wheel. Similarly, Cytronex, one of the earliest and longest-standing ebike brands around, offers a hub-based conversion kit that, while a bit more involved, should take no more than a few minutes of setup. With the explosion in popularity in ebikes, this list is barely scratching the surface. For each of the units mentioned here, there are a dozen or more aftermarket competitors offering similar products. Do your research, talk to the experts at your local bike shop, and decide which unit is right for you and your needs. Once you do that, keep the rubber side down! Power up with unlimited access to WIRED . Get best-in-class reporting that's too important to ignore for just $2.50 $1 per month for 1 year. Includes unlimited digital access and exclusive subscriber-only content. Subscribe Today .
Forbes
41 minutes ago
- Forbes
AI Is Behind 50% Of Spam — And Now It's Hacking Your Accounts
AI is taking over the spam and phishing sectors. getty Artificial Intelligence is, I admit with a certain amount of begrudging respect, impressive to put it mildly. At least when it is used to help make work more efficient and leisure more, erm, leisurely. However, when AI is misused or employed for nefarious purposes, it becomes a concern for all of us. We've already seen this when the first big story broke as an AI attack on Gmail users went viral in 2024. Things have changed since then, and not for the better, as the latest research reveals. Now, half of your spam is generated by AI, and cyberattacks are increasingly using AI-powered methods. Here's what you need to know. It's official: more than half of the spam that you receive has been created using AI tools. That's the finding of newly published research, a collaboration between Barracuda and researchers from Columbia University and the University of Chicago, which found that, in April 2025, the actual number was 51%. This isn't altogether surprising. After all, AI does a better job, for the most part, in producing less spelling and grammatical errors, ensuring that linguistically the messages are understandable across geographies, and can be tweaked to have just the right tone to convince the reader to respond. And that, dear reader, is concerning. Not from the perspective of spamming in the broader sense, but rather when it comes to cybersecurity implications, as such techniques are applied to phishing attacks. The same report found that, already, 14% of the business email compromise attacks analyzed were AI-generated. Extrapolate that across all phishing attack scenarios, and I'm sorry to say, the situation will soon become untenable. Wei Hao, a PhD student at Columbia University, and one of the researchers behind the report, said that 'spam showed the most frequent use of AI-generated content in attacks, outpacing use in other attack types significantly over the past year.' What the research also found was that AI-generated emails didn't differ significantly from human-generated attack emails, at least not in terms of engendering a sense of urgency. It appears that AI, like human attackers, recognises the effectiveness of this method in persuading a recipient to act and become a victim. 'Urgency is a deliberate tactic commonly used to exert pressure and elicit an unthinking response from the recipient,' Hao said, which suggested 'attackers are primarily using AI to refine their emails and possibly their English rather than to change the tactics of their attacks.'
CNET
an hour ago
- CNET
Act Quickly to Nab a Pair of Our Favorite Headphones With 20% Off
We'd all like to own some of the best headphones, because why wouldn't you want better music or audiobooks? They can be pricey, especially if you're looking for things like high-quality noise-cancelling or other features. You can find deals that offset that though. Right now, there's a deal on that knocks 20% off the Edifier W830NB headphones as long as you use the on-page coupon. That brings the price down to $64, and given that these headphones are on two of our "best of" lists, that's an excellent price. The W830NB headphones look sleek, which is always a nice bonus for over-ear headphones. They offer a frankly absurd 94 hour battery life, and have some of the best noise cancelling we've tested, especially at this price. They even manage to squeeze in spatial audio, and an impressive sound quality as well. They're great whether you're listening to music, podcasts or even making calls. You can even fold them away when not in use for easier storage and better portability. Hey, did you know? CNET Deals texts are free, easy and save you money. There's no doubt in our minds that these headphones are worth it. Given how good they are and the fact that the discount is decent too, this is definitely one of the best headphone deals going on right now. Why this deal matters We review a lot of headphones here at CNET, so it takes a lot to truly stand out. The fact that these ones do so then, is worth paying attention too. We think these are incredible headphones at full price, so saving 20% is a no-brainer.
