Artificial General Intelligence
JOSEPH BRONIO, REPORTER: Will AI permanently destroy humanity? It's a science fiction concept that's been explored in Hollywood many times over the decades.
HUGH JACKMAN AS VINCENT MOORE IN CHAPPIE: The problem with artificial intelligence is it's way too unpredictable.
More recently, though, it's been a real question on the mind of many experts around the world, and a risk that many say should be at the forefront of developers mind. When you look at how fast AI Tech has developed over just the last few years. You can't help but wonder where we'll be in the next few years.
AI GENERATED CHARACTER: Where were you on the night of the bubble bath?
And according to Google DeepMind, the company which made the AI tech responsible for what you're watching right now, the next big step will be to a level known as artificial general Intelligence, or AGI, which is…
XUEYIN ZHA, ANU AI RESEARCHER: I'm trying to find a very crisp definition, but it just isn't really there because people disagree on it. To be very honest with you, it's more of a marketing term by tech giants. What it usually means is that it's something that is so intelligent that is able to excel at all domains of intelligence and exceeds human ability in all of those domains. And so not just any specific reasoning task, but spatial, logical, emotional intelligence, decision making, and even physical execution as well.
In a report released in April, researchers from Google DeepMind outlined their belief that AGI could cause severe harm, including potentially the permanent destruction of humanity. And they're not the only ones with these concerns. In fact, around 600 AI scientists and Big Tech names, including the likes of Bill Gates and Sam Altman, signed the statement on AI risk in 2023 amid growing concerns facing humanity from this tech. And last year, open AI safety researcher Daniel Cocatalyst quit his job, claiming that the company was being reckless in its pursuit of AGI.
None of the concern has slowed down development, though. Many companies like Google Deep Mind and Open AI are confident they will create AGI in some form by about 2030 and other companies like Anthropic say it could even be within six months. But if the companies believe there is such a huge risk to humanity, why are they rushing to develop it?
XUEYIN ZHA: Realistically speaking, you know it's kind of already has its own momentum. Even if you stop people who don't even work for a tech giant might just go home and tweak an open weight model and figure something out. Yeah, the ship has sailed.
At this point, you might be wondering how AGI will doom us all and well, not even the Google Deep Mind report could say exactly. Instead, it lists four major categories of risk.
Misuse when people intentionally use it for harm. Misalignments when the system develops unintended, harmful behaviour. Mistakes from design or training flaws and Structural Risks from the ways different organisations or AI systems interact with each other. But these are all risks we already have with the language model base, or LM, AI of today.
XUEYIN ZHA: They couldn't even imagine what the consequences of AGI in particular is, other than "ohh, what LM is already doing".
That means there are already lots of safety regulations and guardrails in place for those risks, but AGI is a tricky field to predict. While entrepreneurs and companies say AGI is just around the corner — and are pouring billions of dollars into scaling or growing their large language models to create it — AI scientists from around the world haven't been able to agree on when we will see AGI in reality, if we will ever see AGI in reality with the current methods we have now or even what it will be to begin with. And while that means that there could be some unexpected existential risks…
ARNOLD SCHWARZENEGGER AS THE TERMINATOR: I'll be back.
… probably not like that — there could also be massive benefits for humanity and experts like Xueyin say that's what we should be focusing on.
XUEYIN ZHA: For high schoolers, I think it will be really important for them to not inherit some of the fears that older generations might have. It'd be interesting for them to think more about how they could shape society with AI right? So, there's a lot of AI native skills or AI native working styles that are actually happening, and I think they actually face, you know, even more opportunities because they grow up, AI native. It will get rid of some existing jobs for sure, but it will also open up new job areas, some of some of which we haven't even imagined. I think whether you're worried or excited depends on how curious you are and how open minded you are. If we really lean into AI with an open mind, we'll see that it actually helps us liberate our human capability more.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
ABC News
28 minutes ago
- ABC News
Brain Rot: Is internet addiction real?
Plenty of people will say they are addicted to the internet. But how well-recognised, scientifically, is an addiction ... to your screen? In episode four of Brain Rot, we dig into how behavioural addictions work. And we hear from self-described internet addicts about the treatment programs helping them manage their relationship with technology. Brain Rot is a new five part series from the ABC's Science Friction about how tech is changing our brains, hosted by Ange Lavoipierre. Guests: Jillian and Kate Internet and Technology Addiction Anonymous members Hilarie Cash Psychologist and Co-Founder, reSTART Anna Lembke Professor of Psychiatry and Addiction Medicine, Stanford University School of Medicine Anastasia Hronis Clinical Psychologist; Author, The Dopamine Brain Dar Meshi Associate Professor, Michigan State University Credits: Presenter: Ange Lavoipierre Presenter: Ange Lavoipierre Producer: Fiona Pepper Producer: Fiona Pepper Senior Producer: James Bullen Senior Producer: James Bullen Sound Engineer: Tim Symonds This story was made on the lands of the Gadigal and Menang Noongar peoples. More Information: Changes and correlates of screen time in adults and children during the COVID-19 pandemic: A systematic review and meta analysis. Internet-addicted South Korean children sent to digital detox boot camp.
ABC News
an hour ago
- ABC News
Marathon popularity leaves runners susceptible to scammers
When Jess Iapella went to buy a resale ticket to the sold-out Gold Coast half marathon she thought she was internet savvy enough to avoid being fleeced by a scammer. Ms Iapella, a 36-year-old executive assistant on the Gold Coast, put out a call for a ticket on a marathon Facebook page that had 4,000 members. She received numerous offers and decided to buy from someone who appeared to be a genuine runner, with a profile picture of them wearing a race medal. Ms Iapella agreed to transfer half the money via PayID, but it soon became apparent there was no ticket. The popularity of the Gold Coast marathon has grown exponentially in recent years. Tickets for this year's half marathon, which cost up to $170, sold out in four hours, while the full marathon, priced at up to $215, sold out in four days. More than 10,000 people are on the waiting list, leaving many desperate to take part. David Tuffley, a cybersecurity expert from Griffith University, said scammers were looking to capitalise on people's fear of missing out. "There's been an upsurge in this kind of thing around the world, any major sporting event seems to attract this, where thousands of people are coming in," Dr Tuffley said. Like most scams, Dr Tuffley said there were red flags to look out for. "If the offer is well below what you would expect, that's the number one point. The second point is if there's a sense of urgency — and there is always a sense of urgency involved in these scams — they get people," he said. Dr Tuffley said people used to be able to spot scams by bad spelling or poor English, but now with the rise of AI helping scammers write copy, scams were harder to detect. Gold Coast Marathon organisers said they had worked to scam-proof the event, with the transfer of tickets done exclusively through the registration portal. While the marathon did not offer an official resale platform, Gold Coast Marathon CEO Ben Mannion said runners could transfer tickets and when done correctly through the portal no money changes hands between individuals. "We want to control the opportunity that people buy legitimate entries," he said. "Just like any event, whether it's a Taylor Swift concert or anything involving ticketing, you're going get scammers. "We want people to be diligent and be smart with what they do so it's really easy to contact us and we can talk them through the transfer process." Jess Iapella said when she reported the fake profile to Meta it was quickly taken down, but she thinks the tech giant should be doing more to protect users. "Couldn't they detect the profile earlier rather than later if they have knowledge that this Facebook profile is not legit and there are pictures taken from another runner?" she said. As for the marathon, she intended to try to run again next year — even though she would approach the event with less trust. "I'm definitely disappointed, but I'm definitely going to be watching it and maybe next year I'll be ready to run the full marathon," she said.
News.com.au
3 hours ago
- News.com.au
Biocurious: CSL's R&D chief says drug development is like falling in love
CSL's hereditary angioedema prophylactic Andembry is one of only a handful of FDA-approved, Australian-developed drugs Andembry reflects CSL's R&D approach of sticking with 'adjacent' competencies CSL spends around $2.3 billion a year on R&D, but it's still outgunned by rivals and needs to play smart CSL (ASX:CSL) global head of research and development Bill Mezzanotte likens drug development to falling in love. 'If you don't put your whole heart into it, it won't work,' the Pennsylvania-based Mezzanotte told Stockhead. 'But when you do, you are at risk of having your heart broken and that sometimes happens.' The blood plasma giant certainly has had its share of heartbreak along the way, including a failed heart attack mega trial. But as a 'hopeless romantic', Mezzanotte remains confident about the life-saving potential of the company's circa US$1.5 billion ($2.3 billion) a year R&D program. Justifying this optimism, CSL last week won US Food and Drug Administration (FDA) assent for Andembry (garadacimab), a new treatment for the severe swelling condition hereditary angioedema (HAE). In commercial terms, it's the most significant win in years for the $116 billion market cap behemoth. Rare local drug success for rare disease Developed over close to two decades, Andembry is one of only a handful of Australian-developed drugs to be approved by the FDA. It's also the first non-plasma derived monoclonal antibody to be discovered and developed entirely by CSL. Monoclonal antibodies are lab-engineered proteins that bind to a specific target such as a cancer cell or a virus. Because they're not produced from painstakingly collected blood, these remedies have much more attractive margins. Fittingly, CSL's new facility at Broadmeadows in northern Melbourne will make Andembry. Regulators in Europe, the UK, Japan, Switzerland, the United Arab Emirates and locally already had approved Andembry. But to re-phrase Paul Keating, if you are not approved in the US – the world's biggest reimbursed drug market – you are only camping out. Same but very different Andembry is intended as a powerful prophylactic for HAE, which affects about one in every 50,000 people. The disease can cause fatal swelling throat swelling. Andembry stemmed from an antibody 'library' of potential targets which CSL in licensed from the Nasdaq-listed Dyax Corp (later acquired by Shire Pharmaceuticals). CSL discovered garadacimab on the 'bookshelves', along with some other molecules. From that base, CSL did all the research and clinical development, including a 64-patient global trial dubbed Vanguard. 'We took the lead very early,' Mezzanotte says. 'We bought it all the way from research to product development.' CSL's HAE armamentum consists of the injected Haegarda and the infused Berinert (for use during attacks). Haegarda requires twice-weekly injections, while Andembry is administered monthly via a 'patient friendly' auto injector. Mezzanotte says Andembry is 'completely different' to Haegarda. Haegarda is a plasma-derived version of the C1 esterase inhibitor, which HAE patients lack. With a more 'upstream' action, Andembry targets an associate pathway called factor XIIa. 'Both are effective but effective in different ways,' Mezzanotte says, adding that Andembry is 'darned effective' with low side effects. The Vanguard trial showed Andembry reduced the median number of HAE attacks by more than 99% compared to placebo. Blockbuster potential But won't Andembry simply cannibalise Haegarda (and Berinert) revenue? Haegarda chalked up US$491 million of sales in the 2023-24 year, with Berinert contributing US$242 million. Mezzanotte says patients may prefer C1 inhibitors 'because they may do a little more than necessary'. Haegarda works well for pregnant women and it possibly is better for more severe attacks. 'Scientifically they live well together,' he says. But management hopes Andembry will steal plenty of share from Takeda's Takhzyro, the market-leading HAE drug turning over US$3 billion but also requiring twice-weekly injections. Bell Potter analyst Thomas Wakim dubs Andembry 'CSL's most commercially attractive near-term new product.' His peer at Wilsons, Dr Shane Storey estimates Andembry will achieve peak annual sales of US$600 million within five years. In the pipeline Andembry exemplifies CSL's R&D approach of not straying too far from what it's good at, whilst not being afraid to disrupt its own products. 'I'm a big believer in adjacency – utilising our plasma platform – or sticking to an area we know like HAE and disrupting ourselves,' Mezzanotte says. 'When we get too far away from our base of operations it gets a little tricky for us.' CSL's annual R&D manifesto outlines a dense agenda of advanced and nascent programs across its Behring (core plasma and specialty) and Seqirus (flu vaccine) arms. An adjuvanted trivalent cell-based flu vaccine – which includes three flu strains to be more effective – is the most advanced. The program is in phase III and may not need more data to win approval. CSL is also trialling CSL300 (clazakizumab) for patients on dialysis for end-stage kidney disease (to reduce inflammation and heart attacks). CSL inlicensed clazakizumab from another company that had tried the drug for kidney transplants without success. Vifor? I'll tell you Investors periodically have criticised CSL for its $18 billion, seemingly left-field acquisition of the Swiss kidney health and iron deficiency group, Vifor. One of CSL's stated reasons for the purchase was to expand its R&D pipeline. 'Vifor gave us an insight into kidney disease, which allowed us to use a drug we had ,' Mezzanotte says. 'It helps us to be efficient with a great chance of success.' CSL's global head of research and development Bill Mezzanotte. Pic: supplied Checking out of Heartbreak Hotel In February last year CSL experienced heartbreak – literally and figuratively – with the failure of its drug program, CSL-112, to prevent secondary heart attacks via cholesterol-reducing mechanisms. The company's 18,000 patient phase III trial failed the primary endpoint. The therapy showed the desired activity, but just not enough. Mezzanotte says the candidate had passed futility analysis, which – as the name suggests – appraises whether continuing a program makes sense. Mezzanotte says CSL took the conscious risk of eschewing a smaller intermediary trial. 'It was a particularly tough choice, a smaller trial would only have led to the bigger trial,' he says. 'We decided the most cost efficient was to go straight to the big trial but it was a risky approach and we failed.' He says the lesson of such setbacks is to 'fail well'. CSL-112 has gone back to the labs for researchers to have another look-see – but it won't be subject to another big-ticket trial. AI speeds up the 'unsexy' drudge work Mezzanotte says while AI is still immature, it could expedite early-stage drug discovery by enabling programs to skip early-stage animal modelling and to go directly to human studies. 'In theory it should improve speed and the probability of success, but we are in the infancy of using that,' he says. 'Automation has improved our productivity, but I'm hoping AI can bridge the gap with all the big laboratory work we can't afford to do. 'We haven't seen the tangible benefit yet, but we see the promise of it.' He says while AI advocates focus on the 'sexy' research side, the technology is helpful for prosaic tasks such as preparing regulatory and safety reports. 'We create a lot of documents in R&D,' Mezzanotte says. 'We used to send it by trailer, now we email it.' Getting bang for 2.3 billion bucks CSL targets an R&D spend of 10-11% revenue, equating to the $2.3 billion at present. While this sounds capacious, rivals are spending up to US$5 billion annually. Given that, the company needs to glean maximum efficiencies from its global complement of 2000 R&D staff, about one-third of which are in Australia. 'We pick and choose carefully,' Mezzanotte says. 'We can't build huge research palaces and we try keep lean and use partners whenever we can to extend our reach.' Having held senior roles at Boehringer Ingelheim and Astrazeneca, Mezzanotte says commerciality should not be a 'four letter word' in R&D. 'We look for areas of unmet need and rational targets to work on,' he says. 'There is plenty of commercial input early on, but the first few years of Andembry all about the science.' Mezzanotte says through 'great successes and some disappointments', investors have long supported CSL's R&D endeavours. 'I hope shows them the power of R&D and what we can do at CSL.'
