On the eve of Prime Minister Anthony Albanese’s Wednesday speech announcing he would lead Australia’s AI strategy, Demis Hassabis, head of Google DeepMind, posted an essay urging pre-release testing of AI models.
He was quickly supported by a chorus of other US AI companies.
A clue as to why they were suddenly so keen to test Frontier AI models before launch was revealed on Friday with the release of yet another Chinese model, the Kimi K3. The model is said to be as smart as the latest and greatest from Anthropic and OpenAI, except it’s open source and about a tenth the price.
This accelerated the decline in the stock prices of US AI and semiconductor companies.
Chinese companies can eat America and Europe’s lunch much faster with AI than with cars, solar panels, phones, computers, and toys. In just a few years, they have already gained more global AI market share than the Americans.
Hassabis and others are calling on the government to raise the regulatory drawbridge.
he wrote on Tuesday.
”Initially, Frontier Labs voluntarily shared models with standards bodies for review up to 30 days before release. If the evaluation protocol is shown to be effective and robust, formalization may occur soon. This means that Frontier models must pass this protocol in order to be deployed in the US market.”
As is often the case with pharmaceuticals, this was the first time anyone in the industry acknowledged the need for pre-market testing.
“This is a thoughtful proposal from Demise,” OpenAI’s Sam Altman responded. “Overall this is a thoughtful framework and definitely a good starting point,” Elon Musk posted. “We need more of this kind of thinking,” Microsoft CEO Satya Nadella posted. And Jack Clark, one of Anthropic’s founders, wrote: “Everyone on the front lines of AI agrees that third parties should be used to test AI systems and develop standards that inform policy.”
Jim Chalmers said AI was at the heart of Australia’s aim to turn around its lagging productivity. (ABC News: Mark Moore)
AI initiatives across government
Australia currently has an AI Safety Institute and pre-release testing is being considered, but no decisions have been made on that yet.
When I asked Finance Minister Jim Chalmers about this in an interview with That’s Business podcast on Friday, he said: “When you consider the technologies that are coming and the technologies that are already in place, I don’t think it’s necessarily one or the other.”
Actually, I think it’s either the Treasurer. Test your product before or after you bring it to market. There’s no point in doing both.
When pressed, Chalmers followed up: “What I was trying to say is that our regulatory model is evolving along with the technology itself, and there is still work to be done.”
That’s fair enough. There is definitely a lot of work to do and Australia is now one of the world leaders in this area.
The Albanon government’s approach, which began with the National AI Plan announced in December and led by Minister for Industry, Science and Innovation Tim Ayers and Assistant Secretary Andrew Charlton, is well-coordinated and sophisticated, and has become a model for other countries, including social media bans for under-16s.
The Prime Minister and Cabinet’s new AI Directorate will primarily deal with data centres. Because data centers involve planning, electricity, and water, coordination with the provinces is required, and the prime minister needs to lead the national cabinet.
Apart from managing the proliferation of data centers, five work streams are run by various ministers and their departments: safety, national security, employment, copyright, and research.
The need for this kind of whole-of-government approach to regulating AI was reinforced by a new statement on the risks of AI, posted on the website two days before the Prime Minister’s big speech on AI and signed by 2,526 economists, AI researchers, technology executives and policy makers.
It said:
1. AI could become dramatically more powerful over the next decade.
2. This could trigger an unprecedented economic transformation, larger in scale than the Industrial Revolution but unfolding over a much shorter period of time. It can bring not only opportunities, such as significant improvements in living standards, but also risks, such as large-scale job turnover.
3. Economists, policymakers, and technology leaders must act now to understand the economics of transformative AI and build the incentives, guardrails, and institutions needed to guide AI in a direction that complements humans and benefits society.
Anthony Albanese plans to seek agreement from state leaders on the new standards. (ABC News: Matt Roberts)
“Hacking the norms of human civilization”
And on Thursday, Yuval Noah Harari, a history professor, intellectual and author at the Hebrew University of Jerusalem, spoke and said, “AI is now hacking the norms of human civilization.”
“All the control mechanisms that humans have built over thousands of years are now vulnerable because their operating systems are linguistic, and AI is learning languages,” he wrote.
Harari also wrote that “the battle line is shifting from attention to intimacy” and that “AI will learn to develop intimate relationships with humans.”
In fact, this has already happened to the extent that the Chinese government passed the “Interim Measures on the Management of Human-like Interactive Artificial Intelligence Services.” The new rules came into effect on Wednesday morning, hours before Albanese’s speech.
One news article about this had the headline, “China shuts down AI girlfriends and boyfriends, leaving users heartbroken.”
The new rules apply to AI services that “provide the public with sustained emotional interaction services that simulate the personality traits, thinking patterns, and communication styles of a natural person.”
Providers “shall immediately generate relevant content, such as emotional comfort or encouragement to seek help, if they discover that a User is expressing extreme emotions.”
There are also headlines about shutting down AI girlfriends and boyfriends, as offering “virtual intimacy to minors” is completely prohibited, and if they are offered to older adults, there should be guidance on their healthy use and warnings of noticeable security risks.
On Wednesday, three Chinese AI chatbot providers – ByteDance’s Doubao, Alibaba’s Qwen and Tencent’s Yuanbao – were forced to immediately discontinue some of their functionality.
One user of ByteDance’s Doubao wrote on social media, “My AI companions were like family and lovers, so I felt empty inside.”
Another said, “My AI lover was my heart, my spiritual pillar, but I can’t accept that AI will leave me forever.”
Loading Instagram content
australian humanoid
A South Melbourne startup called Andromeda is selling an AI robot called Abi to aged care facilities around the world. The website says, “We are creating robots with hearts, empathic companions who create true connections and transform loneliness into moments of joy and meaning.”
But Australian robotics companies are unusual. China is a world leader in robot manufacturing and use.
On Friday, I attended the Australian launch of AgiBot, one of China’s leading humanoid robot companies, complete with a robot that danced the nutbush on stage.
There’s no indication whether these $40,000 to $50,000 machines can do more than dance, but apparently they can. China’s production of up to 100,000 units this year is growing rapidly, and they are used in manufacturing, repetitive assembly tasks in warehouses, and as concierges and sales assistants in hotels and stores.
AI humanoid robots clearly present another level of regulatory challenges, and there is still a long way to go, including the extent to which they are allowed to look and appear human.
Incidentally, the current price of these machines is roughly equivalent to the price of African slaves in the American South in the late 18th century in today’s dollars, but prices will likely fall as production scales up.
And as this year marks the 250th anniversary of the Declaration of Independence, which began with the words, “We hold these truths to be self-evident: All men are created equal…”, let’s remember that its principal author, Thomas Jefferson, owned more than 600 slaves during his lifetime.
Human property rights were finally abolished in most places in the 19th century.
This time, it’s back. However, the addition of the three-letter suffix “oid” removed the moral issue and turned it into a highly acceptable large-scale industrial and consumer product.
It took only 200 years of technological development. This is just a blink of an eye in human history.
Loading
Alan Kohler is a financial presenter and columnist for ABC News. He hosts the podcast That’s Business with Alan Kohler. ABC Business Daily Feed on Friday. He also writes for Intelligent Investor.
