Anthropic, the company behind the Claude series of AI programs, has been wining and dining religious scholars behind a veil of non-disclosure agreements. It is trying to convince them that its AI programs possess a form of consciousness. Building such a consensus would help Anthropic evade criminal or civil charges resulting from harm caused by its software. Australiaโs regulators shouldnโt allow themselves to be conned.
Christopher Olah, a 34-year-old co-founder of Anthropic, has been holding private seminars for months with dozens of religious thinkers to explore the possibility of AI consciousness. Attendees have included various Catholic professors, evangelical writers, a scholar of African Indigenous knowledge, a Sikh human rights advocate, a senior cleric of the Church of Jesus Christ of Latter-day Saints, and an Orthodox rabbi from Israel. The meetings have concluded with dinner at a high-end restaurant and gifts of a handwritten thank-you card, a coffee mug, and a specially bound copy of the AI modelโs โconstitutionโ. Anthropic tried to directly lobby the Popeโs advisers in the Vatican but failed when Pope Leo rejected their overtures.
Talk of near-sentient, proto-moral chatbots is snake oil. Anthropic is merely trying to do what other powerful industry groups have done before: frame a problem and propose a solution that suits its interests. Then its people in Congress and the White House can enact laws that ratify that narrative. It is a familiar tactic known as regulatory capture. Other industry groups have tried previously to influence regulation by reframing the problem to deflect blame from themselves.
In the 1970s, the non-profit group Keep America Beautiful ran advertisements against littering, saying: โPeople start pollution; people can stop it.โ Known as the โCrying Indianโ ad, it showed a Native American in a canoe who later sheds tears because people throw rubbish into rivers. Keep America Beautiful encouraged people to recycle and not throw empty drink bottles away. The commercial won industry awards and is still remembered for its powerful message.
But in reality, the Crying Indian was played by an Italian-American actor and Keep America Beautiful was an industry group created by Coca-Cola, PepsiCo, American Can, Owens-Illinois Glass and other beverage and packaging firms. Their shift from reusable glass bottles to single-use, disposable packaging resulted in billions of plastic containers being discarded. They wanted to shift the responsibility for pollution onto the individual consumer and away from themselves.
Anthropicโs outreach to religious scholars should be understood with this background in mind. Public hostility to risk-laden AI products is growing, and so it wants to grab the microphone and lead the protest. In effect, Anthropic wanted to build a consensus that it ought not be liable for criminal or civil charges. Since it canโt blame consumers, it claims the software itself may be liable. Australiaโs regulators shouldnโt be fooled.
There are three broad philosophical approaches to regulation, each adopted by three of the worldโs superpowers: market-driven, state control-driven, and rights-driven. US President Donald Trump supports the market-driven model, which largely leaves decisions in the hands of the tech companies. Last week, he publicly rejected calls for government regulation, instead calling for โmorally bindingโ commitments. Legally binding ones were unnecessary because the AI companies had โthe smartest people in the world watching over each otherโ. The market-driven approach protects the companies from real-world harms caused by their models. It transfers the risks to society.
The Chinese government backs a state control model, which enables government censorship and mass surveillance, It also preserves stability by ensuring that the ruling Communist Party can take coercive action when it needs to. Some countries find the Chinese model appealing, but a multi-party parliamentary democracy such as Australia would regard intrusive government laws as unacceptable.
Australiaโs regulators may therefore consider the third option, which is a rights-driven model from the European Union. It seeks to protect fundamental human rights, preserve democratic structures, and distribute some of the power and profits from giant digital platforms to smaller tech companies, users and the public at large. The EU has a valuable consumer market of more than 400 million people. It therefore possesses โregulatory powerโ; technology corporations must comply with EU laws that protect citizensโ privacy, human rights and consumer rights, or lose access to a $20 trillion market.
The US hates the EU model and is trying to undermine it because other countries may copy it, thus undermining the profitability of US tech companies abroad. Australiaโs regulators may wish to align their thinking with the EU now, while discussions are still fluid, rather than wait until theyโre already baked in. There is strength in numbers.
Australia can bring our successful domestic experience to these discussions. In the sensitive field of medicines and medical devices, the Therapeutic Goods Administration assesses new drugs before they are sold, tracks their safety once the public starts using them, manages compliance, conducts testing, and takes action against serious, deliberate and repeated non-compliance. Independent specialists in science, medicine, industry and consumer health support the TGA by providing expert advice. AI regulation can be informed by this system, which penalises deliberate, reckless or negligent non-compliance without blocking potentially useful products from Australian shores.
Artificial intelligence will remain useful and important. But buying into fantasies of possibly conscious or โrogueโ AI programs that have their own intentions is ludicrous. Regulators should instead focus on the humans and companies responsible for creating software programs, as they do in other industries.
Professor Clinton Fernandes is part of University of NSWโs Future Operations Research Group which analyses the threats, risks and opportunities that military forces will face in the future. He is a former intelligence officer in the Australian Army.
