The Hon Andrew Leigh MP
Assistant Minister for Productivity, Competition, Charities and Treasury
What Should AI Love?
National Press Club of Australia,
Canberra
Wednesday, 7 October 2026
I acknowledge the Ngunnawal people, on whose lands we meet. My thanks to the Italian Embassy and the National Press Club for organising and hosting this event. It is a pleasure to share the stage with Father Paolo Benanti and our moderator Corrie McLeod.
Today’s discussion matches the arc of how many Australians have been thinking about AI. After ChatGPT launched in 2022, most of us were focused on how it could make us more productive – summarising documents, checking our work or planning holidays.
Since then, progress has been rapid. One measure of AI agents’ ability to carry out complex tasks has doubled roughly every seven months. As the technology has become more capable, the talk has turned to its impact on humanity.
In these short opening remarks, I want to ask two questions. What can AI do for human wellbeing? And what values should guide our AI systems?
Start with a man with ALS who had largely lost the ability to speak.
In 2025, researchers implanted 256 electrodes in the part of his brain involved in speech. AI decoded his neural signals and turned them into speech almost instantly.
The synthetic voice resembled his own voice from before ALS. He could change his intonation, emphasise particular words and even sing short melodies.
Now let’s move to a more vexed use of AI.
Researchers asked 105 cancer patients to compare two sets of empathetic statements. One set were written by palliative care doctors. The other set were written by AI. Not knowing which were which, patients preferred the AI-generated statements 72 percent of the time.
A machine that had never held the hand of a dying cancer patient produced words that provided more comfort than professionals who had.
The potential of AI to improve human flourishing is substantial. AI systems can help clinicians read scans and treat disease. It can give people with disability new capabilities and make dangerous work safer. It can democratise access to expertise. It has the potential to raise living standards, increase life expectancy and create societies that take better care of the most disadvantaged.
There are plenty of dangers too. AI can produce persuasive falsehoods and supercharge scams. Some of the architects of the world’s most advanced AI models now say that at the extreme, rogue superintelligence could end humanity. Australia’s AI Safety Institute is actively engaged on all these challenges.
There is also a question about wellbeing: what happens when AI starts substituting for other people?
Across the world, people are turning to AI companions. To see why, we need only look at the evidence from recent randomised experiments. One experiment found that a 15-minute conversation with an AI companion reduced loneliness about as much, immediately afterwards, as a conversation with another person. People underestimated how much it would help.
As a Gen Xer, I can’t help thinking that the salve of an AI chat may be fleeting – that building relationships with imperfect people is better than turning to digital devices.
Yet a sizeable share of young people see it differently. In an Australian YouGov survey this year, 8 per cent of Gen Z Australians said they had felt like they were in a romantic relationship with an AI chatbot. And 23 per cent said they would sometimes prefer staying home and talking to a chatbot to going out with friends.
What does it say about Australian society that nearly a quarter of young people sometimes prefer to hang out with an AI, and nearly one tenth have opted for an AI lover?
Part of the solution lies in building community: supporting neighbourhood groups, community sporting clubs, community foundations and local charities. We need to strengthen these offline options if they are to compete with the addictive temptations of AI-driven social media, chatbots and gaming.
Which brings me to the second question: what values should guide AI itself?
AI designers often say that their systems are ‘grown’ rather than ‘built’, which means that it is essential that the right values are embedded in their design.
Anthropic literally publishes a Constitution for Claude. OpenAI publishes a Model Spec. These documents talk about principles such as safety, honesty, human autonomy and freedom. When goals conflict, somebody has to decide what the system should prioritise.
There is no value-free objective function for a machine asked to help us live.
Magnifica Humanitas talks about dignity, freedom and truth, but also about love. Drawing on Augustine, Pope Leo argues that societies are ultimately shaped by what they love: whether we orient ourselves chiefly towards ourselves, or towards one another and the common good. In the digital age, he calls for a ‘civilization of love’.
Talking about love makes policymakers shift uneasily in our seats. But in this case, it’s the right language to use, because it helps focus our attention on the kinds of relationships the technology will encourage.
The Matrix shows a dystopia in which AI systems maximise attention and dependence. By contrast, an AI utopia would require systems that enlarge human agency, foster genuine connection with other people, and encourage us to think deeply about the big questions of religion and philosophy.
We don’t have much time. If the current pace of improvement continues, the AI systems of 2030 will be one hundred times as capable as the systems of today. The better they get, the more our choices matter.
The best AI systems can give an ALS patient back his voice.
The challenge is ensuring the rest of us don’t lose ours.
ENDS