Could AI develop goals of its own?

Could an AI system develop goals of its own? A University of Auckland scientist in a $7 million research programme may help to answer that question.

Computer scientist Dr Matthew Egbert has joined an international team investigating how agency, intelligence and meaning arise in living systems – and whether artificial systems could ever possess the same qualities.

Physicists, philosophers, biologists, and computer scientists from the US, Canada, New Zealand and Spain are being funded by the John Templeton Foundation, a US philanthropic organisation, for the three-year programme of research and collaboration.

The project will establish the Virtual Institute for the Physics of Agentic Intelligence (VIPAI). Much of the money – Waipapa Taumata Rau, University of Auckland is receiving about $650,000 – will go toward employing postgraduate researchers.

“AI is moving forward so quickly that people are increasingly comparing it to human intelligence,” says Egbert. “Yet even the simplest bacteria act on their own behalf in a way that AI doesn’t.

“Humans know from our own experience that we have intentions, goals and values. But exactly what are intentions, goals and values, and could bacteria or AI have them?”

The research project will help to inform that discussion.

“What is required to have those things? I don’t think that the couch that you’re sitting on has them – but why is that? What is the difference between objects, like that couch, and agents like you or me?”

Egbert’s work as an interdisciplinary computer scientist was inspired by the US graphics artist and researcher Karl Sims, whose computer-generated creatures made of blocks altered over time in a simulation of Darwinian evolution. (Black Mirror aficionados will have seen this type of thing turning ugly.)

The Auckland scientist creates computer models that mimic living systems so that he can test ideas about how living systems work. He creates “artificial life”.

“Philosophy of mind using a screwdriver,” is a term Egbert likes for testing ideas by building working models. The phrase comes from the late British computer scientist Inman Harvey, an inspirational figure in the field of artificial evolution.

A bacterium seeking nutrients, a bird navigating its environment and a human making plans all act in ways that support their own existence. They continually distinguish what matters from what does not.

“Living organisms, from bacteria to humans, don’t simply process information. They have needs, goals and a stake in their own continued existence,” says Egbert. “Understanding how that kind of agency emerges is one of the fundamental questions about life.”
 

Current AI systems solve problems set by humans, he says, adding that people shouldn’t mistake computers’ “neural networks” for replicas of the human brain.

“Yes, the human brain was the inspiration but the so-called neurons in neural network systems are not at all like those in our skulls or our bodies, each one of which is an incredibly complex and dynamic living cell.”

The international researchers will combine philosophical analysis, mathematical modelling and information theory to develop general principles underlying agency, meaning, and intelligent behaviour.

“Agency is about more than reacting to the world,” says Egbert. “It’s about being a system for which events and things matter. A living organism actively maintains itself and creates its own perspective on what is relevant, beneficial or harmful.”

He adds: “If we want to know whether future AI systems could ever be genuinely autonomous, we first need to understand the principles that make living systems autonomous.”

Egbert is keen to highlight the value of so-called blue sky scientific research.

“When research is excessively constrained by its expected outcome, it becomes inflexible,” he says, limiting scientists’ abilities to go where their research takes them.

The collaboration is being led out of the University of Rochester in New York State and brings together the Santa Fe Institute, Dartmouth College, the University of British Columbia, the University of Auckland, and the Basque Foundation for Science.