An introduction to artificial life: ALIFE 2023 - Day 1
The event is held at Hokkaido University, where the grounds are full of quiet streams and sunlit forests.
Artificial life is a provocative and profoundly ambitious field. It challenges things most of us take for granted: that humans are unique, that consciousness is inscrutable, and that life itself is distinct from other physical processes. “We really want to create artificial life,” said a slide from ALIFE 2023’s opening remarks today. “Anyone can create it. We want to release the recipes for it and change the world.”
“AIs are powerful tools, they aren’t going away, and they can absolutely be abused. They are so easy to abuse. Information theory can help us understand them and see if people are abusing them.”
Sapporo, Japan, is a perfect place to have these hard conversations. Japanese culture has a rich history of science fiction literature and has been profoundly influenced by ideas related to cyborgism, the intertwining of human and machine bodies. Sapporo is a high-energy city of 2 million people in the scenic mountains of Hokkaido, where it’s easy to feel the tension between organic and neon, past and future.
The opening reception included this gorgeous feast, which reminded me of a Google DeepDream.
This week, I’m here to break some taboos around consciousness, cognition, and human autonomy. What happens when we reverse-engineer these concepts and decompose “hard problems” into easier ones? If humans aren’t the end-all-be-all of intelligence, how should we orient ourselves to each other and the world?
Three Key Insights
Understanding consciousness
For hundreds of years, philosophers have been slowly chipping away at understanding what it means to be conscious. In modern machine learning, it’s important that we understand when / if AIs might become conscious (to avoid accidentally torturing them, for example). For decades, overhyped claims of AI consciousness have drawn public ire – but now, it seems like we’ve over-corrected and slowed down an important area of research.
Consciousness as an emergent property
Our current best guess is that consciousness is an emergent property of complex systems stemming from a combination of computational processes and sensory interactions with the world. It does not occur in isolation but is deeply contextual and integrated into its environment. Examining these processes and mechanisms can help us understand how consciousness works at the individual and collective levels!
The demo hall is full of inspiring - and unsettling - innovations in ALIFE.
Difference between humans and machines
For those of us who have teams to lead and don’t have time to ponder the human condition, it can be easy to put off fundamental questions of consciousness. But if we are serious about AI – integrating it into our work and lives – we must also be serious about understanding how / whether AIs are different from us. That means going straight to the heart of what makes us special: our consciousness, cognition, and relationships to our environment.
Three Faces of ALIFE
Rory Grieg is a research engineer at DeepMind who studies ways to make large language models (LLMs) more aligned with human values. He shares my passion for preventing the proliferation of AI-generated bioweapons and managing dual-use technologies well.
Takashi Ikegami is a leading scholar in artificial life and complex systems at the University of Tokyo. His lab seeks to construct artificial life in the real world by exploring the intersections between chemistry, machine learning, and art.
Simon MacGregor has published a large body of research in computational intelligence, informatics, and evolutionary computation. He studies how intelligent systems create and adapt predictive world models to make better decisions.
Two Sessions I Enjoyed
Natural philosopher Pamela Lyon’s opening keynote, “Who’s Afraid of the Ghost in the Machine,” took us on a gripping journey through the history of the science of cognition and what separates life from non-life. Living systems, including cells, display enormous amounts of intelligence and tend towards complexity and integration in ways that are not living cannot. “Organisms,” she said, “aren't dualities of body and mind but hierarchies of interdependent sub-systems, which are both parts and wholes with some individuality or directedness in their process.”
Clifford Bohm shows our class the complexity differences between non-living and living systems.
Clifford Bohm’s tutorial on “Untangling Cognition - how information theory can demystify brains” included challenging technical concepts, which paired well with his natural warmth and humorous communication style. Information theory is an understudied but powerful tool to address the black box problem of large AI models that uses mathematics to quantify and reduce uncertainty in complex systems.
Looking forward to tomorrow
The best conferences help matchmake attendees who are interested in each other’s work! ALIFE is taking that to the next level by hosting a Neuromatch event, where attendees are paired using a machine learning algorithm that uses publication data and user-submitted favorite article abstracts. I can’t wait!
