Science

If a robot can be made to think like a worm, how will we define living and non-living beings? By Gillian Terzis.

I, wormbot: the next step in artificial intelligence

Timothy Busbice, with his Lego robot that is directed by a simulation of a worm’s brain. Inset: Microscopic C. elegans roundworms.
Credit: Main image: courtesy timothy busbice; inset: Teresa Lee / Flickr

Even before they began to stake a claim on our jobs, our boardrooms, our battlefields and our bedrooms, robots have long activated our existential anxieties, forcing us mortals to ponder our own planned obsolescence. Advances in artificial intelligence deepen these feelings.

Supercomputers with artificial intelligence such as IBM’s Watson and Deep Blue have declared emphatic victories on Jeopardy and against world chess champion Garry Kasparov. And just a few weeks ago it was announced that a program created by scientists at the University of Alberta is invincible at heads-up limit Texas hold ’em poker. Not only can the program bluff – a seemingly cognitive trait – but it is said to “learn” from its mistakes through an algorithmic process known as “counterfactual regret minimisation”.

For all these developments, artificial intelligence systems are still fairly primitive. Yet Ray Kurzweil, prominent futurist, Google engineering director and prominent AI hype-man, believes machines could surpass human intelligence in 15 years. Still, the dawn of the so-called “technological singularity” – the point at which the intellectual capacity of machines exceeds our own – often feels more like speculative fiction than reality. Sentient machines, which would exhibit consciousness, curiosity and emotions, remain a long way off. Human–robot relations are stilted; anyone who’s ever shouted at Apple’s Siri will know such interactions are not yet seamless.

Simulating human traits remains the principal bugbear of artificial intelligence developers. But an increasing number of them believe they can design sophisticated and intelligent machines by going back to first principles – that is, by replicating the neural circuitry of simple organisms. Timothy Busbice is one such developer, who is keen to fuse the knowledge of the neural circuitry of a worm with the aim of building intelligent, autonomous robots.

Late last year, Busbice and a team of scientists uploaded a simulation of the nematode worm’s neural networks into a small programmable Lego robot. A video of the result is on YouTube, which shows the three-wheeled robot skating jerkily around on the floor – if you didn’t know the project’s background you might think it is simply being controlled, somewhat clumsily, with a remote.

Busbice claims the robot’s movements had not been pre-programmed, and its behaviour was directed by the simulation of the worm’s brain. For example, touching the robot’s “nose” resulted in the machine beating a spontaneous and hasty retreat, while activating a “food sensor” made the robot advance. The video has elicited vociferous debate about the project’s validity and accuracy, as well as the metaphysical implications.

Busbice emphasises that although his simulation aims for a high degree of biological fidelity, it inevitably lacks the mess and noise of a real-life central nervous system. And this would seem to be the overarching flaw in computational simulations of neural activity, which play a central role in the nascent discipline known as “executable biology”. Monash University associate professor and bioethicist Robert Sparrow believes such simulations are destined to be incomplete portraits of brain activity. “There is still some uncertainty over whether we are capable of characterising all the behaviour of neurons,” he says. “It’s not clear to me that just capturing the neuronal activity is enough to capture consciousness.”

The simulation of multicellular organisms is no easy feat. No scientist has yet managed to create a comprehensive model of a bacterial cell, let alone a living organism with a brain. It’s no surprise that at about 100 billion neurons, the human brain remains something of a black box for neuroscientists; even a mouse has one million neurons. Making a computational simulation of these nervous systems would be an arduous task, but as researchers such as Busbice have proposed, there are simpler places to start: at present, the focus is on the microscopic, soil-inhabiting nematode (roundworm), otherwise known as the Caenorhadbitis elegans.

The C. elegans worm has been the organism of choice for biologists for decades, for reasons that are practical and scientific. It is transparent, which permits scientists to observe each one of its 959 somatic cells and 302 of its neurons under a microscope; its size (one millimetre in length) allows it to be bred in large quantities in a Petri dish; and it shares physiological traits – muscles, a central nervous system, reproductive capabilities – with animals much higher up the food chain.

Twenty-eight years ago, a team of scientists led by John White and Sydney Brenner published a map of the C. elegans’ neural connections, otherwise known as a connectome. Tracing cross-sections of the worm’s anatomy and figuring out where the neurons connected was a painstaking process that had taken close to 13 years.

Today, an open-source science project named OpenWorm, of which Busbice is a co-founder and former member, is trying to create a 3D computer simulation of the C. elegans, which has also spawned similar initiatives: scientists trying to create simulations of the common fruit fly and the jellyfish.

Another of OpenWorm’s founders, Stephen Larson, says that while he thinks the Lego robot was a “fun and interesting application of the open science approach”, their efforts are concentrated on computer simulations.

“It’s exciting that folks are getting creative,” he says, but adds that he hasn’t seen the finer details of Busbice’s robot and couldn’t speak to its scientific validity. “We feel very strongly about peer review. We want to be doing real science.”

A computer simulation may be a simplified model of reality, subject to certain controls, but Larson believes it still has value. He hopes that as computer simulation technology becomes more advanced, physics and chemistry can more closely approximate reality. “We now understand enough about living systems to appreciate that they are built on the foundations of physics and chemistry, that they are carrying out physical operations and transformations that are knowable.”

Equally intriguing is what remains unknowable, and most likely unquantifiable. While these experiments in executable biology explicitly challenge the dichotomy between the living and non-living, the scientists themselves are reluctant to delve into the slippery liminal space in between. Busbice does not believe his robot to be alive in a biological sense, and both he and Larson appear content to leave the task of defining life to philosophers.

“As a scientist, I don’t have any feelings towards it,” Busbice says of his Lego experiment. “I do kill it.”

Nonetheless, during the course of our conversation he occasionally speaks about the robot with the sort of fondness one might reserve for a family pet. “I liken it to a cat,” he says. “You can try to entice it with food and things like that, but a cat pretty much does what it wants to do.” The robot often scurries about his office, “wandering around like an animal, observing and interacting with its environment. It kind of makes you think it is alive to some degree – as much as a worm is alive.”

The degree of “aliveness” of things is surprisingly debatable. Is a worm alive in the same way as a mammal? As Sparrow notes, there are organisms that exist in the natural world in which a binary categorisation of life and non-life seems unsatisfactory. Viruses, for instance, share some characteristics of the living: they can reproduce (albeit within the cells of other living organisms) and they have an evolutionary history.

Perhaps a more pertinent question, Sparrow suggests, is whether virtual organisms are worthy of moral consideration. If a virtual organism were to reach the functional equivalent of a living one, “there would be questions about whether it would be wrong to cause it pain”. Most humans would place animals in an ontological category similar to our own, but the same cannot be said for computers or robots.

Many of us would hope that our existence is more than a pneumatic network of neurons, valves and ventricles. Such biological determinism may seem depressing – and has been vigorously contested by critics – but Busbice is unfazed. For him, the ghost is the machine. “What I’ve done with the technology, I guess, reduces humans to a bunch of connections. That scares people, that we’re not this benevolent creature, unique in the universe ... we’re just a bunch of wires connected together.”

This article was first published in the print edition of The Saturday Paper on Feb 14, 2015 as "I, wormbot". Subscribe here.

Gillian Terzis
is a San Francisco-based writer.