In a room the size of a small kitchen, three researchers work amid a crush of computers, wiring, books and papers. At the centre of the room is what looks like the head and torso of some modern Frankenstein's monster with the skin removed: a mass of wiring drooping off a skeletal structure, with wheels instead of legs and feet.

The contraption at the computer science department at the University of Essex bears little resemblance to the sleek industrial robots familiar from films and factories; but it is one of the world’s most advanced automatons.

And the researchers, led by Professor Owen Holland, are addressing a question as fundamental as any posed at the multi-billion dollar particle accelerator at Cern: can a machine be conscious?

Philosophers and scientists have for decades discussed this question in terms of the Zombie enigma: if you were to build a machine that is functionally identical to a human (no matter whether you can; this is a thought experiment) would it be aware, in the way that we associate with being alive?

Three years ago, with what seems like remarkably little fuss, Britain’s Engineering and Physical Science Research Council gave Holland and his team a £314,000 grant to build something of the nature of a zombie: the beginnings of a machine sophisticated enough to experience consciousness. “There is no point in philosophers sitting down and talking about these ideas. Someone has to try them out,” he said.

Holland, drawing on ideas going back to the1940s as well as some of the latest advances in neuroscience, is not simply building a mechanical model of a human on the off-chance that it will exhibit signs of life. He is incorporating into his robot, called Cronos, what he believes could be the architecture of consciousness.

This is where it gets scary for anyone attached to conventional ideas of what it is to be alive. Holland believes consciousness is an illusion, that it rests in a mind model of ourselves interacting with a mind model of the world around us: a virtual person in a virtual world.

The real world exists (though not precisely as we see it) and our bodies really interact with it - but only, so to speak, by acting out our fantasies. We experience consciousness when our virtual person switches on and begins registering and making sense of our rich sensual data.

It turned out to be surprisingly easy to implement this architecture in software because the tools are already available in the games industry. The internal models for Cronos have been constructed in a module called Simnos, using the same Ageia PhysX software used by games coders. This defines how entities behave based on weight and other physical properties.

The idea is that Cronos will gather information about its environment, mainly using its single ‘eye’, which will be used to build its interior world model. It will need to know, or discover by interaction, something of the physical properties of what it is looking at. Its ‘virtual person’ - again a Simnos model - will know what actions it can perform in the environment and on the objects he sees.

Team member Hugo Marques is building in ‘functional imagination’, the ability to rehearse a task in a virtual world to find a solution that can be acted out in the real world (see here for images). Another software model covers the interactions between the virtual person and the virtual world.

“We go through processes like this all the time,” said Holland. “It’s like the old party trick where someone asks you to pick up a model of something that looks much heavier than it is. Your entire body is primed for the task and you get a shock when it is not as you had anticipated.”

Marques plans to build memory into his system so that the robot can remember goals it has already achieved. Holland explained: “One of the issues is deciding when to use imagination. If you have done something before, you don’t need to imagine. The only thing is, how long do you spend imagining?

“Some people are like this. They spend so much time imagining they never get round to doing. Instead of imagining, you might think ‘I’ll go with what I have done so far.’ Nobody has actually built systems that face up to these problems.”

Two approaches are being taken to controlling the robot, or rather allowing it to control itself. One is probabilistic, using statistics to deal with real-world uncertainties.

Richard Newcombe, who is currently refining the motor control of Cronos, explained: “This is not a highly engineered system, like you might find in a racing car. Instead it is more faithful to biology, because biology gets over the rawness of the system, and the noise, and the inability to control everything very precisely, with a system that deals with this uncertainty and can take advantage of it.”

The parallel approach being developed by another team member, David Gamez, links Simnos to a form of neural network which simulates the behaviour of neurons and is closer to the way our brains work.

Cronos differs from industrial robots not only in having this complex autonomous ‘mind’, and the fact that it has not been designed for a specific set of tasks. It is what Holland calls anthropomimetic: its structure is based on that of a human, even down to the fact that some of his joints are straight out of Gray’s Anatomy. You may wonder what physical structure has to do with consciousness, but our own inner life is not confined neatly to grey matter. E motions, for instance, clearly have a physical component.

“I don’t think you could have any meaningful consciousness without having a complicated body,” said Holland. “I wanted to get something that was the same order of complexity as the human body. But that does not mean that I have to copy everything faithfully.”

The Cronos body was made by another team member, Rob Knight, who runs a company called The Robot Studio. “We made a few simplifications - the neck is longer than a human’s and the spine has fewer vertebrae. We had some problems getting enough motors in the space so we have made some engineering approximations. But, qualitatively, controlling Cronos is almost as big a problem as controlling the human body.”

Cronos has 35 degrees of freedom (the number of ways his body parts can move) which is much the same as a human from the waist up. Also as in our bodies, movements are not powered by rigid connectors: Cronos uses bungees as ‘tendons’ (see Why robots are best held together with bits of string).

The motor control on Cronos is not yet up to scratch and Newcombe is currently refining it. For this reason, and to allow the software to be developed without endangering the robot, its ‘mind’ is separated from its body. This requires an additional layer of modelling within Simnos, which will be replaced by the ‘real’ Cronos when he is tuned up for the task.

Holland does not claim that this initial work will produce anything like we understand as conscious. He sees it as empirical science: you try something, see what is missing, adapt your machine, and try again.

The question left hanging over all this is: even if you did produce consciousness, how could you tell? We cannot even know that another person is conscious, though we may infer it with near certainty; in the case of someone in a coma, or even sleeping, it is not obvious.

Igor Aleksander, emeritus professor of neural systems at London’s Imperial College, has set out what he considered to be five signs of consciousness: a sense of self, imagination, focused attention, forward planning and emotion. Holland takes these as a starting point. “Igor says if you find they’re wrong, change them; find out what’s missing and add it in.”

The German philosopher Thomas Metzinger, whose work influences much of Holland’s, has listed another 11 criteria, and neuroscientist Giulio Tononi has developed an informational metric called Phi which he takes to be a measure of consciousness.

There is a certain circularity in defining criteria for consciousness, building a machine to meet them, and saying it is therefore conscious. But Holland says he is not setting himself up as judge and jury in his own case. “I can say [Cronos] is conscious by Aleksander’s standards, it meets five out of 11 of Metzinger’s criteria, and its Phi is this and the human Phi is that… It’s up to other people to say whether it is conscious or not.”

Implicit in the idea of Tononi’s Phi is that there are degrees of consciousness: a rat would be less conscious than a human because it can’t process as much information. Australian academic David Chalmers claims that even a thermostat has a degree of consciousness because it is processing information.

“I don’t agree with that,” said Holland. “Any system that is going to be conscious is going to have to have the capacity to model itself, to model the outside world, and to model the interaction between the two. Consciousness is a by-product of the development of intelligence.”

He believes that many animals simply don’t have enough brain to support the modelling required for consciousness. These include the smallest mammal: “Whether the 10-to-the-seventh (10 million) neurons of the bumble-bee bat can do that I have got serious doubts.”