I turned off the computer as I left my office, deleting the neural network simulation that I had been working on. Then a thought hit me: had I just killed something? I rationalised that the simulation was simpler than the systems within the bacteria I was doubtlessly squashing on the floor. If they did not matter, neither did it. But the doubt remained…
SCIENCE has a problem. If we want to find out what really happens in living beings, or how to cure diseases, often we must experiment on them. Digital simulations offer a way out.
Ever since the squid giant axon was modelled in the 1950s using a mechanical calculator, our ability to simulate biological systems has skyrocketed. Today we can run neural simulations on supercomputers that contain hundreds of millions of realistic neurons and billions of synapses. Cells and their chemistry have been modelled to a similar extent. This potentially offers an alternative to animal experiments. Instead of subjecting a living creature to pain when testing a painkiller, why not simulate the pain system and check whether the treatment works? The logical endpoint of this is an emulation, where every part of the brain – and body – is simulated digitally.
Advertisement
The challenge is to map the connectivity in real brains. It will be years before we can create even a proper insect brain, but work is already under way to create the . The worm is a good candidate for this because it has one of the simplest brains of any organism, with just 302 nerve cells. In 2012, researchers at the University of Waterloo, Canada, announced the creation of a large functional brain simulation, SPAUN, with 2.5 million neurons. And the Human Brain Project, a European collaboration, has the ultimate aim of simulating a whole human brain.
[youtube]https://www.youtube.com/watch?v=EPNL914XnLI[/youtube]
Although these digital emulations could resolve many existing ethical dilemmas, they raise new ones. The first is that many real animals must be sacrificed to create a virtual one. We may one day scan the final lab rat, which will become Standard Lab Rat 1.0, and rely on simulation from then on – but there will have been years of basic neuroscience to enable that simulation. The second problem is that we need to be certain our simulations are right if we want to trust them with our drug testing or other research.
It is the third problem that really interests me. Do we have to care for them like we do for animals or humans involved in medical research?
This hinges on whether software can suffer. For example, lets users observe the behaviour of rats given electric shocks to teach the psychology of learning without using live animals. Yet few of us would think there is any real pain there: it is essentially an interactive cartoon, similar to a virtual pet toy. We may empathise with it, but it is similar to a talking doll. Whole brain emulations, which recreate the neural connections of animals and even humans, are a different matter.
In his 1978 paper “Why you can’t make a computer that feels pain”, philosopher Daniel Dennett argued that we don’t have a rigorous enough definition of pain, so we cannot build a machine that feels it. But he also believed that we may eventually figure it out, and at some point thoughtful people would refrain from kicking robots. Other philosophers, such as John Searle, have argued that no matter how sophisticated the simulation is, it will always be mere numbers updated in complicated ways: there cannot be real intentions or consciousness in pure software. It might also be that the system needs to have a body to ground it in the real world.
But what about , created by the neuroscientist Rodney Cotterill as a model of his theory of consciousness? It is a virtual simulated infant with a brain and body model based on real biology. It has internal states such as blood sugar levels, and activity in different brain areas. It responds to these internal states, it can learn, it needs food – if its nutrient levels are too low it “dies” – and it can cry and flail its arms. Yes, it is a very simple organism, but it is intended to be conscious. There is something eerie about it: assuming Cotterill’s theory is right, in principle this being could have experience.
“Turning off the neural network simulation, it hit me: had I killed something?”
We know brains exist for motivating actions that lead to better outcomes for the organism: this is the whole point of pain, pleasure and planning. If we were to make a perfect copy of the activity of a brain, we would get the same behaviour, based on the same pattern of internal interactions. There is no way of telling from the outside whether it has any real experience, whatever that is. There is considerable disagreement about whether software can suffer, or whether it matters morally. So what should we do?
My suggestion is that it is better to be safe than sorry: assume that any emulated system could have the same mental properties as the organism or biological system it is based on, and treat it accordingly. If your simulation just produces neural noise, you have a good reason to assume there is nothing in there to care about. But if you make an emulated mouse that behaves like a real one, you should treat it like you would treat a lab mouse.
I agree this is inconvenient for computational neuroscience. But it is probably the moral thing to do. Once we get to simulated vertebrates, we ought to apply government animal-testing guidelines. We should avoid generating virtual suffering by not running experiments that produce pain signals. But we can also improve on biology, because in simulations we can (temporarily) leave out pain systems, simulate perfect side-effect-free painkillers or just block neural activity related to suffering. We could in principle monitor the emulated brain for any kind of suffering and stop as soon as we detected it. There is also the issue of quality of life. We have begun to recognise that giving animals good environments matters; building equally good virtual environments may prove a hurdle. Virtual rats would plausibly need virtual fur, whiskers and smells to feel at home.
What about euthanasia? Living organisms die permanently, and death means the loss of their only chance at being alive. But an emulated brain could be restored from a backup: Lab Rat 1.0 would awake in the same way no matter how many copies had been tested in the past. The only thing lost when restoring it would be the memories of the previous experiment. There may still be pleasures and pains that count. In some ethical views, running a million supremely happy rat simulations in the background might be a “moral offset” for doing something painful to one.
In the long run I believe we will create human brain emulations. Their moral status will in many ways be easier to determine than for animals: just ask them. Take an eminent philosopher doubting software can be conscious, scan their brain, and ask the resulting emulation if it feels conscious. If the response is: “…yes. Darn. I need to write a paper!” we have pretty good evidence that there is a being with enough intellect, introspection and moral value to deserve rights. But until then we should treat our software animals well. Just in case.
This article appeared in print under the headline “Death and pain of a digital brain”
