The Analog Brain: Why You Can't Download Your Brain Onto a Computer - Part III

This post is a continuation from this previous post.

< Previous      Next >

View fullsize

The weather is a classic example of a chaotic system.  Try as we might, we just cannot predict the weather well.  Even with computers the size of villages we haven’t been able to produce a model that mimics the real weather beyond a few days’ time.  Why is that? 

For the answer, we have to turn to Edward Lorenz; mathematician, meteorologist and Father of Chaos (theory).  Lorenz tried to predict the weather by creating a computer model that could calculate how meteorological variables such as temperature, wind-speed, air-pressure, etc. would interact over time.  The idea was that if he could represent each of those variables mathematically, and calculate the effect each would have on the others, there was no reason why he couldn’t calculate the future behavior of the weather.  Just as long as he could get his model right.  To this end, Lorenz took extraordinarily accurate measurements of the weather, so that at the starting point his model and the real weather were perfectly matched.  Once everything was in place, Lorenz hit ‘Enter’, let the model run, and compared the model’s predictions to the real weather conditions.  

Initially, the two systems mirrored each other quite well.  However, within a few days the model started behaving erratically, and would no longer give him a reliable prediction.  Confused, Lorenz went back and checked his data.  The variables he put into his model were exactly the same as the weather measurements, and his model matched the real climate perfectly at the point at which he pressed ‘Go’.  So how could the two systems possibly diverge if their starting conditions were the same? 

It dawned on Lorenz that his model was only a perfect match to a certain level of accuracy.  The measurement he had taken of temperature, for example, was accurate only to a limited number of decimal places. So whereas in his model the starting temperature might be set to, say, 22.00000000000°C and not 21.99999999999°C, in reality the actual temperature could lie somewhere between the two.  This is what’s known as a rounding error, and Lorenz’s model was full of them.

Like the hairs on our snooker table, these tiny rounding errors have very little effect over short periods of time, but after a while they add up to produce drastically different behaviours.  This is the famous ‘Butterfly effect’, where a tiny change in one variable (a butterfly flaps its wings in Hong Kong) can lead to huge differences (a hurricane in New York instead of sunshine) given enough time.  Because these tiny errors are catastrophically additive, a digital model of a chaotic system will only behave exactly like the original if its starting parameters are exactly the same as the original, totally free from rounding errors. 

Which brings us back to the digital brain…

< Previous      Next >

The Analog Brain: Why You Can't Download Your Mind Onto a Computer - Part II

This post is a continuation from this previous post.

< Previous      Next >

In the curious case of the brain-download it is not enough to create a computer model that is similar to a human brain.  For a successful download of your brain we need to recreate, to a high-level of accuracy, 'Your' exact brain.  Your unique and wonderful brain, with all its imperfections and hard-won connections which you’ve trained and pruned your whole life.  The computer model has to be exactly the same as your biological brain in terms of how it behaves, otherwise it simply doesn't count.  A half-baked, inaccurate model would not really be ‘You’, or at least would cease to behave as ‘You’ would after a short amount of time. And who wants to live forever as an approximation of their former self?

The issue then is not whether we will ever have the technology to reproduce your brain digitally, but whether a digital copy of your brain could ever be good enough to be indistinguishable from your real brain. We might almost dismiss the notion on principle because a computer model is, by necessity, a summary of a real scenario.  Since a summary could never contain the same amount of information as the original (otherwise it would not be a summary), the premise is necessarily false.  However it would be foolish to suggest that a model cannot meaningfully represent reality.  After all, we use models all the time.  The map on your phone, for example, is a digital model of your surrounding geography.  The important information, the roads and place names, are contained in the map, but everything else is ignored. The textures and smells of the real world aren’t needed for the map to be useful, so they are ignored. This reduces the amount of information in the model while preserving its behaviour. Theoretically we should also be able to reduce the amount of information we need to represent in our computer-brain-model while still preserving your brain’s behaviour. However, the behaviour in this case is vastly more complex, meaning that the more information we leave out, the higher the chance of producing a bad model.

Models of relatively simple systems can afford to be superficial.  A simulation of a snooker game, for example, doesn’t need to take into account the effect every hair on the snooker table has on the moving ball.  It can afford to ‘zoom out’ and represent the net contribution of all the hairs with a summary variable, such as 'Friction'.  Because the system is simple, our model will be able to make accurate predictions of the ball’s trajectory despite summarising reality to a great extent, and ignoring all the tiny variations between the individual hairs on the table.  Despite this blunt approach, our model will still be a true representation of the real system.

In reality, of course, those little hairs do have an effect on the snooker ball’s trajectory, but their contribution is so infinitesimally small that within the confines of our snooker table the effect is completely unnoticeable. If we had a long enough snooker table though, and a perpetually moving ball, these tiny effects would eventually become very relevant.  In other words, on an infinite table if we plucked a single hair from the ball’s path and re-struck the ball in the exact same way, the new trajectory would noticeably diverge from the original given enough time.

Modelling simple systems is relatively easy because we can afford to be inaccurate.  That is why you can play snooker right now on your phone for 59p if you wished. For complex systems like the brain however, the situation is very different.  Complex systems are determined by lots of different factors, each of which might be influenced by many others.  The number of possible interactions means that complex systems are very hard to model accurately, because an error in any one of these factors will lead to bigger and bigger errors with every interaction. Like a virulent sneeze on a train, the single error infects all factors it comes in contact with, and eventually blights the whole model.

And speaking of colds, we all know first-hand how futile it is to try to model complex systems.  In fact, there is one particular chaotic system that we talk about pretty much every day…

< Previous      Next >

The Analog Brain: Why You Can’t Download Your Mind Onto a Computer - Part I

Leading thinkers such as neuroscientist David Eagleman and philosopher Nick Bostrom believe that one day, you will be able to download your mind onto a computer.  A sophisticated brain scanner will record all the connections in your brain and a computer will then recreate them all digitally.  The digital ‘brain’ will then begin to behave exactly like your real brain, which means it will essentially become you, and allow you to live beyond the death of your body in an eternal 'transhuman' existence.

If that sounds too good/bad to be true, that’s because it probably is.  Replicating the trillions of dynamic connections that exist in your brain digitally would be a truly miraculous feat, and one which is definitely beyond us at this time.  Nonetheless, we should never underestimate the future’s potential to wildly exceed our expectations. With our technology and scientific methods steadily improving, one day we will surely have the capacity to create a computer model with comparable complexity to a human brain.  Progress towards this has already begun, with the 2.5-million-neuron SPAUN brain model recently being created, and the Human Connectome Project working diligently to map all the connections of a single human brain.

However, even if we were able to map and model all the connections of a brain, translating this into a personality-download is a whole different ball-game.  Whether or not identity is embodied (i.e. attached to a particular body) will keep philosophers occupied even while the digital-human-brained robots take over civilization.  The problem being that the minute we recreate our minds in another location, that duplicate mind will begin to have its own experiences and perspective, and will therefore necessarily have a different identity to the original.

This problem is perhaps insurmountable.  But for the sake of argument, and because it would be very cool, lets assume that if we created a perfect digital replica of someone’s brain we will have transferred their identity to a computer.  After all, that in itself would still be a mind-bending feat, even if both resulting individuals remained convinced their parallel twin was an impostor.  In this situation, could we ever be confident that the digital version was faithful to the original brain?

In this ‘Analog Brain’ series I will argue that it is impossible to perfectly replicate a brain digitally. And the problem lies in the chaotic nature of complex systems and tiny unassuming things called rounding errors.

 Next >