Mini-Review: Shadows of the Mind

I have just finished reading Shadows of the Mind, by Roger Penrose. It's something I wanted to do a long time ago, specially after reading many other books about mind and artificial intelligence. As one of the best parts of reading is to comment, I will write about my impressions of the book here.

On the bright side, you always learn something interesting when reading Penrose's books. As all the others, this book is sprinkled with interesting bits of mathematics and physics. The objective of Penrose is to argue in favour of an idea that he has been pursuing for a long, long time - the idea that computers cannot (and will not), even in principle, think.

The chain of arguments goes like this. He argues that Goedel's Theorem implies that there are mathematical truths that cannot be accessed in any computable way, which means that no algorithm that can actually be written in a computer can prove their truth. Then, he says that mathematicians can actually identify any mathematical truth. This, according to him, is because mathematicians can understand mathematics, what computers cannot do. This "understanding" must then be some uncomputable process. Therefore, computers cannot and will never be able to really think or understand.

This makes the first part an interesting journey through the realm of computer science, computability and logic. It's the mathematics part of the book. He even writes about polyminoes. I really like it, except for the extremely biased dialogue between a robot and its creator. 

Because of the argument that anything that relies on computable functions, like usual computer algorithms, cannot think, he must find a way for uncomputability to enter into the human brain. As physics up to now is computable, he then argues that computability must be hidden in what has not been understood yet. The most straightforward non understood fundamental problem in physics - and incidentally the one he worked so much with - is quantum gravity. Therefore, there should lie uncomputability. He doesn't really know where in quantum gravity it is, even because we don't still know quantum gravity (despite string theorists claims). Still, he makes some interesting suggestions and relate them to some biology.

Biology, of course, needs to kick in because the brain is... well... biological. So, he proposes that there can be a kind of collective physical phenomenon, much like superconductivity or superfluidity, happening in the brain. At this point, he seems not only worried about uncomputability, but also in explaining conscience. He argues that the quantum gravity proposal is compatible with the microtubules, structures that play an important role in cell structure, being the key organelles where uncomputability and collective phenomena should happen. This implies that conscience might even be present, in some level, in a unicellular organism. But not in a machine!

As I said, it's worthwhile to read the book even if only for the interesting things you learn with it. Penrose also has the nice habit of including references to articles and books everywhere, so we can go after them and learn the details if we want to. Now, the not-so-nice comments.

The problem with the book is the premises. It seems that Penrose has a religious belief that man-made machines will never be really sentient and he then searches desperately for a way to justify this belief. He is very competent, knows a lot of physics and mathematics and argues very well. Still, the problem is in the hypothesis. Right at the beginning, Penrose assumes at least three things which are not true:

1. Mathematicians know mathematical truths.
Although this might be true, it's highly improbable given the amount of disagreement on fundamental issues. Even if there was no disagreement, that is something for which there is no proof at all. Remember that agreement doesn't imply truth (yes, consensus is not enough...)

2. The human mind is consistent.
Do you really want me to start giving examples? Scientists that are religious fanatics. People who give advice and do the opposite. If there is one place where inconsistencies can coexist without problems is inside our brains.

3. We understand things and, by the way, we don't need to define exactly what is "understanding" in this discussion.
It is absolutely beyond my comprehension how a mathematician like Penrose can argue that we can assume that something so fundamental for the whole discussion doesn't need to be defined precisely. Every scientist knows that, when we don't define things precisely, everything goes. Think about all those people talking about positive and negative energy.

But these three hypothesis are central to his whole thesis. The second one, for instance, completely breaks his line of arguments. Once that humans can be inconsistent, this immediately violates the conditions for Goedel's Theorem to be applicable.

In summary, I would still read the book. There's a lot of things to learn in there. And even if his whole argumentation is flawed from the beginning, someone needs to explore those possibilities. It doesn't matter if it's for the wrong reasons. Although the conclusion is wanting, many things can be salvaged during the journey. Yes. I'm still a Penrose's fan. Even disagreeing with him.