Popular idioms aside, birds can actually be quite intelligent, with some species demonstrating the ability to use tools and to develop complex social dynamics. This despite some substantial differences in the anatomy of bird brains compared to those of mammals, particularly humans. A recent set of publications revealed stronger similarities at the level of cellular organization and the way neurons are connected even without the same higher-level anatomical organization. More specifically, bird brains lack a cerebral cortex, an anatomical feature of mammal brains, but they do have pallium which were found to be wired like the cortex.
How those connections come to be made is another question. The number of connections in our brain is estimated around 100 trillion. That’s far too many to be specified explicitly by our genes, which number only in the billions (and also inform the rest of our anatomy and physiology). Not to mention that the nature of genes are such that they cannot directly indicate “connect cell A to cell B” anyway. Genetics do play a role in shaping the development of our brains, of course, and likely there are genetic differences that contribute to the anatomical distinctions of bird brains and mammalian brains. But we also know that the connections between neurons develop as a result of activity, feedback and sensory input as well as the molecular biology of proteins and gene expression.
I think it is interesting to consider these findings in conjunction with what we are discovering via deep learning. Here I don’t mean so much the specific results of better Go and chess programs but the general finding that deep learning works at all. The idea behind it is to take little computational units that simulate features of how individual neurons fire and connect a whole bunch of them. Even without putting a lot of structure into the resulting network and starting with random firing patterns, these systems can learn specific tasks like recognizing pictures of cats, defeating world champion chess players, and driving cars. The important features are enough of the little neuron simulators to have sufficient computational power and the feedback to reinforce useful output.
Now, I’m not suggesting that deep learning has achieved the same level of intelligence as humans, or even necessarily birds. Nor do I think deep learning has answered or will answer all our questions about cognition and how our brains mediate our minds. But I do think programs like AlphaGo have something to contribute to those conversations. And alongside neuroscience like this finding in birds, I think a picture is emerging in which our genes influence how much of a brain we have and how it is connected to our sensory systems, but perhaps not as much of how our brain goes on to process information. That part may be more influenced (but not fully independent of genetics) by the input and feedback our brain goes on to receive.
That reminds me of a book on language evolution I reviewed a while back. Instead of our brains having evolved specific language modules or structures, the thesis of the book is that language has adapted to our brains. And so our language skills have a lot to do with simply growing up hearing the language of others. Perhaps this also has to do with the gradual increase in IQ scores, which seems to be happening too fast to be explained by biological adaptation but could be explained by our brains having the same potential but increasing opportunities to learn more as human knowledge expands and the complexity of life expands. Or maybe not; maybe I’ve misunderstood or overreached. Either way, I look forward to having more of these questions answered by chess programs and our fine feathered friends and everything else that can help.
The ceremony recognizing Francis Collins’ Templeton Prize was held last week. If you missed the live stream, here is his address.