Can we build a functional brain using computers? In order to answer that question, we need to know how the brain is built in nature. The human brain is composed of more than 10 billion cells called neurons that can be electrically activated upon stimulation. Neurons produce special proteins called ion channels that are inserted in their cell membranes. These channels allow ions (i.e., electrically charged molecules) to flow in and out of the cell, which generates an electric current. Each neuron is connected to thousands of other neurons by structures called synapses. Groups of neurons with similar functions can communicate closely with each other to perform more sophisticated functions, and such a cluster of neurons is called a “column”. About 10,000 columns form the neocortex of the brain (the outer layers on the surface of the cerebral hemispheres) (Figure 1). When information arrives at the brain, it transforms into electric signals and travels through a large network of neurons to allow our brain to perceive, interpret, think, and generate physiological or motor responses to the surrounding world.
Figure 1. The hierarchy of brain structure starting from a single neuron that contains ion channels. 10,000 neurons form a neocortical column. There are 10,000 columns in the neocortex of the brain. (Illustration by Shan Lou)
Imaging the brain: Techniques and limitations
Despite the knowledge that scientists have about the brain, there are still many big questions in the field of neuroscience that remain unanswered. One of these questions is how perceptions (what we detect through our senses) transform into conceptions (our interpretation and response to what we perceive). For example, if you are presented with a piece of pizza, how does the image trigger the word “pizza” in your mind, how would it be defined and categorized as food, and how does it trigger your hunger? There is a big leap from the functional neocortical column to the thinking brain. Several techniques have been developed to look at the big picture of brain function, such as functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and brain field recordings. But these technologies can only be used in limited situations. For example, individuals have to remain still during these examinations, meaning that many brain activities that occur during active movements cannot be studied. Furthermore, these techniques have low resolution, such that the activities of single neurons cannot be visualized.
To work around the limitations of studying the brains of living organisms, scientists have instead focused on building a computer model of the brain that can be used to help unravel the complex mysteries behind how it functions. But is it possible to build an accurate working model of the brain? Building such a model is a huge challenge in the field of neuroscience, but it is one worth pursuing because it would allow researchers to test various hypotheses about how the brain perceives and processes information. Building a brain model will also help researchers better understand brain diseases, such as stress, autism, schizophrenia and neurodegenerative diseases. It could even be used as a simulation for drug tests and screening.
Making computers think like humans
But the question remains, how are scientists going to build such a brain? It is hard to avoid thinking about computers when talking about the human brain. The close correlation between the way computers work and the way our brain thinks is not a coincidence. In 1936, when Alan Turning was designing his early computer called the ACE (Automatic Computing Engine), he used the human brain as a model. Rather than building a program based on the adult brain, Turning instead suggested basing it on a child’s brain, which could then be trained through learning to function like an experienced adult brain. It was also during the design of the theoretical computing system that Turing proposed the standard of an “intelligent” machine. This standard is based on the Turing test, which identifies a machine as “intelligent” only if a human interrogator cannot tell the machine apart from a human based on the answers it gives to questions [1].
Computer scientists have always dreamed of making “intelligent” machines that mimic the human brain, or even surpass it. In one well-known example, an IBM group developed the Jeopardy! champion computer “Watson”. As evidence of its capacity for artificial intelligence, Watson was able to understand vague questions and search for correct answers by looking for correlations in the fields of humanities, culture and sociology [2,3].
The Blue Brain Project
It’s one thing to make computers think like humans — it’s another to use computers to gain insight into how the human brain works. Yet, scientists are trying to do just that to explain brain activities using specialized computer programs. They first developed hypotheses for how specific neurons might work and used them to generate computer models. These computer models were then verified by comparing them to the real responses given by the same neurons in mice.
Now that scientists have successfully modeled individual neurons using computer programs, will they be able to use computers to model the whole brain? If one computer cannot, tens of thousands of computers might be able to. This idea comes from Henry Markram, a neuroscientist who is famous for discovering how the relative timing of electrical signals in two connected neurons affects the strength of their connection [4]. Markram proposed simulating each neuron with one laptop, basing the parameters of the laptop “neurons”, including their size, ion channels and capacity to be excited, on data from real neurons that were described in scientific papers or studied in his own lab [5]. This is to say that any electrical signals put into the computer should have exactly the same output as that of the actual neuron they are mimicking. Each laptop neuron is connected to thousands of other laptops in the same fashion as the actual brain to form a functional neocortical column. This project is called the “Blue Brain Project”.
Figure 2. Henry Markram and his Deep Blue Brain model design.
In 2005, the first single neuron model was built. In 2008, the first neocortical column was built. It is expected that in 2014 the whole rat brain neocortex could be modeled. The ultimate goal of the Deep Blue Brain Project is to replicate the same procedure with the human brain. It faces challenges in data collection, but with advances in technology, Markram expects it to be completed in 2023 [6].
Yet, as a newborn brain, the Deep Blue Brain might not know what to do with its potential. It is the “child brain” in Turning’s words. For the child brain to grow and develop, it needs to be trained and tuned to the correct output by various algorithms of machine learning, so that the smell of pizza will mean “food” rather than “toy”. How will scientists judge if the project is successful? Markram suggests linking the Blue Brain to a machine rat with certain sensors and motor outputs. The real rat should be able to learn, be able to perceive and respond to its environment, change in response to experiences and, most importantly, be unpredictable. Unpredictability and autonomous decision-making are the main differences between a well-programmed machine and a real biological brain [7, 8]. What if it does not work? Maybe this will give scientists reason to look back and ask the question, what is missing? And maybe it is the answer to this question that will give us insight into how the brain’s perceptions lead to its conceptions.
Shan Lou is a PhD student in the Program in Neuroscience at Harvard Medical School.
References:
1. “Alan Turing.” Wikipedia, The Free Encyclopedia. Wikipedia, The Free Encyclopedia, 18 May. 2012. Web. 21 May. 2012
2. “Watson (computer).” Wikipedia, The Free Encyclopedia. Wikipedia, The Free Encyclopedia, 9 May. 2012.
3. “IBM Watson Jeopardy.” Youtube (2011) http://www.youtube.com/watch?v=qpKoIfTukrA&feature=related
4. Markram, Lubke, Sakmann. “Regulation of synaptic efficacy by conincidence of postsynaptic APs and EPSPs.” Science 275 (1997): 213-215
5. Markram. “The blue brain project.” Nat. Rev. Neurosci. 7 (2006) 153-160
6. Miller. “Blue Brain Founder Responds to Critics, Clarifies His Goals.” Science 11 (2011): 748-749
7. Waldrop. “Brain in a box.” Nature 482 (2012):456-458
8. Markram. “Henry Markram builds a brain in a supercomputer.” TED Talks (2009) http://www.ted.com/talks/henry_markram_supercomputing_the_brain_s_secrets.html