Is the brain a computer? That is, does neural computation explain cognition? One reason to believe it does is that, much like a computer, the brain functions as an input-output system. Stimuli received as input from sensory systems are processed and a corresponding response is generated. But even among those who do believe the brain computes, there exist various theories about which type of computational system the brain might be. Cognitive scientists on the other side of the debate believe that the brain could not possibly be a computer, for its operation doesn’t follow any pattern of behavior we would expect if it were one. The debate between these two groups continues even today and little progress has been made toward answering the question.
However, researchers at the University of Missouri in St. Louis, Gualtiero Piccinini and Sonya Bahar, say that while the brain is in fact a computer, it is not the kind of computer that traditional computationalists make it out to be. In a new paper published in Cognitive Science, the authors argue that the nervous system fulfills four criteria that define computational systems. First, the nervous system is an input-output system. For example, the nervous system takes sensory information such as visual data as input and also generates movement of the muscles as output. Second, the nervous system is functionally organized in specific way such that it has specific capacities, such as generating conscious experience. Third, the brain is a feedback-control system: the nervous system controls an organism’s behavior in response to its environment. Finally, the nervous system processes information: feedback-control can be performed because the brain’s internal states correlate with external states. Systems that fulfill these four criteria are paradigmatic computational systems. But how does the brain satisfy the criteria?
Computation in a generic sense involves the manipulation of medium-independent “vehicles” or variables based upon the information they carry. Operations that process information do so based upon rules that are sensitive to this information. To see how the brain performs computational operation on variables, the authors focus on “spike trains,” the means by which neurons send signals to one another as well as muscle fibers. If a neuron transmits a spike above a certain threshold, a signal is transmitted. These spike trains are medium independent: they are similar throughout the nervous system regardless of the type of input stimuli. But these types of signal presumably could be implemented on a silicon-based circuit. Thus brains are generic computers.
The authors explain that many of the traditional objections to computation don’t work. One of the most common objections is that the brain exhibits certain behaviors, such as conscious experience, that cannot be explained by computation. The authors reply that, even if computation cannot explain some behaviors, all the objections shows is that computation is not enough—but this does not mean computation plays no role at all. Another objection is that neural processes are non-electrical processes: these systems don’t only use electricity like computers do; much activity involves chemical substances such as neurotransmitters and hormones. The authors reply that computational systems can accommodate chemical signals as well. There is no principled reason why electrical and chemical signals cannot interact in such a way as to provide for computation.
Perhaps the most pervasive objection to neural computation is that, while the brain is an analog system (a continuous system that operates with the world in real time), computation is digital (occurs through a series of discrete steps that are insensitive to time). The authors argue that while the brain is computational, it is neither analog nor digital. It is true that certain processes evolve continuously. For example, the release and uptake of neurotransmitters and hormones varies continuously. And so does transmission by dendrites and some axons. But the most functionally significant signals transmitted by neurons occur through spikes, which are discrete in nature—the spike is either there or it’s not. These discrete events fall outside of the realm of analog computation.
But the brain cannot be digital either. Information transfer within a digital system occurs through the transmission of sequences of data, or strings. For example 1234 is a string that consists of the numbers 1, 2, 3, and 4 in that order. When I type 1-2-3-4 on my keyboard, each keystroke is transmitted to the computer as a distinct binary element (0001, 0010, 0011, 0100). It is much easier to electronically transmit information through a series of 1’s and 0’s because these two numbers can be represented simply by switching electricity on and off. But while research has found that neurons in the brain do fire regularly, it turns out they can’t possibly transmit strings. The firing rates of the same neuron may speed up or slow down or stop firing altogether. In order to decode a string like 0001001000110100, the receiving axon would need to “understand” when time between two spikes each representing a ‘1’ indicates ‘00’ or ‘000’. However, the way neurons fire don’t allow for this.
The authors conclude that if the brain is a computer, but neither analog nor digital, then it must be of it’s own kind, or sui generis. This may have important implications for how we understand artificial intelligence. If the brain differs in computational type from the electronic computer in such a way that the electronic computer can only model brain function, strong artificial intelligence may not be possible. But it may also turn out the differences between electronic and neural computers are small enough that a real-life HAL is on the horizon.