You go into a new computer store. Everything seems normal: you see Apple computers, Dell computers, HP computers … the regular computer brands you are familiar with. But then you notice a weird area. On one shelf, there are a bunch of rocks labeled “Putnam’s rock.” On another, there are pieces of brick wall labeled “Searle’s wall.” On yet another, there are buckets of water labeled “Hinckfuss’ pail.” The price of these unorthodox items is lower than the price of Apples, Dells, and HPs — much lower. But something seems wrong. For starters, no specs are listed next to these philosophical computers.
Question: are they a good deal?
The answer depends on which physical systems perform which computations. If rocks, pieces of wall, and buckets of water can perform the same computations that ordinary Apples and Dells can, then these philosophical computers are a good deal — especially considering that you could easily build them by yourself. Otherwise, they may be a scam.
It’s easy to scoff at these putative “computers” and declare that they are not computers at all. You’d be foolish to buy them (qua computers), and even more foolish to sell them (qua computers). The philosophers who argued that such things compute — Ian Hinckfuss, Hilary Putnam, John Searle — seem very wrong. But it’s a lot harder to answer the question that gave rise to these examples: which physical systems perform computations, and which computations do they perform?
This is not an idle question. Consider how quickly computer circuits evolve. In 1965, Intel co-founder Gordon Moore predicted that the amount of transistors on a chip would approximately double every two years. This amazing progression, which has held true until very recently, requires that the size of transistors become smaller and smaller. But nothing can get smaller forever.
In recent years, Moore’s law has begun to bend. The rate of transistor shrinking is no longer what it used to be. The industry is approaching the physical limits of conventional computing technology.
Current microprocessors are built from transistors that are fourteen nanometers wide, and the industry is transitioning to ten nanometer components. A silicon atom is about 0.2 nanometers in diameter, so current transistors correspond to about one hundred layers of silicon atoms. IBM recently announced transistors that are seven nanometers wide — about fifty layers of silicon atoms. At such dimensions, thermal noise and quantum effects will seriously disrupt the classical processes that give rise to conventional digital computations. In other words, the components’ states won’t be reliably ones or zeroes but messy classical and quantum mixtures of ones and zeroes. And then what?
Some physical information theorists, such as Neal Anderson at the University of Massachusetts Amherst, are hard at work trying to figure out the fundamental physical limits of conventional digital computation. They are developing theories of classical digital computation at the nanoscale, where quantum effects are inevitable. To develop such theories, they need to define the boundary between physical processes that count as digital computations and physical processes that don’t.
(Classical digital computation at the nanoscale should not be confused with quantum computation, which is a different theoretical enterprise that also requires appreciating the boundary between physical computation and non-computational physical processes.)
Meanwhile, the computer industry is trying to circumvent the slowdown in the shrinking of computer circuits, adding specialized chips to their current technology and seriously considering unconventional computing methods that they largely ignored until recently.
One such method is neuromorphic computing — building chips that compute similarly to how brains compute. Wait, how do they know that brains compute?
In 1943, Warren McCulloch and Walter Pitts, two neuroscientists of sorts, proposed that neural processes are computations and neural computations explain cognition. Since then, the computational view of brain processes has become dominant in neuroscience and psychology. The mainstream view is that brains compute differently than digital computers — for instance, neural circuitry changes over time in response to feedback, whereas conventional computer chips have a fixed architecture. But the computational view of brain processes remains controversial to this day.
Once again, searching for unconventional computing methods as well as for a neurocomputational theory of cognition requires knowing what does and does not count as computing. A question that may appear of purely philosophical interest — which physical systems perform which computations — shows up at the cutting edge of computer technology as well as neuroscience.
Ian Hinckfuss and his followers were naïve to conclude that buckets of water, rocks, and walls perform any computation you like — although I’m sure even they would not buy such putative “computers” in a computer store. But the question they asked is genuine and deep and it deserves our attention. Figuring out what counts as physical computation is no trivial matter.
Featured image credit: ‘Blue screen of … Envisionware’, by Jeffrey Beall. CC BY-SA 2.0 via Flickr.