|Volume 1: No. 06|
Larry Press, in his 4/91 CACM column, mentions some significant advantages of personal computers as research tools. Since large and small schools alike use the same hardware, research progress can be distributed as software; results are thus easily demonstrated or replicated. Students can take their work with them after graduation. And conversion to commercial products is relatively painless. (Fred Brooks, though, once estimated that the cost to finish a programming system is nine times the cost of the prototype.)
I used to develop computer vision algorithms, as did most of the people I knew. Some are still at it, mapping ever-simpler algorithms onto ever-more-parallel hardware. They dream of the day that their customers will buy them pipeline machines, teraflop CPUs, or custom-VLSI neural simulators.
My own enthusiasm faded when I realized what a limited market I was serving, and how unlikely it was that I would ever buy a Connection Machine and go into business for myself. I bought a Mac instead, and have spent the last three years learning to use it -- and I'm still nowhere near exploiting its full potential. I have no regrets whatever, and I hope that Larry's words will lead others to the same career choice.
I will offer one word of warning, though. Common hardware and shared software make it easy to build your research on top of code written by others. For academic research, this is a Good Thing -- provided that significant findings are checked with independently written code. For commercial developers, this is also a Good Thing -- provided that licenses are available for all incorporated code. The potential for trouble arises if you try to move academic software into the commercial world. You may find yourself unable to license critical components. I find this unlikely, but Richard Stallman, Will Tracz, and others concerned about intellectual property rights have lately been raising the alarm. Keep it in mind.