INTELLECTUAL PROPERTY PROTECTION FOR NEURAL NETWORKS

The principal forms of intellectual property protection for neural networks in the United States include patents, copyrights, trade secrets, and mask works. As with previous forms of new technology, some aspects of neural networks transcend existing legal categories. This is primarily due to their dynamic nature, as well as the impossibility of predefining the trained state of the system. As a result, these aspects of neural network technology may be left with limited protection until Congress or the courts respond by customizing current laws to fit this technology, much as they have already done with computer software. This article discusses the ways in which neural networks pose novel issues in intellectual property law, issues that will challenge the ability of the legal system to provide adequate protection by stretching the current categories. A strategy is recommended for inventors and attorneys in this field to mitigate the weakness in current laws by making optimum use of a combination of the existing forms of protection. © 1990.