I’m uncomfortable with this article for several reasons. Firstly, if you’re trying to reach a broad (even just within software engineering) audience and convince them of your point, you use a lot of very strange technical language. You write,

“In the case of neural networks, we restrict the search to a continuous subset of the program space where the search process can be made (somewhat surprisingly) efficient with backpropagation and stochastic gradient descent.”

I tried using google to figure out what ‘continuous subset of the program space’ even means, and I found nothing. In fact, every mention of the phrase comes from this article and related pages. But assuming this actually does mean something, the only people who understand it are people who probably already agree with whatever you’re saying since they’re as enswathed in neural-net research as you are.

Second, this article seems to make the assumption that all software problems amount to subjective recognition problems, like understanding what someone is saying to the computer as often as possible. I don’t even know what to say if you think people don’t need web browsers or file managers or, well, operating systems? But I’m not an expert, and maybe there is a way to do these things. If so, please point me in the right direction to read about it.

As a final point, I think I’d say that neural nets will be useful, and will probably make great additions to existing technology, but I’d guess that no more than 20% of software 50 years from now is ‘software 2.0.’ This article reads like you wrote it to convince a clueless investor to pay you for your research or something.

Also that ending. ‘I do not know what technology software 3.0 will be made with, but software 4.0 will be made with sticks and stones.’