I developed my software idea and worked on my business plan for over a year during my MBA program. As I was explaining my ideas, I had a number of students who expressed doubts based on “common sense” heuristics about my ability to produce the technology my business is based on. The doubts were centered on two things:
- The technical difficulty and “millions of dollars” in costs of creating a new desktop application.
- If it were so easy, Microsoft would squash you like a bug…
- The heavy reliance of AI technologies such as natural language processing, coupled with the difficulty that people have in comprehending that computers can actually “understand” natural language.
- Japanese companies spent and lost billions during the artificial intelligence craze in the 1980s.
- Companies and universities have research labs and tons of resources that you can’t compete with. Why haven’t they come up with anything yet?
One student remarked that if you can pull this product of, the fact that potential competitors do not even believe that the product is possible to build is a major competitive advantage.
Rather than follow common sense and empty generalities, I repeatedly question assumptions and investigate the issues involved. I come up with proof of concepts and shoot down impossibility arguments by offering counterexamples.
One of the early pioneers of computer technology remarked that his supervisor asserted that computers would never be able to perform mathematical calculations. Everyone would agree that computers can perform calculations today, but most still probably believe that computers will never acquire language ability because such activity would seem fundamentally human and conflict with notions of consciousness.
In assessing the technical difficulty of delivering software, people forget that I was once a software developer at a very big company. Microsoft has a much higher bar to pass than all other software vendors—internationalization, compatibility, accessibility, etc. I know that in the two and half years that I spent to add a few PivotTable features to Excel 97, I could have developed a serious application. I should also point out that I am licensing decades of work from various institutions.
As for the failed Japanese experiment of the 1980s, AI is such a broad term, anyway, and the Japanese appeared to have been focused on unrelated areas like fuzzy logic. There’s also the impact of Moore’s law—more computer processing power and memory, better and more productive tools and languages. My machine readable dictionary, which takes several megabytes of memory, would not be able to fit inside either the high-end RAM or external storage of the time.
There’s also the efficiencies of a focused development process and a holistic application design. By “holistic,” I mean that any weaknesses in the AI can be ameliorated by the design of the user interface—something that I will talk about in a later post.
I have noticed that researchers often try to obtain the general solution and don’t think about creating commercially viable software. In particular, I look at the OpenCyc project with its massive knowledge database, and wonder if they even know what their goals are. Companies like Microsoft and Google have limited vision and apply their research narrowly to search engine queries and command and control.