WolframAlpha is a beautiful, amazing computer program. As with any truly novel idea, on first encountering it, one struggles to put it into context. Here is how I view it, from a Bayesian Networks perspective.
Quote from the WolframAlpha website:
What is the core technology of WolframAlpha?
There are many parts to it, each with significant innovations. Four key general areas are the data curation pipeline, the algorithmic computation system, the linguistic processing system, and the automated presentation system.
I view the current WoframAlpha as an intermediate step towards full AI. To answer my own question in the title, I would say
WolframAlpha + Bayesian Networks = Hal 100 (not yet 9000) possible in five years
| AI task | WolframAlpha equivalent task | limitations of current WolframAlpha |
| remember | data curation pipeline | Can’t self-add data (for instance, data acquired from internet) to curated data. In this sense, it’s ability to remember what it hears is very limited. |
| analyze | algorithmic computation system | It’s a deterministic, rule-based analyzer. If it also used bayesian networks and statistics, it could analyze cases, common in real life, in which some of the input data is missing or contradictory. |
| hear | linguistic processing system | Can’t hear except from command line. |
| speak | automated presentation system | Limited ability to say things in meaningful prose. |
| self-teach (learn independently) | not yet | Can’t learn new rules from its own analyses. Bayesian network learning could help here. |
