Machine learning and artificial intelligence are not the "same thing". To say so does not "oversimplify things just a little bit": It is grossly inaccurate. Expert systems, pattern recognition, robotics and fuzzy logic- all part of A.I.- do not, per se, involve learning (though they may).
By far the most common is supervised learning (in which there is a target variable). Less common is unsupervised learning (in which there is no target variable, but solution quality still might be measurable). Occasionally, one comes across reinforcement learning (long-term performance is measurable, but little or no short-term feedback is available), and a variety of more special-purpose techniques like association rule discovery / link analysis, anomaly detection, sequential patterns mining, frequent pattern mining and probably several others I've forgotten.
Excel's pseudorandom number generator is weak: Agreed. The claim that someone, presumably a well-situated person in the Canadian government, might be able to put their finger on the scale for this applicant versus the others seems reasonable. Otherwise, the notion that a bad PRNG, even the one in Excel, will "favor" one group over another, or one person over others, seems a stretch.
I think that's an interesting question: It seems that many people assume that taxation and regulation should be the norm.
One argument often made by governments in favor of the taxes they collect is the services they provide, especially, in the case of sales taxes, the legal protections they offer buyers. I wonder how well that justification holds up in the world of on-line retail?
In viewing the page titled 'The Criminal is Disproportionately Likely to be a Foreigner', I am reminded of the "New Yorker's View of the World": http://i.imgur.com/A3JwoFr.jpg
The interesting thing about this paper is that the "intelligent" component does not select the binarization threshold directly, but is contained within a loop (see figure 1) which iteratively adjusts the threshold.
I feel that this discussion has wandered. Are you arguing that the Chinese government is concerned with false positives in their social scoring system?
- At least occasionally read something that isn't on a screen (in a book).
- Do some experimentation on your own. Too many pretenders in this field parrot whatever they heard elsewhere. It's hard to argue with actual results.
- Don't mindlessly chase whatever's trendy: that's an endless treadmill and "the crowd" wastes a lot of time.
- Once you understand a few machine learning algorithms, you probably have enough. A bigger toolbox is better, but I recommend paying attention to things other than the machine learning activity itself, such as application of machine learning to the problem, data collection, data preparation and model validation.
