[dxbydt]

pardon my ignorance, but what is the specific correlation of math skills to dominance in AI ? yes, AI/ML/dnn uses substantial amounts of linear algebra, much the way accountancy uses large amounts of arithmetic. but we don’t see mathematicians racing to be accountants. i have specifically written to several math departments in the US asking if they do any sort of AI/ML related graduate research. Even schools that offer courses like “math of deep learning” are happy to confess that this is just a bait - they are using that course title to hook more students to sign up, but the course content is plain old eigenvalues and matrix decompositions and what used to previously be called “advanced linear algebra”. you aren’t likely to get any AI breakthroughs from math folks - that is simply not the focus of math depts. i have correspondended with several COLT folks who deal with the theoretical end of sgd/neural nets/ statistical learning, and even there the correlation with math is expendable.

now if you are talking statistics, in particular applied statistics, most of those departments are retooling by hiring ml folks from csee departments. but the bread and butter courses required to get a stat phd remain the same as they were a decade ago - 2 core inference courses, 2 core math-stat measure theoretic courses, 2 courses on experimental stats. literally no stat dept mandates an ML course, though a few do allow ml as an elective. now, all of this can and will change over the next few years, but its very early days.

essentially, predicating ai dominance on raw math skills is a mistake. there’s no significant correlation.

[yorwba]

The PISA test (at least when I took it) measures math skills in calculation, i.e. the kind of applied mathematics that academic departments write off as too trivial. Nonetheless, those are the skills that are required for Deep Learning as well as accountancy.

You could make an argument that high performers on the PISA test would find AI and accountancy too boring and would go into pure mathematics instead; but I don't think those proportions would be different between countries, so a higher potential (schoolchildren with good math skills) should still translate into more AI researchers, on average.

[AlexCoventry]

> those are the skills that are required for Deep Learning

Why do you believe that?

[yorwba]

Rereading that, I should have worded it as "those are skills that are required for Deep Learning research". There are of course a bunch of other skills you need to be successful, but you won't get far without being comfortable with linear algebra and calculus.

To answer your question, I believe that because I haven't yet seen a Deep Learning paper which did not couch its results in terms of those, even if it might not have been strictly necessary. If you do not understand the basics of current approaches, you'll have a hard time developing them further.

[nopinsight]

We are not talking about the level of professional mathematicians or ML theory researchers you mentioned [1]. An average person, in the overall population, is unlikely to pass or possess the prerequisites (e.g. two semesters of college calculus) to attend the linear algebra course in the first place [2] [3].

For simple, routine applications, one can call machine learning APIs without deep understanding of the math behind them. In most real world applications, however, input data tend to be messy and there are many complications and constraints to satisfy. For the foreseeable future, applying machine learning in the real world requires input from humans who understand the why's and the how's behind an API.

For more advanced work, reading papers is necessary and most of the recent machine learning papers contain substantial math (from the viewpoint of an average person, not a professional mathematician).

In addition, math skills, at the basic level that PISA measures, are highly correlated with logical thinking, which will be relevant for AI engineering for a long time to come.

[1] There are perhaps fewer than 100,000 professional mathematicians in the US out of over 100 million knowledge workers there. https://mathoverflow.net/questions/5485/how-many-mathematici...

[2] https://www.noodle.com/articles/the-problem-with-college-cal...

[3] http://hechingerreport.org/high-failure-rates-spur-universit...

[averagewall]

I'm going to get smashed into the ground for saying this, but a lot of psychology research has found that East Asians have a higher average IQ than caucasians. That's known to be a strong predictor of success at math at school and surely has something to do with ML.