Machine learning won't 'solve' the economics problem (a problem which the TFA doesn't really define). The problem with math in economics is that economic time-series is extremely chaotic -- a practically infinitesimal change in initial conditions vastly changes the outcome of the system. Hey, remember how we can only predict the weather out 15 days *max* (using big ass supercomputers along with lots of soil moisture content, temperature, wind and other seasonal data)?... well the weather is just one of the tiny effects that propagate through economic time-series. Don't forget about psychology, trading strategies, oh yeah, and the fact that people are actively trying to trick your trading strategies into losing money.
No, machine learning is only natural to takeover a human's limited computational ability, but it doesn't solve the problem of unpredictability. In fact, it will make the market harder to predict for joe blow.
That said, TFA did rightly point out that economics is filled with lots of bullshit conjecture and over-rigorized high-brow nonsense.