On your first point, sorry but I really disagree. You can teach...

On your first point, sorry but I really disagree. You can teach a machine how to compare income statements, cashflow statements and balance sheets, but how can you teach it how to read a whole annual report, or a whole decade of annual reports, and detect patterns such as consistency of business plan, honesty of management, value added organically as opposed to coming from exogenous factors, etc.?

It might be possible theoretically, but it is prohibitively difficult in practice. That's why I personally think that robo-advisory will prove a failure when the next crisis hits, for the very reason that no machine can reliably detect that sort of nuances, and all machines will end up recommending the same things.

But on your last point I do agree with you: automation will not necessarily favour new entrants.