I think Machine Learning will soon be applied successfully to stock pricing. I doubt I will be the one who programs this, but am interested in the topic. What effect will Machine Learning have on Mechanical Investing in the next 5 years?

My own expectations:

People have been trying REALLY hard to beat the market for centuries.
Every idea that has been invented has been tried, and there is no shortage of capital to fund the attempts.
All the fancy ideas in the last few decades have been automated, many with very fancy adaptive software methods.
A new type of automation is relatively unlikely to come up with new ideas that work so well they change the landscape.

There are presumably still lots of market inefficiencies to exploit.
I suppose machine learning bots might accelerate wiping out the few remaining obvious ones that still work and are subject to being arbitraged away, if there are any left.
But other than that? Just another player in the market.

I add the phrase "subject to being arbitraged away" because some things aren't.
Momentum is the simplest example: if someone tries to front run a momentum factor, it just adds to the momentum at a very slightly shorter lookback.
The exact mix of momentum strategies that work best will continue to shift, but I can't see any way they can be eliminated by having more participants.
It's probably much the same for the end-of-month effect. When there are more participants it might start (and end) a couple of days earlier, but doesn't cease to exist because of that.

Speaking of well funded quant software projects, I guess we all wish we had invested in the Medallion Fund and been allowed to remain.
$100 invested at the start of 1988 was worth $398 million by 1998...and that's after the shocking fees of 5 and 44 for most of that time.
Allegedly: Average hold period ranging from a-day-and-a-half to a-week-and-a-half, percentage of winning trades around 50.75%.

Jim