Even basic ML/AI strategies can be easily profitable with the current volatility. It drastically reduces the risk of the HOLD strategy, and while not necessarily the top profits, can be much safer.
I've been running various algorithms for the past 4 months with varying success (whole system homemade in Python), all profitable.
I'd also note that as mentioned elsewhere in the thread, what algorithm is very important. This market is like few others, and thus I have found lends itself to very different strategies.
When you place an order which is not immediately matched by an existing order, that order is placed on the order book. If another customer places an order that matches yours, you are considered the maker and your fee will be 0%.
When you place an order that gets partially matched immediately, you pay a taker fee for that portion. The remainder of the order is placed on the order book and, when matched, is considered a maker order. The fee is 0% for that part of the total order."
GDAX has 0% fees on limit orders. Say you set a limit sell at +1% and a stop loss at -1%. You could trigger a buy whenever you predict that it's >80% likely to hit +1% before -1%.
Plug in everything you can get (trades, order book, etc) from GDAX's WebSocket API into an RNN, and I'd guess you can be right more than you're wrong. Aka profitable.