An open network allowing to use massive computing power would be incredible. But to be honest, I'm not sure it can be easily done in an open network.
Proof of work or systems like storj work in an trustless network, because there's an asymmetry between the work to be done and the verification. Doing the work is expensive while verification is cheap (e.g. mining is expensive while checking a PoW is cheap and fast). This asymmetry is important to avoid "cheaters".
Main problem with a trustless distributed network providing computing power for ML: task and verification have the same cost if you want to be 100% sure that the work done is correct and check it by brute-force.
Note that some cryptographic concepts like
zero-knowledge proof may help. I'm not sure they're mature enough for everyday use cases but some people on the forum should be able to tell you more about this.
At last, you might be interested by some concepts from project Ethereum (a platform for distributed applications).
Just be aware that these applications are very "simple"... not intended to implement ML algorithms.