Incentivized machine learning

One feature of Bittensor that Matt hamilton mentions in Microsoft Reactor London is that in any mining case all miners around the world provide POW to solve the puzzle in order to get the reward. When any specific miner can solve the puzzle the efforts put into mining the same block by other miners essentially go to waste. What Bittensor does is actually incentivize those efforts, utilize them through AI, and reward them.

How does that work? what would it look like in tangible terms? Are those people miners as well a AI contributors? I am unable to understand how the work that is intended for one specific purpose which is to solve a very specific puzzle can be utilized for anything else. Also, who will be awarding them tokens if they were unable to mine any block?

AI contributors can be in direct connection with the clients in Bittensor (correct me if I am wrong). In that case, if you are a contributor then mining is different from being a contributor. Although I think that there could be a specific code for each of them or if you want to combine them. But I am not sure about it.

I am not sure of the connection between the contributors and the clients but the miner aka AI contributors are the same as one. I have not learned the minute details of the subject nor how the code looks like for this but they can both be run on the same system. As the meter of the connection between the client and miner is dependent it could be anonymous using code but still be kept direct.