There's probably some utility to it: a lot of problems involve hacking together datasets, sometimes in dubious ways. There's also value, especially for startups that are looking to build simple neural net applications (ex: identifying plates of food from different restaurants) which are very data-dependent. Researchers may also want to reflect the cost assembling datasets (ex: MTurk, processing power) and open up datasets that may never have been open before.
My general sense on this though is that I'd like there to be more of an incentive for people to open up their datasets to the larger public. Maybe I'm being idealistic but a crowdsourcing type function where you pay for X dataset together with other users and then it's released under MIT, forever free etc.
As others have mentioned that'll probably bump against usage rights issues, a larger problem you'll have to deal with independent of your need to sell or distribute the datasets in question.
My general sense on this though is that I'd like there to be more of an incentive for people to open up their datasets to the larger public. Maybe I'm being idealistic but a crowdsourcing type function where you pay for X dataset together with other users and then it's released under MIT, forever free etc.
As others have mentioned that'll probably bump against usage rights issues, a larger problem you'll have to deal with independent of your need to sell or distribute the datasets in question.