5 ms·

> We need to move to companies seeing that data as a liability, not an asset this could lead to serious limitations on legitimate future progress including on

> We need to move to companies seeing that data as a liability, not an asset this could lead to serious limitations on legitimate future progress including on anything built using statistical methods that require large training data sets.

CaptSpify 9y ago
too bad, so sad. If companies were trustworthy with data, than that wouldn't be an issue. Why should they get a free lunch on my dime?
claytonjy 9y ago
That's certainly a concern given the way things are currently, but I'm hopeful concepts like Google's federated learning [1] will become increasingly popular, and a change to data-as-a-liability would help drive development and adoption of such approaches. [1] https://research.googleblog.com/2017/04/federated-learning-collaborative.html https://research.googleblog.com/2017/04/federated-learning-c...
- FridgeSeal 9y ago
  I really, really like this approach to data, and there's so many advantages beyond the ones outlined there: if you could turn that collection approach into a universal/OS-level API you could have some kind of setup where user data by default stays on the device, then to use it, the app/app server has to request it explicitly (allowing the user easy and powerful, but granular control over who gets what data) and then user data is further protected by the other measures talked about in their paper. Some level of public education is also required I think, to make people aware of the value/threat of just giving everything up every time someone asks for it.
Espressosaurus 9y ago
And nothing of value was lost, except to marketers.
nkrisc 9y ago
We choose to limit medical progress by restricting the experimentation that can be done on humans because we've deemed such protections worthy of the cost to progress. If protecting personal information limits or slows some progress, so be it. Also it could still be done, just more carefully.
mattgreenrocks 9y ago
Yeah, cause they're doing so much with it now /s
eeZah7Ux 9y ago
I see very little evidence that personal data is useful for that. Can you name real life examples? Most machine learning is done against anonymous dataset and works fine.
- chadgeidel 9y ago
  Machine learning can also de-anonymize that anonymous dataset. How does one protect against that? https://en.wikipedia.org/wiki/De-anonymization https://en.wikipedia.org/wiki/De-anonymization

CaptSpify 9y ago

too bad, so sad. If companies were trustworthy with data, than that wouldn't be an issue. Why should they get a free lunch on my dime?

claytonjy 9y ago

That's certainly a concern given the way things are currently, but I'm hopeful concepts like Google's federated learning [1] will become increasingly popular, and a change to data-as-a-liability would help drive development and adoption of such approaches. [1] https://research.googleblog.com/2017/04/federated-learning-collaborative.html https://research.googleblog.com/2017/04/federated-learning-c...

FridgeSeal 9y ago
I really, really like this approach to data, and there's so many advantages beyond the ones outlined there: if you could turn that collection approach into a universal/OS-level API you could have some kind of setup where user data by default stays on the device, then to use it, the app/app server has to request it explicitly (allowing the user easy and powerful, but granular control over who gets what data) and then user data is further protected by the other measures talked about in their paper. Some level of public education is also required I think, to make people aware of the value/threat of just giving everything up every time someone asks for it.

Espressosaurus 9y ago

And nothing of value was lost, except to marketers.

nkrisc 9y ago

We choose to limit medical progress by restricting the experimentation that can be done on humans because we've deemed such protections worthy of the cost to progress. If protecting personal information limits or slows some progress, so be it. Also it could still be done, just more carefully.

mattgreenrocks 9y ago

Yeah, cause they're doing so much with it now /s

eeZah7Ux 9y ago

I see very little evidence that personal data is useful for that. Can you name real life examples? Most machine learning is done against anonymous dataset and works fine.

chadgeidel 9y ago
Machine learning can also de-anonymize that anonymous dataset. How does one protect against that? https://en.wikipedia.org/wiki/De-anonymization https://en.wikipedia.org/wiki/De-anonymization