Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

What they've inferred about you: https://www.google.com/settings/ads


that website doesnt show anything they inferred about me, just some settings.


That's very little of what their models have on you to power each product.


What makes you think they have more than that?

I'm pretty sure Google has plenty of users willing to share their data that they wouldn't risk the legal and political shitstorm that would come out of secretly storing data about you...


Open up Google Inbox or Google Now -- all these 'smart' features are extracted and interpreted from your underlying data. Or do a google search logged in vs not logged in.. the results will be different.

The models are very complex and often not something that can be meaningfully represented, especially when you're a data point inside a neural network.

I have no idea why I'm being downvoted... anyone who has done any machine learning can attest to these basic facts, not to mention I've worked at Google.


I never said it wasn't complex or could be meaningfully represented, but you were implying that they have more "data" than they are showing.

If you are worried because they are making "hidden associations" from this data, my question would be how to display that? How would they possibly show you all of the assumptions, associations, interpretations, and calculations they did, do, or will gather from your data?

If you are going to fault them for not displaying that data, shouldn't it be possible to display it in the first place?


A fine grained list of thousands of items can have summaries. These summaries can reflect actual models, or be new ways of visualizing and digging into the data.

Existing models can be depicted in a wide variety of ways. They could either be the results of previous decisions (description through action), they could be expressed as is should the model be more human-readable (e.g. text clusters), they could be visualized somewhat directly [1], or they could visualized indirectly [2].

"My activity" seems like a nice idea, but the execution ends up providing little value to few people. If the idea is transparency over Google's personal data on you (how they framed it in an email to me), it does a poor job. If the idea is a better understanding of yourself (quantified life), it does a poor job. If the idea is finding some youtube video you watched recently, then it's mediocre depending on how long ago it was. In what scenario is it some spectacular & useful product?

[1] http://static.azinman.com/pdfs/dataportraiture.pdf

[2] http://yosinski.com/deepvis


I would be more surprised if they were not asked to store publicly not accessible information about everyone.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: