5 ms·

Even if they did there have been many, many HN stories of data being deanonymised absurdly easily, and health data tends to unique very quickly indeed. How man

Even if they did there have been many, many HN stories of data being deanonymised absurdly easily, and health data tends to unique very quickly indeed. How many females in a (county|country|postcode} born on 5 March 1942 are there with a kidney problem? Particularly in the UK where a postcode, even the locality selector (the first half, e.g. M50, NW11) is more precise than many US zipcodes that can cover many square miles. A full UK postcode narrows it down to within the street - a dozen or two houses usually. What if there is associated medical data like medications or chronic conditions - diabetes, arthritis etc? To make it suitable for study a) keep it within the NHS or relevant health authority - perhaps allowing an airgapped DeepMind server to come in process locally and afterwards get wiped or shredded, or b) remove all personal data including gender, dates, associated medical info of any type - i.e. render it almost useless for study. Oh and c) imprison all the management who agreed to share this data without patient consent and in breach of DPA, and require Google to remove all trace under audit.

mda 7y ago
Who said that proper anonymisation doesn't remove exact birthdates and fine granular location? Proper anonymisation techniques are much harder to attack.
- NeedMoreTea 7y ago
  Anonymisation after it's all gone to Google is rather moot, don't you think? Besides in this particular case they rode roughshod over both DPA and our entire structure of medical ethics and sent across entirely non-anonymised data. "proper anonymisation" is a set of goalposts that keep moving - it's like password length or bcrypt work factor. What's enough today won't be enough with CPU, or in a year or three, or perhaps a suitably fancy AI from the world's largest data hog. Proper anonymisation likely requires all associated data - any dates, blood group, medications, chronic conditions etc removing, hence mention of making it so bland as to be almost useless for study.
  - mda 7y ago
    Heh you go on an on about an imaginary weak anonimisation technique then say all is moot and again go on and on about pure speculation, It is you who is moving goal posts.
GhostVII 7y ago
There are mathematical ways you can guaruntee different levels of anonymity though without removing all identifiers. Ex. you can increase age and location buckets until for every unique set of identifiers, you have a good distribution of attributes.
- zwaps 7y ago
  And yet, you know this will not happen, because it wont yield the best and most profitable model for Google. Google will do what they have always done, ignore privacy, train models on full data, apologize later ans delete data when no longer needed. You dont know this, but people at Google are Gangsters and bullies They dont give a hoot about laws in Europe, about privacy or any of that. They think they are right and that everyone else is wrong. See the guy above who said this is for the collective good, even though everyone who knows even a little bit stats and machine learning knows full well that these sort of sensitive data will lead to unimaginable grief for outlier patients. In particular, because Google cares about money, and not the collective good! Everyone should be outraged, it is the only rational reponse.
  - GhostVII 7y ago
    What is this "unimaginable grief" you think outlier patients will receive if anonymized data is provided to Google? Seems a bit hyperbolic.
kapuasuite 7y ago
What’s your goal, though? Is there some harm you’re seeking to prevent? Are you interested in privacy for privacy’s sake? Have you considered the potential benefits of this research that might be lost under your approach?
- sorokod 7y ago
  Companies (like Google) exist for the purpose of making a profit from your data. Their interests and yours are fundamentally not aligned.
  - kapuasuite 7y ago
    As opposed to farms, banks, utilities, service providers of all types, manufacturers of just about every good in existence?
    - sorokod 7y ago
      two words make the difference: "your data"

mda 7y ago

Who said that proper anonymisation doesn't remove exact birthdates and fine granular location? Proper anonymisation techniques are much harder to attack.

NeedMoreTea 7y ago
Anonymisation after it's all gone to Google is rather moot, don't you think? Besides in this particular case they rode roughshod over both DPA and our entire structure of medical ethics and sent across entirely non-anonymised data. "proper anonymisation" is a set of goalposts that keep moving - it's like password length or bcrypt work factor. What's enough today won't be enough with CPU, or in a year or three, or perhaps a suitably fancy AI from the world's largest data hog. Proper anonymisation likely requires all associated data - any dates, blood group, medications, chronic conditions etc removing, hence mention of making it so bland as to be almost useless for study.
- mda 7y ago
  Heh you go on an on about an imaginary weak anonimisation technique then say all is moot and again go on and on about pure speculation, It is you who is moving goal posts.

GhostVII 7y ago

There are mathematical ways you can guaruntee different levels of anonymity though without removing all identifiers. Ex. you can increase age and location buckets until for every unique set of identifiers, you have a good distribution of attributes.

zwaps 7y ago
And yet, you know this will not happen, because it wont yield the best and most profitable model for Google. Google will do what they have always done, ignore privacy, train models on full data, apologize later ans delete data when no longer needed. You dont know this, but people at Google are Gangsters and bullies They dont give a hoot about laws in Europe, about privacy or any of that. They think they are right and that everyone else is wrong. See the guy above who said this is for the collective good, even though everyone who knows even a little bit stats and machine learning knows full well that these sort of sensitive data will lead to unimaginable grief for outlier patients. In particular, because Google cares about money, and not the collective good! Everyone should be outraged, it is the only rational reponse.
- GhostVII 7y ago
  What is this "unimaginable grief" you think outlier patients will receive if anonymized data is provided to Google? Seems a bit hyperbolic.

kapuasuite 7y ago

What’s your goal, though? Is there some harm you’re seeking to prevent? Are you interested in privacy for privacy’s sake? Have you considered the potential benefits of this research that might be lost under your approach?

sorokod 7y ago
Companies (like Google) exist for the purpose of making a profit from your data. Their interests and yours are fundamentally not aligned.
- kapuasuite 7y ago
  As opposed to farms, banks, utilities, service providers of all types, manufacturers of just about every good in existence?
  - sorokod 7y ago
    two words make the difference: "your data"