4 ms·

There's consent, and then there's informed consent. It's not really possible for people to have given informed consent to AI model training on their data in an

There's consent, and then there's informed consent. It's not really possible for people to have given informed consent to AI model training on their data in an era when AI had much more limited capabilities relative to what it has now. It's one thing to give consent for gmail to train on your data to develop better spam filters or profile your interests; it's another to give consent for Google to convincingly impersonate your writing style, your facial tics, and your speaking voice. I think few people on Earth in 2012 really understood that their data could ever be used in such a way, and so their consent was not informed.

squigz 2y ago
> I think few people on Earth in 2012 really understood that their data could ever be used in such a way, and so their consent was not informed. I think we've been able to see this coming for many, many years.
- godelski 2y ago
  Post hoc ergo propter hoc You may think this now, but I'm willing to bet a lot of money that you didn't believe this in 2012 and even 2015. Be careful to not rewrite history. It's an easy thing to do.
godelski 2y ago
> I think few people on Earth in 2012 really understood that their data could ever be used in such a way, and so their consent was not informed. Arguably this is just not consent, plain and simple. There could be no reasonable expectation for a person to believe they are agreeing to this. Whereas uninformed consent, I'd argue, is more like how people were agreeing to giving away their data but do not actually understand how the data is taken and used (let's be real, this includes even a lot, if not most, tech people).
Workaccount2 2y ago
If you use Google, your data actually is their data. If you're not paying for it, you don't own it
skeledrew 2y ago
The "informed" aspect tends to be covered by a "future yet unknown uses" blanket clause.
mr_toad 2y ago
> It's not really possible for people to have given informed consent to AI model training on their data in an era when AI had much more limited capabilities relative to what it has now. I can’t see that argument holding up in court. You’re using a definition of informed and consent that are rather novel.
- mitthrowaway2 2y ago
  These definitions are standard in medical studies, but AI training on personal data is the wild west by comparison. That might well change. It took a few ethical calamities for medical law to adopt these practices. https://legaldictionary.net/informed-consent/ https://legaldictionary.net/informed-consent/
jillesvangurp 2y ago
Most legalese is ignored by the people that are supposed to read that when they do online things. But that doesn't mean it doesn't apply. Ignorance is not a valid excuse. Arguably, the average tik tok user probably did not bother to read a single word of the terms of use they agreed to. The textual content of these things does not even matter at this point when users simply don't read it at all and just blindly agree to anything you put in front of them. Funnily enough, you can actually use LLMs to make sense of these things. We're more empowered than ever to process legal text. It used to be that your only option was to consult somebody with legal training. Which of course isn't something a lot of users would do. In any case, you are using a moralistic argument in a legal context. That's simply not how the law gets interpreted in court rooms. It might motivate politicians to write new laws though. Which then get interepreted in court rooms. But given the complexity of the matter at hand, the outcome might not be something that is a net positive. In fact that is extremely unlikely.

squigz 2y ago

> I think few people on Earth in 2012 really understood that their data could ever be used in such a way, and so their consent was not informed. I think we've been able to see this coming for many, many years.

godelski 2y ago
Post hoc ergo propter hoc You may think this now, but I'm willing to bet a lot of money that you didn't believe this in 2012 and even 2015. Be careful to not rewrite history. It's an easy thing to do.

godelski 2y ago

> I think few people on Earth in 2012 really understood that their data could ever be used in such a way, and so their consent was not informed. Arguably this is just not consent, plain and simple. There could be no reasonable expectation for a person to believe they are agreeing to this. Whereas uninformed consent, I'd argue, is more like how people were agreeing to giving away their data but do not actually understand how the data is taken and used (let's be real, this includes even a lot, if not most, tech people).

Workaccount2 2y ago

If you use Google, your data actually is their data. If you're not paying for it, you don't own it

skeledrew 2y ago

The "informed" aspect tends to be covered by a "future yet unknown uses" blanket clause.

mr_toad 2y ago

> It's not really possible for people to have given informed consent to AI model training on their data in an era when AI had much more limited capabilities relative to what it has now. I can’t see that argument holding up in court. You’re using a definition of informed and consent that are rather novel.

mitthrowaway2 2y ago
These definitions are standard in medical studies, but AI training on personal data is the wild west by comparison. That might well change. It took a few ethical calamities for medical law to adopt these practices. https://legaldictionary.net/informed-consent/ https://legaldictionary.net/informed-consent/

jillesvangurp 2y ago

Most legalese is ignored by the people that are supposed to read that when they do online things. But that doesn't mean it doesn't apply. Ignorance is not a valid excuse. Arguably, the average tik tok user probably did not bother to read a single word of the terms of use they agreed to. The textual content of these things does not even matter at this point when users simply don't read it at all and just blindly agree to anything you put in front of them. Funnily enough, you can actually use LLMs to make sense of these things. We're more empowered than ever to process legal text. It used to be that your only option was to consult somebody with legal training. Which of course isn't something a lot of users would do. In any case, you are using a moralistic argument in a legal context. That's simply not how the law gets interpreted in court rooms. It might motivate politicians to write new laws though. Which then get interepreted in court rooms. But given the complexity of the matter at hand, the outcome might not be something that is a net positive. In fact that is extremely unlikely.