3 ms·

In 2007 or 2008 or so, when I saw that they had implemented a feature that let people tag you in photos without permission, I assumed it was to gather data for

In 2007 or 2008 or so, when I saw that they had implemented a feature that let people tag you in photos without permission, I assumed it was to gather data for identifying people with AI at some future date (albeit based on my rather different understanding of AI at the time). So, I deleted my account: FB didn't provide me any value, and had introduced an unknown future risk by taking away control of my identity. Ever since then, I've asked friends and family not to upload pictures of me to Facebook, with the natural amount of awkwardness and eye-rolling that request evoked. I assume my friends continued to add me to Facebook without telling me about it, just as I assume that deleting my Facebook account did not actually remove me as a uniquely identified individual in their system. But, what am I supposed to do, just silently assent? Turns out, we're not the product, we're further back on the chain. We're more like ingredients that can be used and reused to create an infinite variety of different products.

echelon 3y ago
> Ever since then, I've asked friends and family not to upload pictures of me to Facebook, with the natural amount of awkwardness and eye-rolling that request evoked. In the future, systems will probably infer what you look like based on connections. DNA relatives, etc.
- edgyquant 3y ago
  How accurate do you expect that to me? If I tell you two people have children how well do you think you can predict what that child looks like?
  - echelon 3y ago
    1. With the child's complete DNA or key markers, I bet we'll eventually have algorithms for DNA -> adult phenotype 2. With only parental or relative DNA, if we are able to gather an enormous amount of extra information -- social group, gym use, hobbies, dating circles (mate selection probably yields a huge amount of info!), profession, fashion purchase activity, etc. -- then I bet we can get pretty reasonable confidence intervals on a lot of physical characteristics. Height, eye color, eye distance, face shape, lip thickness, etc. etc. Perhaps even elements such as voice timbre, etc. If we had their familial phenotypes, photos and phenotypes of all a person's dating history, then I bet that lets us cut the search space down tremendously. If we add music listening habits and hobbies, I bet we'd also be able to cut it down. I'm hypothesizing that physical appearance might correlate from behaviors, social graph information, and limited DNA information. I have nothing to show for that, however. (I would love to read literature on this topic.) I'd imagine that constructing such a predictive model would require an enormous amount of data. Facebook + 23andme could probably pull it off.

echelon 3y ago

> Ever since then, I've asked friends and family not to upload pictures of me to Facebook, with the natural amount of awkwardness and eye-rolling that request evoked. In the future, systems will probably infer what you look like based on connections. DNA relatives, etc.

edgyquant 3y ago
How accurate do you expect that to me? If I tell you two people have children how well do you think you can predict what that child looks like?
- echelon 3y ago
  1. With the child's complete DNA or key markers, I bet we'll eventually have algorithms for DNA -> adult phenotype 2. With only parental or relative DNA, if we are able to gather an enormous amount of extra information -- social group, gym use, hobbies, dating circles (mate selection probably yields a huge amount of info!), profession, fashion purchase activity, etc. -- then I bet we can get pretty reasonable confidence intervals on a lot of physical characteristics. Height, eye color, eye distance, face shape, lip thickness, etc. etc. Perhaps even elements such as voice timbre, etc. If we had their familial phenotypes, photos and phenotypes of all a person's dating history, then I bet that lets us cut the search space down tremendously. If we add music listening habits and hobbies, I bet we'd also be able to cut it down. I'm hypothesizing that physical appearance might correlate from behaviors, social graph information, and limited DNA information. I have nothing to show for that, however. (I would love to read literature on this topic.) I'd imagine that constructing such a predictive model would require an enormous amount of data. Facebook + 23andme could probably pull it off.