4 ms·

so then hire 200 customer service people to handle the queries when the algorithms go awry? surely there is an algorithm they can devise that will identify when

by pyvpx 10y ago

so then hire 200 customer service people to handle the queries when the algorithms go awry? surely there is an algorithm they can devise that will identify when the other algorithms have made mistakes. right? :-)

misingnoglic 10y ago
The problem is that if a human is available to talk to, people will normally tend to that option even if their answer is on the help page.
- pas 10y ago
  Then let those 200 people pick from the email firehose? For example if the sender is a Chrome ext. developer with more than 10 000 downloads?
Merovius 10y ago
Why do you assume that this is not the case? Humans have a failure rate for repetitive tasks between 1 and 10% (depending on who you ask). So even (unrealistically) conservatively and including having two steps of appeal, you would need to assume a minimum false positive rate of at least one in 10K cases (you can even make that a million, if you want, by adding another layer) that get appealed. Now, what do you estimate, how many chrome app store takedown notices, youtube copyright claims, android bugs… whatever, does Google get in any given week? It is literally unavoidable that there will be a significant number of cases that fall through the cracks. And as long as it only takes one of those to post a blog post and get on HN, I don't see much value in increasing the cost for this any further, tbh (I just love how that article completely ignores any statistics when comparing this with the case of Apple btw. Not only is Safari only a tenth of the size, basic statistics should tell you that you won't be the unlucky bastard in both stores simultaneously). The issue isn't algorithms or automation. The issue is, that we can't yet completely replace some jobs with them. And humans suck, especially at scale.

4 ms·

so then hire 200 customer service people to handle the queries when the algorithms go awry? surely there is an algorithm they can devise that will identify when

by pyvpx 10y ago

misingnoglic 10y ago
The problem is that if a human is available to talk to, people will normally tend to that option even if their answer is on the help page.
- pas 10y ago
  Then let those 200 people pick from the email firehose? For example if the sender is a Chrome ext. developer with more than 10 000 downloads?
Merovius 10y ago
Why do you assume that this is not the case? Humans have a failure rate for repetitive tasks between 1 and 10% (depending on who you ask). So even (unrealistically) conservatively and including having two steps of appeal, you would need to assume a minimum false positive rate of at least one in 10K cases (you can even make that a million, if you want, by adding another layer) that get appealed. Now, what do you estimate, how many chrome app store takedown notices, youtube copyright claims, android bugs… whatever, does Google get in any given week? It is literally unavoidable that there will be a significant number of cases that fall through the cracks. And as long as it only takes one of those to post a blog post and get on HN, I don't see much value in increasing the cost for this any further, tbh (I just love how that article completely ignores any statistics when comparing this with the case of Apple btw. Not only is Safari only a tenth of the size, basic statistics should tell you that you won't be the unlucky bastard in both stores simultaneously). The issue isn't algorithms or automation. The issue is, that we can't yet completely replace some jobs with them. And humans suck, especially at scale.