I have read some pretty big files in less, but I don’t know about 20GB.
@stran4 has joined the channel
Who comes up with these and thinks “that’s a good idea”. Likewise, I bet those same people think the inverse feature would be a bad idea.
I work on a competing product. I can’t defend Microsoft’s specific choices of models to build.
In general, though, this kind of surveillance is required by law in certain industries throughout the world. Purposes include discouraging corruption, money laundering, and market abuse.
I also used to work as a monitored individual in one of these industries, and my employer was completely forthright about the nature and purpose of the surveillance. We were even introduced to the compliance officers who would be doing it as part of our onboarding.
A major purpose of AI in this space is actually pro-privacy. Traditional tools for this purpose were very blunt and resulted in compliance officers reading a lot more of people’s communication than was necessary due to spurious alerts. A good document classifier can significantly reduce how much of that is happening through increased precision.
I have to believe the false positive rate is considerably high. Even something like 0.1% in a large org would be insane. Curious how you keep updating the model to reduce them?
It is very high. 0.1% of communication alerts meriting further attention (so, more like 99.9% false positives) would be considered quite good. Communication surveillance is very expensive and time-consuming. I don’t think most firms would do it if they weren’t required to by regulators.
So…it’s like reading employee email randomly?
The traditional approach was to read anything that hit on certain triggers, like using keywords the compliance team deemed suspicious, or communicating with unknown parties. But a company might choose to supplement that with some random sampling.
I should hasten to say, if you work in the USA, you should assume this is already happening as a part of litigation. If you work for or wirh the government, your communications could even be released to the public as part of a freedom of information request. It’s been a long time now that privacy in workplace communication has been taken to be against the interest of civil society. For better or for worse.
(It’s not nearly as bad as that in the EU. I don’t know about other jurisdictions.)
It is a good idea though, isn’t it? Your company emails can be audited by an admin by design. This just makes it less likely that they’d ever pick you out of the bunch.
こんにち。他の日本人はありますか？
Just noting that the work @seanbunderwood is talking about - while the same technology - is about regulation and possibly even state secrets. You sign up for that with the job and no one should be surprised by that.
Secondly, companies do own the hardware and emails. There is no right to privacy with anything done on those machines, emails sent, etc (at least in the States).
But the link I posted was about mining the data to try and guess when an employee is going to start looking for another job or outright quit. I think it’s couched around protecting company IP, etc. And that’s valid, but it’s going to be used by companies for other things that can only negatively impact the employee.
The employer wouldn’t want it turned inward providing the employees with possible insights into things like possible layoffs, closings, etc. Because the only possible outcome is negative and detrimental to the relationship between employer and employee.