Comment by 4oo4 | Hacker News Reader

Comment by 4oo4 | Hacker News Reader

I think if you have an E5 license there is already thoughtcrime functionality built-in. I remember someone demoing this to me in a Teams user group, and no one seemed to think it was creepy at all. In addition to flagging keywords it also used AI to detect undesirable thoughts and emotions, under the guise of anti-harassment and compliance. Unfortunately I can't remember the name of the feature but I think it might be this:

https://docs.microsoft.com/en-us/microsoft-365/compliance/co...

So I think if Microsoft existed in the world of 1984, they would easily be the preferred tech vendor for IngSoc.

Side note, do you think this would also detect the money laundering and bribery going on within Microsoft itself?

https://www.theverge.com/2022/3/25/22995144/microsoft-foreig...

Side-side note, I think the reason why that is allowed to still keep going on given that the SEC knows about it and that there's ample evidence has to do with national security reasons.

It's extremely troubling that given all this corporate authoritarian AI tech they built that Microsoft is still trying to be the voice of reason about the dangers of AI.

> It's extremely troubling that given all this corporate authoritarian AI tech they built that Microsoft is still trying to be the voice of reason about the dangers of AI.

Just speculating, but this phenomenon could either be explained by 1.) diverse internal opinion; the parts of Microsoft responsible for warning against AI are not the same parts pushing authorarian AI software, or 2.) Moat-building/ladder-pulling; Microsoft is warning people of the danger of _other people's_ AI, but of course you can trust _their_ AI, because they're the ones warning you after all!

Emails aren’t thoughtcrimes. This is nonsense.

Everything in corporate email has always been subject to read by others, there is no expectation of privacy.

As we’ve seen from countless court cases, they range from boring nothingburgers, to evidence of actual crimes.

Right, and this functionality is to punish you for nothingburgers as much as actual crimes. "Leaver" detection is something that is entirely sane for corporations to do but will be abused by the usual suspects in HR to instill fear and to retribute.

Of course it is. But this is how it has always been even without this technology. The cost is lowering because the demand for is pervasive.

I think it's that you can be considered effectively guilty (there's grounds to fire you or take disciplinary action against you) for tripping an ML routine without further evidence or proof, and that it seems more important to have "clean" corporate communication rather than actually act in good faith (as long as bad stuff happens on off channels, no one cares).

Hopefully it doesn't make it outside of the corporate world though.

They want you to know and realize that you’re being monitored so that you take the “bad” communications to where subpoenas can’t get them.

Be good cogs; don’t leave logs.

It already has left the corporate world in China.

Technology, like advanced weapons, doesn’t solve political problems for long, as the other side eventually gets their hands on it.

There is a difference between an investigation under subpoena for example and

An automated process that alerts whomever is chosen as overseers to all possible missteps and misdeeds.

One is a very targeted and conscious effort the other is automated and pervasive everywhere all the time.

Building and selling AI software to do this is also a targeted and conscious effort.

My view is this kind of thing is inevitable and pervasive because there’s a lot of internal risks that companies and governments are worried about. The only solution is to be so valuable that it doesn’t matter.

Ok but on the one hand one choice is blanket pre-approved for all time for everyone with no time expenditure.

The other one is instigated and deliberate at the official request of legal and can take a lot of time.

It’s very different. It instills a climate of untrust. Everyone is “guilty”. In the other scenario everyone is innocent till a specific and circumscribed “matter” is started.

The targeted normally implies that you have a specific incident and a specific person suspected of a misdeed. Building an AI for blanked surveillance is the opposite of that, you aren't looking for a specific incident, have no specific person to suspect. You are basically accusing everyone of being a criminal without any evidence of wrongdoing. Most people don't want to spend their whole lives treated as criminals, anyone who does is free to live in north korea or russia.

> Everything in corporate email has always been subject to read by others, there is no expectation of privacy.

Depends where you work? I expect my work emails to be private.

Reply on news.ycombinator.com