Privacy watchdog urges companies drop emotional analysis AI software

Companies should think twice before deploying AI-powered emotional analysis systems prone to systemic biases and other snafus, the UK's Information Commissioner's Office (ICO) warned this week. Organizations face investigation if they press on and use this sub-par technology that puts people at risk, the watchdog added. …

[Why Not?]

Why do the persist trying to put it back in the box?

It's coming, it won't be perfect at the start but honestly it will save lives and fight crime. Nurture it and see it grow.

[Anonymous Coward]

Re: Why do the persist trying to put it back in the box?

Worse than those mostly useless lie detectors - which still are used as infallible divining rods.

[Anonymous Coward]

Re: Why do the persist trying to put it back in the box?

It won't be used to save lives or fight crime.

It will be used to track your reactions to advertisements and products - a massive, virtual, unpaid, focus group.

[Kane]

Re: Why do the persist trying to put it back in the box?

"it will save lives and fight crime"

Oh yeah? What's next, the Pre Crime Division?

[Anonymous Coward]

Sort of, but not like you think.

This pre-crime division has no pretense of being able to solve the basic incompetence of the rank and file, so they have concluded that the only hope is for the watch commander to tell the knuckle-draggers which corner to stand on at the beginning of their shift if they hope to make an arrest.

They aren't there to prevent crime, after all. If that was the case they would be required to intercede if they witnessed a crime in progress, or protect someone who is being threatened.

[Anonymous Coward]

Apparently the other posters are using the same text analyzers.

I just ran that statement through one and it indicated it is 100 not sarcasm

What do they say about pictures again. Wait, I've forgotten again, they must be worthless.

[Andy Non]

I wonder what companies would

make of the expression on my face while I'm stuck on the telephone for half an hour or more waiting to talk to someone while listening to the same ABBA record in a loop punctuated with "Your call is important to us, please hold the line...".

[Anonymous Coward]

Inscrutable

Such AI would never be able to read beyond the Brit stiff upper lip. The AI is better off looking for greener pastures - perhaps the more transparent US culture?

Hold the the thin red line men! Whatever you do - don't smile - it is a sign of weakness!

[Will Godfrey]

Why HR likes it

No person then becomes responsible for hiring decisions that go seriously (possibly expensively) wrong.

[SoulFireMage]

Stupid buggers

Who are the stupid sods relying on this without a ton of solid evidence that it's worth using?

[Killfalcon]

Re: Stupid buggers

The guy from the sales team said a lot of things that sounded like evidence.

[Khaptain]

Terrible usage of software

Human beings write the software.

Human being can't analyse other human being very effectively.

How can we expect software to be anything less than hyper-sub par for this kind of analysis.

[Killfalcon]

Re: Terrible usage of software

in principle, you could have it learn from the best humas, and then it'll be better than most people.

However odds are that the trainign data will be tagged by the lowest bidders of Fivver or Mechanical Turk, so I'm sure that'll go well.

[Anonymous Coward]

In practice

You can't train it on the best humans because they can't even train other people, and few of them can maintain that status over time and any degree of scrutiny.

The TSA helped create that "Super observer" meme where people presented themselves as human lie detectors. Most of the original cohort have been debunked at this point, and the ones that haven't are just hiding to protect their image. They have since had to reverse their position on the program but some of the same faces are now showing up linked to these fake AI projects.

So an AI could observe your hand selected cohort, but without anyone who can explain how it works you just amplify random traits with a weak statistical correlation until your biases are all that is left. If these systems get the right answer some of the time, it's by accident. The basic methodology is wrong.

[Anonymous Coward]

Please keep it on the IVR systems, though.

The more advanced systems seem to use this - start yelling at the machine and you'll get transferred to a human PRONTO. (But speak calmly and politely to the human.)