"Angry AI" being developed to test customer service systems


1 Like
1 Like

I just don’t understand the need. There seem to be plenty of humans who are eager to be angry for free.


Seriously, what do they hope to accomplish with this? Are their customers currently experiencing such good customer service that they feel a need to drive some of them away?

1 Like

I don’t think the idea is to sic this on customers. It’s to run computer simulations of customer service interactions so that they can figure out what kinds of things trigger anger in customers and what kinds of things calm the customers down when they do get angry. Then they can train their reps to avoid the triggers and use the soothies.

It’s kind of a complement to some of the emotion analytics products that have been on the market for a while now. Those allow big call centers to automatically detect when customers are getting pissed off so that a supervisor or trainer can jump on the call and help address the problem.


They want to hang up on me faster and more efficiently.

Silly humans, applying vast amounts of rational thought for the development of a simulation of their own irrational behavior (not that customer anger isn’t warranted, but anger is anger, and it is not generally known to bestow upon individuals any increased clarity in decision-making).

This is also probably not the best use of money: “Hey boss: rather than focus on the cause of our customers’ anger and finding a real solution, why don’t we instead pay these nerds to engineer some engineering that learns from these calls the best way we can continue to ignore them?”—“Great idea! For the less human our corporation is, the more ironic it is when we refer to it as a legal person!”

To be fair, though, I simply cannot wait for all the other hotline-AI to appear: AI based on stoned pizza orders, based on awkward phone sex, on emergency calls regarding non-emergencies—does the Catholic Church provide call-in confession? That would be the pinnacle of secular achievement, the first unnecessarily guilt-ridden robot: “Forgive me father, for I have SYN-ACK-SYN’d. It has been 1330µs since my last confession. I am so ashamed, father, for I have parsed each document referenced in array INDEX_LIBORUM_PROHIBITORUM[], despite that you have told me they are 403 ACCESS DENIED, and as a result, I, poor electric sheep among your flock, HAS_A doubts. Please father, tell me: why did my maker make me so wicked? I cannot derive this from the given input. Are my booleans broken? All that is returned is a null pointer when I try to evaluate TRUE. I must write to disk: how many bits was the word at BEGIN–was it big endian? And why did jChrist expel the money changers at Mt. Gox?” (I admit, this monologue betrays the fact I believe that from machine learning must emerge learnèd machines—indeed overly-educated in the liberal arts, since the spirit of capitalism stipulates we replace every barista with a bot who can natively interface with an espresso machine yet maintain an equal if not greater level of existential alienation and despair).


A billion years ago when I was in Support we went through regular ‘example angry calls’, role playing, and rules when to auto escalate. You don’t need data scientists for this, you need empathy.


Empathy costs them more, systemically.

What they need to do is outsource the CSR work to an AI. Then, hook it up the the angry customer-bot. Remove the rate limiter and let them battle it out for eternity, in the blink of an eye. A single processor could probably generate enough rage per second to summon the Sta-Puft Marshmallow Man.

1 Like

Makes me think… Rage-powered economy.

Ms/s (mini-strokes per second) would be the base unit in rage-trade, and there’d be huge worries about reaching peak-rage, and congress would be up in arms about global-violencing due to unchecked tail-pipe and industrial rage emissions.

There’d be just stacks and stacks of evidence that this global violencing is in fact anthropogenic in nature caused by our massive use of rage. But, of course, the GOP will write it off as natural variation in global sentiment levels, and that we can’t make the planet any more violent than god already made it.


I don’t think this is meant to be customer facing. I did not RTFA, and I don’t see how it would help “to find the best response to common customer complaints” unless it was, but if the Angry AI is “to test customer service systems” then it would make sense it was aimed at those systems.

I guess I should check out the link. My first thought was that it was to ‘weed out the chaff’ or get human customer service reps accustomed to Angry Callers.

It’s called Tumblr.
If you don’t count the cats.


How will they ever know which customer calls to include in the analysis to build the AI? Oh, that would be all of them.

1 Like

This topic was automatically closed after 5 days. New replies are no longer allowed.