Thursday, June 23, 2016

Tech support and the Turing Test

The Turing Test dates to the very earliest days of computing.  Posed by pioneer Alan Turing, the test says that we will be able to declare the existence of Artificial Intelligence when a person talks to a computer and can't tell whether they are talking to a computer or a person.

Robert Graham thinks that this AI will appear in Tech Support first:
The idea is that in the future, tech support will be replaced AI bots that use natural language processing to answer questions like. But that's what we already have: tech support search text, finds plausible answers they don't understand, and regurgitates them back at us.

In other words, when the Turing Test is finally won, it's going to be in tech support, where a well-designed bot will outperform humans on answering such questions.
Replicating annoying and not particularly helpful people with annoying and not particularly helpful computer programs seems a low bar.

3 comments:

R.K. Brumbelow said...

Most stage 1 and 2 tech support is already based on a flow chart, I fail to see how that is really any different using a human or a voice responsive computer program. The advantage of using people is that for some small percentage of the operators they will learn enough to move up to level 3, but fewer than 10% of all calls ever need to get to a level 3 tech.

Ted said...

...... and those calls that do reach level 3 tend to get the "Microsoft answer"


--- Technically accurate but of no practical value.

R.K. Brumbelow said...

No, not in my experience. But what would I know, I was just a senior trainer, master tech blah blah blah for one of the largest tech support companies back in the day.

Disparaging comments where they are deserved is fine, even encouraged, disparaging comments when they are undeserved relflect on the presenter, not the target.

I wonder how many times you ever engaged a level 3 tech via the phone. The goal set for each stage is to complete at least 75% of all calls, our center shot for 90%. Even at 75%, though, that means only 6.25% even make it to level 3 and none of them would be helped by giving and standard answer, if a standard answer would have worked, it would have been expressed at level 1&2. No level 3 tech is going to waste his stats on repeating what he expects 1&2 to have done. Stats are how we earned bonuses and got raises.