There’s a chance, a very small chance, that I’ve got this wrong, and that I actually dealt with a human being. There’s a much bigger chance, though, that I dealt with a prototype AI. In either case, the service was absolutely terrible.
So, some background. I have paid-for Google Drive storage. For some weird account-related reason, Google has failed to invoice me for it for the last 3 months, so Gmail now thinks that I’m out of storage and will soon be unable to send emails.
Naturally, I contact support, using the Chat option, because I prefer IM. I explain my problem, and the support rep tells me they’re escalating it, and to expect a response within 24-48 hours. Fair enough, it’s a complex billing query that will take some senior involvement, I’m fine with that.
Few hours later, I’m browsing the web when I come across an interesting project – IPSoft’s Amelia. It’s an AI that can apparently teach itself based on the interactions of colleagues, stored knowledge on intranets (and the internet) and is apparently able to read the emotional state of the end user and react accordingly.
The video talks Amelia up like it’s about the end the tyrannical reign of human-staffed call centers TOMORROW. But then that’s marketing, and marketing is usually BS.
That’s when I spotted this:
Now that sounds suspiciously like the exact encounter I had with the Google support rep earlier. Is there a chance that one of the “multinationals” testing Amelia, that IPSoft coyly refuses to name, might be Google?
Ok, so, simple test. I figure: I’m gonna contact support with a fake issue. It will seem real, and I will provide the answer in the original question. Any human would be able to iron it out in a few minutes.
So I go to the support page, select the Chat option, and populate it thus:
I include my email address there to prove that I’m authenticated into this session, and would hope that this authentication would be leveraged to provide some basic insight into my account standing to the agent I’m about to speak to.
Assuming I reach a real human being:
- They’ll immediately spot my bogus support query (if they’re smart), or
- They will respond to the built-in instruction (if they’re not) and ask me to close PowerPoint, or
- They’ll possess sufficient initiative to look up my account status and find I’m using 19gb/15gb available storage, and so will not be able to upload more files until I purchase more storage.
I hit Submit, and a few seconds later I’m connected with “King R”. And so begins my quest to resolve my support issue.
I should stress that, as I begin this journey, I’m only about 20% confident that I’m dealing with a bot of some sort. Watch how quickly that deteriorates into absolute certainty.
I provided the (bogus) error message right upfront. It’s a common error message in programs that require a file lock in order to read. Chrome requires no such lock – it can read files even though other programs are accessing them.
A human engineer would know this. Or, if they didn’t know, would react immediately to the obvious solution hidden in the question: Close the offending program. That did not happen.
However: I asked a clarification question, when I couldn’t find the Upload button. Humans do this all the time – we say short things when we mean longer things. In context, I’m trying to upload a file, so obviously I need the File Upload button. The agent (or bot, 40% confidence) doesn’t understand that.
That detour seemed unnecessary, since I was able to successfully follow prior instructions. At this point I’d start getting irritated with this agent, but we forge ahead.
We have now managed to manually upload the file – or, not. Now, note the several cues in this exchange so far:
- Based on my first response, PowerPoint is still open. I haven’t been asked to close it yet, which would theoretically resolve the issue “the file is open in another program”
- The error I get is a generic upload error, for a file type that Drive is known to support. The issue cannot be the file.
- If it worked earlier, but it doesn’t work now, and I’m trying the same process (which by inference I am), then the state of SOMETHING has changed between then and now. Which flies over King’s head completely.
- Finally – I’m getting irritated, using words like “I don’t understand” and “it worked” – any human would read that as Annoyance. Not our brave King though!
Note that I helpfully told it that I got an error, provided the error in quotes, and told it that I got “the same error as before”. As a human, you understand that to mean”the last error message I shared”.
Irritation levels = rising. King either can’t keep up, or is reading from a badly-malformed script.
But, we forge ahead.
At this point, huge ass red flag: maildrivetest.
If I were setting up mailboxes to test a new support system for Drive, I’d call it pretty much that. Very, very good chance we’re dealing with a bot here.
Also note, it asks me for a screenshot. Now, in this scenario I’m obviously playing the dolt. Nowhere in the help interface does the means exist for me to take a screenshot. I might be smart enough to push the Print Screen key, but what do I do from there?
No instructions, even when I explicitly ask for it. Had it simply tried Googling for “How to take a screenshot” + “on Windows” (the platform being broadcast in my user agent) it would have resulted in:
That could literally have been provided point-by-point. Of course, King doesn’t give me a chance to take the screenshot, and seems to forget about the screenshot immediately when I tell it that I have no idea how to take the screenshot.
At this point, I get the message “Please give me one moment”, which I think is a smokescreen to quietly connect my session with another agent, either real or virtual. That would be the most seamless way to dial in a real human being and solve this thing once and for all.
But no, it gets funnier.
King’s sibling (let’s call it “Rook”) correctly asks me to close the offending program. However, it misses the huge contextual question that is, “In PowerPoint?”. The answer is obviously Yes – we haven’t dealt with any other programs or files so far.
However, Rook is a bit smarter – it guides me, as a human would, towards the X:
So far so good! Also note the typo, “uploa dit” – that’s the sort of thing a human might do. Bots are capable of responding in annoyingly perfect written English, so the best way to obscure that is to introduce the odd spacing error.
Don’t worry, though – Rook suffers from the same intractable disease as King: Absolute failure to read the human condition. Which we’ll expose in due course.
I get the same error as before, helpfully provided in quotes. It then asks me for the error message, and for any numbers – obviously, you’ll get a more confident search result with an exact error code. This would be the first thing I’d ask for, since those cryptic codes usually yield exact results.
And now, ladies and gentlemen, the Caravan of the Mystical takes a hard left:
Why on earth is the Drive App suddenly a factor? If I were an inexperienced and irritable human, I might take offense at being asked to install additional software all of a sudden. So I do just that.
Observe the first-class EQ module of this bot fail to recognize that I’m getting irritated. That’s the first exclamation mark I’ve used all session, should be a hint, right? Instead it asks me what browser I’m using – a curious question since it can just read my User Agent and find out I’m using Chrome, but OK:
Now it suggests Incognito?
For the uninitiated: Incognito is a safe mode in Chrome that lets you browse anonymously. It forgets everything – your history is wiped when you close the browser, and it doesn’t bring over any information from your regular Chrome browser when it runs. Information like, being logged into Drive.
Any Google employee, whether on-site or remote-outsourced, would likely have gone through basic training on how to use Chrome, and some of the essentials. Someone in Support would have definitely had training in how to use the basic features of Chrome, of which Incognito is one.
Said agent would immediately recognize that switching to Incognito would solve nothing. I’m using the same browser, with the same file, and am still facing the same issue uploading to the same account, which is still over-quota, an issue that neither King nor Rook have considered yet.
I know for a fact that Incognito won’t solve my problem, so I don’t bother trying it. Instead I wait a few seconds and:
And now, ladies and gentlemen, the moment at which the true nature of this digital beast is exposed and understood. Marvel at the deft and utter mishandling of one of the most common scenarios among technologically-challenged users.
I’ll be here all week.
I’ll give you a moment.
Ok, here are the critical mistakes that Rook has made so far:
- Asked me to try Incognito in the first place
- Does not understand that a password is required to authenticate
- Does not understand that I do not have the password
- Does not understand that I simply could have asked my wife
At this point, as both the fictional high-maintenance user I’m roleplaying, and the actual technically-savvy user frustrated at the pace of AI development, I pitch it a wide.
Note how I did two things in that sentence:
- Ended the conversation – “don’t have time”
- Solved my own problem – “Dropbox instead”
This is case closed. This is when any human support tech would apologize that they were unable to solve the issue, and wish me a good day. Then they’d close the window, exhale, and remind themselves that it’s only a few days left until Friday.
Rook don’t give a fuck.
So at this point, you might share in my frustration on many levels:
- It’s not my job to troubleshoot your software
- I’m clearly annoyed at you, and have my own alternative
- I do not wish to continue this conversation
Which Rook picks up on about half a second later, without giving me a chance to respond:
Wow. So it did two things at once:
- Read that I was upset and tried to calm me down (good)
- Also read that I said I don’t have time anymore and wanted to end the conversation
We’re 110% into badly-written deep-learning bot territory at this point. I’ve got what I came for – no human agent I’ve ever met would handle such a simple support request so badly. So I decide to leave an ambiguous message for the developers of whoever put this AI together.
That of course is not a reference to Drive. At this point I’m hoping that a human is either watching, or will later review this log and realize that I’m referring to the bot AI, not Drive.
Note, again, that I’m giving Rook an out: I’m giving it the keywords “not getting the help I need”, which is generally what humans say when they need their request escalated.
But there are no supervisors. There are no escalation paths.
There is only Rook.
The single most offensively-human line I’ve heard in this entire exchange yet.
I know it doesn’t know the answer at this point. Rook knows that I don’t know the answer. At this point, even the most basic SQL query would know to take it upstairs.
But this is Rook. And Rook does not surrender.
And as I feel the hopes and dreams of a thousand Ray Kurzweils melt away in the sheer furnace of Rook’s stubborn inability to solve a question that anyone, possibly even a man off the street could solve, I call it a night.
I do not wait for Rook’s response.