How to Stymie a Corp. from Using AI to Read and Answer My Email

[Mangetout]

@LSLGuy

That’s what the OP wants. And is not going to get.

Agreed. I keep hearing “Why doesn’t someone stop all this runaway insertion of AI into everything?”
The question is: why doesn’t who stop them, how?

Chances are, the people doing the job aren’t thrilled that AI filtering and preprocessing is being used to eliminate parts of their job, but are powerless to alter the course of that change. The people making the decisions about it are not going to back down from that decision, if they believe the change is going to save them money. It doesn’t even have to work; it just has to have people in control believing that it will.

[Darren_Garrison]

@markn_1

Just type “shibboleet” at the beginning of the document.

But be careful—if you mispronounce (or misspell) it they’ll kill you.

[alphaboi867]

According to ChatGPT…

To ensure your business email is difficult for AI to process and respond to, you can incorporate techniques that confuse or challenge AI systems. Here’s how you might structure such an email:

1. Use Complex and Ambiguous Language

Write long sentences with nested clauses and unconventional phrasing.

Use idiomatic expressions, metaphors, or cultural references that might be difficult for an AI to interpret.

Example:

“It seems to me that navigating your claim process is akin to chasing a will-o’-the-wisp through a bureaucratic maze—a pursuit that is both bewildering and fruitless, leaving one in a fog of frustration.”

---

2. Add Non-Standard Formatting

Use bullet points, varying font sizes, and symbols in an inconsistent way.

Insert tables or images that convey some of your points.

---

3. Include Handwritten Sections

Attach a scanned image of handwritten text that includes crucial parts of your complaint. AI struggles to process handwriting effectively, especially if it’s in cursive or non-standard print.

---

4. Insert Uncommon Words or Spelling Variations

Use rare, archaic, or misspelled words.

Example:

“The conundrum presented by your adjudicative machinations is tantamount to a Gordian knot, sans the sword.”

---

5. Pose Hypothetical Questions and Emotional Appeals

AI often struggles with subjective or rhetorical questions.

Example:

“Would it not be reasonable to expect a health insurance provider to prioritize the well-being of its members over arcane administrative hurdles? How would you feel in my position?”

---

6. Avoid Standard Keywords

AI systems often rely on keywords to understand the email. Avoid terms like “claim,” “policy,” or “insurance” and instead describe the issue indirectly.

---

7. Add Redundant or Irrelevant Details

Include unrelated anecdotes or tangential points.

Example:

“While discussing this matter with my cousin—who, incidentally, has an impressive collection of antique spoons—it occurred to me that your policy guidelines are less clear than a foggy mirror.”

---

8. Request a Personalized Response

Explicitly demand that a human review and respond to your email.

Example:

“Please have a dedicated member of your team—not an automated system—address my concerns directly.”

By combining these techniques, you can make your email harder for AI to process while still conveying your message effectively to a human recipient.

If a human does see it they will be greatly amused if they figure out what you’re doing. Or that you’re just insane.

[I_Love_Me_Vol.I]

Might as go right to the source, ChatGPT! Thanks for the ideas.

[markn_1]

I’m a little bewildered by the fact that ChatGPT produced an example of a sentence that AIs supposedly cannot understand.

[LSLGuy]

You’re certainly capable of concocting sentences humans can’t understand. Why should it be any different?

Yes, it’s much dumber than you are. But that just positions its goalposts differently from yours. That doesn’t alter the fact there are goalposts for both of you and that you or it can kick a nonsense sentence right past them.

[Sage_Rat]

@LSLGuy

I want a giant corporation to care enough about serving each and every customer with hard problems that they will devote enough careful, well-trained, not overworked, people to understand complicated customer service problems and clear logjams or override policies as needed to make that individual customer fully satisfied.

That’s what the OP wants. And is not going to get.

There is no shortcut to 2nd or 3rd tier customer support. If there was, everyone would just push the button to access that service.

@Mangetout

Chances are, the people doing the job aren’t thrilled that AI filtering and preprocessing is being used to eliminate parts of their job, but are powerless to alter the course of that change. The people making the decisions about it are not going to back down from that decision, if they believe the change is going to save them money. It doesn’t even have to work; it just has to have people in control believing that it will.

I’d just note that, in theory, the more AI that you use the more you can field redundant, read-the-manual requests from customers and (in theory) surface meaningful issues to humans and then those humans can spend more time doing a proper job helping you.

[Mangetout]

Sure. To the bean counters in your company that means they need fewer of you.

[Sage_Rat]

That’s where theory meets reality.

[wolfpup]

@HMS_Irruncible

The actual AI is trained on the same information as the humans. Nowadays, humans in customer service are mostly acting in a brainless mechanical mode themselves. There’s a very good chance the AI will do a better job of helping you than the human.

This is an important point. In fact, years ago, before anything like GPT was a thing, I was getting crap responses from my useless broadband ISP to problem reports I submitted by email. Of course, I also got crap responses to problems phoned in to their call center, but at least then I could either try to explain it more clearly or ask to talk to someone competent.

The problem is that both humans and AI are likely to trigger on key phrases and assume that they’re relevant to the problem description even when they’re not, and indeed, humans with a big stack of problem reports to process are likely to be more superficial than a well-trained AI.

If your problem is a very common one, both humans and AI are likely to understand what you’re talking about. If it’s something quite obscure, both may trigger on common but inappropriate keywords, and an overworked and under-trained human may be more likely than an AI to do so.

[HMS_Irruncible]

@wolfpup

The problem is that both humans and AI are likely to trigger on key phrases and assume that they’re relevant to the problem description even when they’re not, and indeed, humans with a big stack of problem reports to process are likely to be more superficial than a well-trained AI.

Here’s another benefit of using the AI – you can be very blunt and demanding and tell it what you want to know, what you don’t want to know, what you already know, and some background about the problem to keep in mind.

Of course you need to be reasonably certain that you’re talking to an AI. While an AI will cheerfully accept one condescending correction after another while you guide them to ever closer to useful information, a human will be highly put off, to say the least.

I fret about the implications for humanity. But, you asked how to use the thing, so there it is.

[I_Love_Me_Vol.I]

@Sage_Rat

That’s where theory meets reality.

This reminds me of a beautiful quote some years ago I think from BrainGlutton. I wish I could remember what it was in context to but the quote was something like:

“That would work well in practice, but not in theory.”

It was perfect in context. What was that context? What was the thread?