Why Sam Altman got fired

Okay this post is not really about Sam Altman. But it is about the current state of generative AI and how it is being used.

AI is already out of control, and some important thinkers have been talking about this for awhile.

The organization where I work has a reading room, and we have several subscriptions to print magazines. Among them is National Geographic. According to Wikipedia, Disney acquired a controlling interest in NatGeo in 2019.

Some issues of the magazine contain a poster folded within the magazine. The cover of the October 2023 issue said there was a poster with a map of the night sky inside.

The magazine we received did not contain a poster.

The magazine is delivered inside a paper sleeve. The sleeve was intact and unopened, and the magazine itself had nothing wrong with it. There were no signs of damage or tampering. The poster simply wasn't there.

So I went to the NatGeo website and reported the problem using a contact form. There was a field to include our account number, which I obtained from address block on the sleeve. Here is the message I sent:

We received the October 2023 issue, and the cover says there is a poster
inside. But there is no poster in the magazine. Please send us the poster or send a replacement magazine that contains the poster.
Thank you


Here is the response I received three days later:

Thank you for contacting National Geographic Magazine.

We are sorry to hear you have received a damage copy. Unfortunately, the
issues in question are now out of print and no longer available.

To compensate you, we are extending your account. Your account is now
valid through the issue dated September 2025

Again, please accept our apologies.

If we can be of further assistance, please let us know. To ensure your
future concerns are handled in a timely fashion, please include all
previous email correspondence.

Thank you,
Angeline
National Geographic Magazine Customer Service
natgeomagazine@cdsfulfillment.com


There are multiple red flags here. The phrase damage copy is ungrammatical. The proper use is damaged copy. The next sentence says that "the issues in question are now out of print," using the plural form, even though the previous sentence appears to acknowledge that we only reported a problem with one issue.

But the real problem is that the issue in question was not out of print. It was the current issue, and I quickly discovered that it was offered for purchase as a single copy on the NatGeo website. But the fact that it was offered did not necessarily mean that it was actually available. With modern web-based ordering systems, sometimes you don't find out that an item is out of stock until you actually try to order it.

And that, of course, is exactly what I did. I ordered a single copy of the October 2023 issue, and paid $8.00 for it using a credit card. And the order was accepted.

I then sent the following reply to their message:

In your response below, you stated that "the issues in question are now out of print and no longer available."

This is not true. The October 2023 issue is available for purchase on your website at https://ngsingleissues.nationalgeographic.com/national-geographic/2023

We placed an order for the October 2023 issue at a cost of $8.00. We have attached a screenshot of your website, and we have also attached a copy of the e-mail confirmation that we received. These documents clearly show that the October 2023 issue of National Geographic is NOT out of print, and that this issue IS AVAILABLE.

Your response to our inquiry contains false and inaccurate information. This is unacceptable. THIS DISPUTE IS NOT RESOLVED.

To resolve this dispute you must immediately take one of the following actions:

(1) Send us a check for $8.00, payable to
[redacted], to reimburse us for the cost of the defective issue, OR
(2) Extend our subscription by an additional 12 months at no cost.

We are requesting that you respond to this e-mail message by 5:00 PM eastern time in the USA on Monday, October 30, 2023.

If this dispute is not resolved to our satisfaction, we will commence the dispute resolution process set forth in Section 8 of your Terms of Use.

Please be aware that by providing false and inaccurate information in your response to our inquiry, you have violated federal and state consumer protection laws. If this dispute is not resolved to our satisfaction, we will file a complaint with the US Federal Trade Commission and the Michigan Attorney General.


Section 8 of their TOS provides for arbitration.

This time the response took four days. Here's what came back:

Thank you for contacting National Geographic Magazine.

We are sorry to hear you have missed an issue. We work very hard to deliver our magazines in a timely manner, but, unfortunately, problems do sometimes arise.

To compensate you, we are sending a replacement for October 2023 issue which will arrive within two to four weeks for U.S. and Canadian addresses and within six to eight weeks for international addresses.

Please accept our apologies for any inconvenience or concern this matter has caused you.

If we can be of further assistance, please let us know. To ensure your future concerns are handled in a timely fashion, please include all previous email correspondence.

Thank you,
Masyl
National Geographic Magazine Customer Service
natgeomagazine@cdsfulfillment.com


Once again, there are multiple problems here. The message refers to a "missed" issue, which is not what happened. And the message completely fails to address the fact that I already ordered a replacement, and fails to address my request for compensation.

And this message came from Masyl, while the previous message came from Angeline.

It is certainly possible that the process used to formulate these messages did not involve artificial intelligence as that term is commonly used today. But if it was not AI, then it was something else that actually resembles a primitive form of AI. If there was direct human involvement in the creation of these messages, then they were sent by representatives who are required to select from a library of canned responses, and are they not permitted to use any other text. If they can't find a response that is appropriate and actually addresses the customer's problem, then they are trained to either (1) select a response from the library that is as close as possible and is likely to resolve the problem, or (2) escalate the case to someone with a greater level of authority, who is permitted to write original text.

But if they escalate too many cases, that hurts their performance metrics, implying that they are not good at their job. And that could eventually get them fired. So there there is massive pressure to choose from one of the available canned responses, even if none of them are a good fit.

These reps are required to use an algorithm that is essentially a low-grade form of AI.

I then sent the following reply:

Why did Angeline, in her e-mail message of October 26, claim that "the issues in question are now out of print and no longer available"?

This time the response took eight days. Here's what came back:

Thank you for contacting National Geographic Magazine.

Please accept our apologies for any inconvenience.

A copy of this correspondence is being forwarded on to management, for quality purposes.

If we can be of further assistance, please let us know. To ensure your future concerns are handled in a timely fashion, please include all previous email correspondence.

Thank you,
Sherry
National Geographic Magazine Customer Service
natgeomagazine@cdsfulfillment.com


This response came from Sherry. It was at this point that I may have finally broken through the AI, or the algorithm, and begun communicating with a human agent who had the authority to reply with something other than a copy-and-paste response. But they still completely failed to address my request for compensation, and they still did not even appear to recognize that I had already ordered a replacement copy.

So I sent the following reply:

Thank you for your reply.

This matter is NOT resolved. This case should NOT be closed. I am NOT satisfied with the way this has been handled, and you have NOT answered my questions.

Please have a member of your management team call me at xxx-xxx-xxxx. If you call and I am not available, you must leave a message with your first and last name and a regular telephone number where I can call you. I will not return your call if the number is a toll-free number that rings into a customer service queue at a contact center. If you do not call to discuss this matter, we will begin formal dispute resolution proceedings pursuant to your terms of service.


This time the response took 13 days. And most tellingly, the response came from Sherry—the author of the previous response--and not from whatever rep just happened to be working the e-mail queue at that point.

Thank you for contacting National Geographic Magazine.
Please accept my apologies for any confusion.

If you not yet received the replacement copy we mailed to you, we do
request that you allow up to 4 weeks for delivery. The issue should be
received this week.

Your complaint, that one of our Customer Service Representative provided
you with incorrect information, is being addressed.

We are not able to honor your request for a free one year subscription.
If you still wish to cancel your subscription, please let us know, and a
prorated refund for the unserved issues left on your order will be
processed.

If we can be of further assistance, please let us know. To ensure your
future concerns are handled in a timely fashion, please include all
previous email correspondence.

Thank you,
Sherry
National Geographic Magazine Customer Service
natgeomagazine@cdsfulfillment.com


Here we are finally beginning to have some meaningful communication, in which she flatly informs me that they will not provide a free one year subscription, and says that my complaint about the false information in the original response "is being addressed."

And yet. The next sentence says "If you still wish to cancel your subscription, please let us know."

Excuse me? I never said anything about canceling our subscription. And this message still fails completely to address the fact that I paid for a replacement copy and requested a refund of $8.00.

So I sent yet another response:

Sherry--

This matter is NOT resolved, and this case should NOT be closed.

On November 7, 2023, I requested that you have a member of your management team call me at xxx-xxx-xxxx. I indicated that if you call and I am not available, you need to leave a message with your full name and a direct telephone number--not a toll-free number that is answered at a call center--where I can call you back.

Do you plan to honor my request to discuss this matter by telephone?

If so, please reply to this e-mail message and provide your direct telephone number.

You have failed to address our concerns and answer our questions. We are providing an opportunity to resolve this dispute informally through telephone communication. If you do not agree to discuss this matter with me by telephone, we will commence the dispute resolution procedure identified in your terms and conditions, which includes our right to an in-person arbitration hearing.

Please note that we do NOT intend to cancel our subscription at this time, and there was never any reference to cancellation in any of our previous communications regarding this matter.


The next response came in less than 24 hours. It was only at this point that someone--probably Sherry--took the time to read the entire thread of communication and try to understand what I was complaining about, and what I wanted.

Thank you for contacting National Geographic Magazine.

Please accept our apologies for any confusion.

A refund has been processed for your purchase of the October 2023 issue.

Your request to have a manager contact you directly by telephone was
forwarded on to management.

If we can be of further assistance, please let us know. To ensure your
future concerns are handled in a timely fashion, please include all
previous email correspondence.

Thank you,
Sherry
National Geographic Magazine Customer Service
natgeomagazine@cdsfulfillment.com


Some of those who are still reading are probably thinking okay, this guy's completely fucking nuts.

He's going to go to arbitration over a missing poster? Over a magazine that cost $8.00? And file a complaint with the FTC and the state attorney general??

Well, maybe. I don't think I would have actually paid fees to file an arbitration case. But if you win the arbitration, the other party usually has to pay all the fees. And their terms require a written notice of dispute—on paper by certified mail for God's sakebefore filing an arbitration case. Sending that notice probably would have gotten me on a phone call with a human paralegal. And I wanted to make a point. They fucked up, and I wanted them to acknowledge it and apologize. I wanted my $8.00 back, and I wanted someone to understand just how incredibly bad the customer service experience can be when you use AI or algorithms like this. And I also would have demanded reimbursement for the cost of sending the certified mail.

Yes, this was, relatively speaking, a lot of time and energy over something of very little significance. Yes, paying for a single copy was probably not the best approach, and in any event I could have disputed the $8.00 charge through my credit card issuer.

But I have too much time on my hands, and this was an interesting experiment. It is pretty clear that the representatives who responded prior to Sherry were either nonhuman, or they were humans who were heavily constrained by an algorithm that was simply not equipped to address my request.

No, I am not going to pursue this any further, assuming that I do in fact get a refund of $8.00. If I get a phone call from management, it might be interesting. The original question remains unanswered: Why were we told that the issue was out of print?

Human error? Seriously? How much human involvement was there in the original response?

The biggest danger with these systems is that in some cases, there is no human available on the back end. Or even there is, it may be impossible to "break through" the AI and actually get to someone who can interact with the customer without the algorithmic constraints.

Some systems are hybrid human/AI models, where an AI program reads the customer's message, and suggests a few responses, with a human agent selecting the one that seems most appropriate. Some of these systems use very low-paid workers who have very limited proficiency in English.

The offer to cancel our subscription actually would have made sense without the word still. The use of that word implies that I had previously expressed a desire to cancel, which I did not do. And that suggests that even Sherry probably copied and pasted some parts of her final response to me.

My experience is a relatively harmless, benign example of an algorithm that simply doesn't work when the problem doesn't fit into a predefined category.

What's going to happen when someone in charge at a very small, underfunded local agency gets the brilliant idea that they can save money and improve service and response times by using AI to take 911 calls?

People will die.


There are already many documented cases of innocent people getting arrested and even prosecuted on the basis of facial recognition programs.

And lots of documented accidents involving self-driving vehicles.

This is only the tip of the iceberg.

AI is already out of control.
OK, so here is a related experience I had recently. My Microsoft lap top (or whatever the damn thing is now called) was misbehaving and I couldn't figure out how to fix it. Seemed to be a software bug. Contacted Microsoft technical support. "They" first asked for all manner of irrelevant information. I got annoyed as I wanted to get to the task at hand rather than waste a few minutes entering clearly unrelated shit. I told the "person" on the other end to fuck off, and accused "it" of being a BOT. The BOT protested and said "I am a human." I responded with, "Liar!". I explained what I needed to know how to fix. Short and sweet. BOT told me not to use profanity, but said "it" would help me. So I said "OK, Do it then. You don't need to know my mothers maiden name to fix this problem." They persisted to annoy me with irrelevant platitudes and other shit, just as though reading from a script, in the nature of the script that "Bob in Mumbai" will read to you. At this point I was absolutely convinced, because of the canned nature of the responses I was getting, that there was absolutely no human on the other end. I explained very clearly and with great specificity the problem I was experiencing. "They" suggested doing the usual, that thing where you ask the computer to "trouble shoot" and you get nowhere. I told "them" I already did that "thing", and got nowhere. Finally I decided to go all the way and just insult the hell out of whatever "it" was at the other end. That did it. Suddenly "it" said, "Let me see if I can fix your problem. But first I need your permission to take over your computer." I said, "have at it asshole". So, I did what they told me to do. Then I just sat there and watched the screens flash by at light speed, way too fast for any human to follow. And bingo everything worked as it was supposed to. Bang. Done.

Obviously I was not dealing with a human during the repair as no human could have made that series of alterations to the software at such blazing speed.

My advice to Microsoft would be to cut out the crap, stop fishing for superfluous information not needed. Be up-front about the true nature of the "thing" on the other end, and once the user has given a clear description of the problem, immediately ask for permission to take over the users computer and fix the damn problem. If "they" have the ability to do this, and "they" obviously do, then for god sake just do it. If "they" don't do this for you, then insult the hell out of "they" in an attempt to invoke the useful response I finally dragged out of "they".

What an interesting three quarters of a century lies ahead of us! I'll be dead in ten years, but the rest of you blokes can look forward to many annoying "conversations" with integrated circuits.
 
The BOT protested and said "I am a human."

Some experts in the field of AI have suggested that laws or internal self-regulatory policies or codes should include a requirement that any AI program adequately disclose that it is not human.

However, that is certainly not the current state of affairs.

We frequent get robocalls from a highly deceptive organization that solicits donations that supposedly support police and firefighters. The organization is a PAC (polictical action committee). It is not a charitable organization, and donations are not tax-deductible under US federal law.

But they never claim that they are a charitable entity, nor do they claim that donations are tax deductible.

They use robot call systems that are guided by a human. What you hear is a series of recorded messages, asking for a donation and then prompting for name and credit card number.

There is a human on the back end, selecting which messages you hear.

I have experimented with this system extensively, because we get a lot of these calls. On many of them, the call ends with the artificial agent saying they will put us on the do-not-call list, but the calls keep coming.

When I ask, "Am I speaking with a live agent?" or "Are you a human representative?", the response I get is "There's a real person here."

But that doesn't answer the question that I asked.

Only once or twice have I managed to get through to a real live agent. To do that, you have to express a desire to make a donation, and then make something go wrong that they believe can be fixed if they take over the call and speak to you live. This is not easy to do, and I have only pulled it off a couple times. At least once I did it by providing a nonfunctional credit card number. But it takes a lot of patience, because they will use the bot to ask for the card number two or three times, and then use the bot to ask you to use a different card. And sometimes they give up rather than taking over the call, and they just hit the recording that claims they will put you on the do-not-call list.

On the rare occasions that I have gotten through to a live rep, I almost immediately say something really obscene. And at that point they usually hang up.

For me it's a form of entertainment. It's a fun, challenging game to see if I can get the AI to hand the call over to a human.
 
OK, so here is a related experience I had recently. My Microsoft lap top (or whatever the damn thing is now called) was misbehaving and I couldn't figure out how to fix it. Seemed to be a software bug. Contacted Microsoft technical support. "They" first asked for all manner of irrelevant information. I got annoyed as I wanted to get to the task at hand rather than waste a few minutes entering clearly unrelated shit. I told the "person" on the other end to fuck off, and accused "it" of being a BOT. The BOT protested and said "I am a human." I responded with, "Liar!". I explained what I needed to know how to fix. Short and sweet. BOT told me not to use profanity, but said "it" would help me. So I said "OK, Do it then. You don't need to know my mothers maiden name to fix this problem." They persisted to annoy me with irrelevant platitudes and other shit, just as though reading from a script, in the nature of the script that "Bob in Mumbai" will read to you. At this point I was absolutely convinced, because of the canned nature of the responses I was getting, that there was absolutely no human on the other end. I explained very clearly and with great specificity the problem I was experiencing. "They" suggested doing the usual, that thing where you ask the computer to "trouble shoot" and you get nowhere. I told "them" I already did that "thing", and got nowhere. Finally I decided to go all the way and just insult the hell out of whatever "it" was at the other end. That did it. Suddenly "it" said, "Let me see if I can fix your problem. But first I need your permission to take over your computer." I said, "have at it asshole". So, I did what they told me to do. Then I just sat there and watched the screens flash by at light speed, way too fast for any human to follow. And bingo everything worked as it was supposed to. Bang. Done.

Obviously I was not dealing with a human during the repair as no human could have made that series of alterations to the software at such blazing speed.

My advice to Microsoft would be to cut out the crap, stop fishing for superfluous information not needed. Be up-front about the true nature of the "thing" on the other end, and once the user has given a clear description of the problem, immediately ask for permission to take over the users computer and fix the damn problem. If "they" have the ability to do this, and "they" obviously do, then for god sake just do it. If "they" don't do this for you, then insult the hell out of "they" in an attempt to invoke the useful response I finally dragged out of "they".

What an interesting three quarters of a century lies ahead of us! I'll be dead in ten years, but the rest of you blokes can look forward to many annoying "conversations" with integrated circuits.
I have a similar painful experience with MSFT. I upgraded my MacBook Pro. Excel stopped working after I ported my software over. Purchased a new Office license. Had difficulty, called Tech support, a BOT, told me to do it online. Online help wasn't helpful, and asked me to call tech for help. It was an endless do loop.

Finally asked Google, directed me to a non MSFT site. Done in 5 min.

MSFT hates Mac users.
 
I have a similar painful experience with MSFT. I upgraded my MacBook Pro. Excel stopped working after I ported my software over. Purchased a new Office license. Had difficulty, called Tech support, a BOT, told me to do it online. Online help wasn't helpful, and asked me to call tech for help. It was an endless do loop.

Finally asked Google, directed me to a non MSFT site. Done in 5 min.

MSFT hates Mac users.
It would seem so.
 
The original post was not written by AI. I am the writer. I am open to constructive criticism, but I don't think my post is filled with grammatical errors. There might be one or two typos. You are welcome to point out any specific errors.

The text of the e-mail messages from NatGeo is quoted verbatim, and the errors are part of the whole discussion.
Well done.

And then too funny, getting corrected by someone who capitalizes Please in the middle of a sentence.
 
I have a similar painful experience with MSFT. I upgraded my MacBook Pro. Excel stopped working after I ported my software over. Purchased a new Office license. Had difficulty, called Tech support, a BOT, told me to do it online. Online help wasn't helpful, and asked me to call tech for help. It was an endless do loop.

Finally asked Google, directed me to a non MSFT site. Done in 5 min.

MSFT hates Mac users.
Don't blame Sam hehe:

Early life and education
Samuel Harris Altman was born on April 22, 1985, in Chicago, Illinois,[3] into a Jewish family,[4] and grew up in St. Louis, Missouri. His mother is a dermatologist. At the age of eight he received his first computer, an Apple Macintosh.[5] He attended John Burroughs School, a private school in Ladue, Missouri. In 2005, after one year at Stanford University studying computer science, he dropped out without earning a bachelor's degree.[6]
 
Back
Top