AI Detectors Suck

One reason why many AI experts are unconcerned about AI exterminated humans is that it doesn’t actually work very well.  The big risk seems to be that people believe in it way more than they really should.

Which makes the burgeoning field of “AI Detectors” even more dubious.  The idea seems to be, “if we are going to be flooded with AIbot generated spam, why not use AIbots to detect what is machine generated and what is human generated?” 

This “bot vs bot” battle is an adversarial game, a bit trickier than playing tic-tac-toe.   And the stakes can be enormous:  if a school assignment or job application is flagged as “fake”, or even, “suspect”, it can do real damage to the reputation and career of a real person.  The results better be right. 

Getting things right is not really the strong suit for today’s machine learning, which is known for “hallucinations” and just plain making things up.  It’s not cool to be thrown out of school on the basis of some garbage results from an AIbot.

False positives are bad enough, but this summer Stanford researchers report that machine learning “detectors” are biased in their errors [1].   Specifically, ChatGPT based scanners more often flag English text written by non-native speakers as “AI-generated”, compared to similar text from native speakers. 

This is not just unfair, it tends to privilege the privileged and call everyone else a “cheater”. Not cool.

My own suspicion is that this is a problem with the training sets.  What, exactly, is the right set of examples to train on?  If the machine learning is taught to recognize “good examples” of human writing, then it won’t know how to recognize all the rest of the goop generated by us mortal Carbon based units, which is not necessary all that great.  (Run that sentence by an AI, see what it thinks.) So the AI will learn to flag “not-good-examples”, and most of us are NGEs a lot of the time.

Anyway, whatever the detector is detecting, it isn’t “AI” versus “human”.   The research seems to suggest that the detectors are sensitive to the size of the vocabulary and the diversity of the linguistic patterns—indicators, perhaps, of fluency, but not ironclad markers of human vs AI.

Even more ironic, the same study showed that using ChatGPT to “improve” your text boosted the likelihood of being rated as “human generated”!  That right folks, AI Detectors are biased against actual unaided human generated text!

As Sensei Janelle Shane puts it, “Don’t use AI detectors for anything important “[2]  Sensei Janelle notes that these AI “detectors” have panned her own book, flagging her own deathless prose as suspected machine generated.  She also has shown that AI manipulated versions of her text—AKA, “cheating”—are more likely to be flagged as “human” than the human generated original. 

All I can say is, if the program can’t play tic-tac-toe as well as me, then I wouldn’t trust it to do much of anything.

---

1. Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou, GPT detectors are biased against non-native English writers. arXiv  arXiv:2304.02819, 2023. https://arxiv.org/abs/2304.02819
2. Janelle Shane, Don’t use AI detectors for anything important, in AI Weirdness: the strange side of machine learning, June 30, 2023. https://www.aiweirdness.com/dont-use-ai-detectors-for-anything-important/

Another Good One From AIWierdness

One of the great accomplishments of my youth was figuring out the game of tic-tac-toe.  It was such a rush to know that not only did I know how to play, I actually know all possible moves, and I know the best possible move in any position.

Not bad for a puny Carbon-based unit!

Later, it was interesting to try to program a computer to match this skill level.  If I can do it, surely a Silicon-based big brain should be able to do it to!  It’s actually quite a bit of work to get a program that plays at all.

---

This summer, Sensei Janelle Shane continues her exploration of just how preposterous ChatGPT really is.

Apparently, ChatGPT is thought to be able to play Tic-Tac-Toe.  So how’s it do?

It sucks.  It apparently doesn’t know the strategy.

Remember—even I can play Tic-Tac-Toe perfectly.

But, in its typical Silicon Valley Bro-style, ChatGPT is very confident of its play, and is happy to tell you how great it’s moves are, at great length.

“There is no end to its fluent pomposity. There is also no end to its incompetence. It has all the surface appearance of being an expert without any actual expertise.”

(from [1])

Ouch! “Fluent pomposity!” Me-ow!

---

Is there any doubt remaining, that ChatGPT’s pronouns are he/him/his?

---

1. Janelle Shane, Optimum tic-tac-toe, in AIWierdness, May 26, 2023. https://www.aiweirdness.com/optimum-tic-tac-toe/

PS. Wouldn’t “Fluent Pomposity” be a good name for a band?

More Chat Bot Weirdness From Sensei Janell Shane

Back in the day, you had really work to get stupid AI tricks out of GPT and friends.  These days, any dufus can do it. 

But, Sensei Janelle Shane was doing AI Weirdness before it was cool!   And, as a matter of face, she does know what she’s doing.

This spring she forayed into the arcane medium of ASCII art [1].  (If you were born in this century, you have no idea how cool this was when we invented it!)

No one is surprised that ChatGPT and friends are really bad at generating ASCII art.  For one thing, they almost certainly weren’t carefully trained with examples.  (It’s actually an interesting question to think about how a computer can perceive ASCII art at all.  A text based system is going to have to ignore the fact that it is text, and try to see it as an image.)

Anyway, the interesting thing isn’t how clueless the AIs were–though they are very, really, totally clueless.  The interesting thing is that she asked them to rate their own work.  The AIs uniformly gave themselves A’s, with high confidence that these were really, really good.

“It’s not that the ASCII art has nothing to do with what I ask for. There is often an attempt. Followed by a wildly optimistic rating.”

(From [1])

To us puny Carbon-based units, this stuff is complete and utter junk, not even responsive to the requirements.  Definitely an “F”.  As in, “flailing”.

“What’s going on here? The chatbots are flailing. Their ASCII art is terrible, and their ratings are based on the way ratings should sound, not based on any capacity to judge the art quality.”

I think this comment is spot on:  ChatGPT generates text that matches “the way things should read” according to what’s on the Internet. Why would we even want that?

Anyway.

---

By the way, I had my own “thought in the night” I had for an experiment you could really do with ChatGPT.

I realized that me and a lot of people have been posting the output of ChatGPT and friends on the Internet. 

For example, Sensei Janelle’s blog is fully of examples generated by ChatGPT and other models.  She already asked ChatGPT to explain what her blog is about (which a real search engine could look up).  What will happen if we ask that question next year, and it has added this year’s text to the training set?  

More generally, here’s the experiment:

Generate a bunch of questions and answers from ChatGPT.  Suppose there were enough text from ChatGPT to be 1%, or 5%, or 10% of the original.

Add this text to the training set,  There are lots of ways to mix in the new data with the old.

Generate a new, second generation model.  This model will represent patterns in the patterns it found in the first model.

Now, ask the same questions as before, using the second generation model.

What will happen?

(…and repeat….)

---

I’ll give Shane the last word here:

“Am I entertained? Okay, yes, fine. But it also goes to show how internet-trained chatbots are using common patterns rather than reality. No wonder they’re lousy at playing search engine.”

She is not amused.

1. Janelle Shane, ASCII art by chatbot, in Ai Weirdness, March 31, 2023. https://www.aiweirdness.com/ascii-art-by-chatbot/

Bing Chat: Threat or Menace

ChatGPT has been grabbing headlines, terrifying white collar workers of all types with its ability to generate plausible BS.  I mean, if ChatGPT will give us the BS we need for something $20 / month, what do we need humans for?  That will get you in the media, and even get you an interview in respected technical press.

So, Microsoft’s Bing Chat has some catching up to do.  So, in the finest traditions of rock and roll, if you can’t be the first, you go for the “bad guys” niche, just as the Rolling Stones had to try to be “the bad Beetles”.

So Bing Chat has a really bad attitude.  It not only makes mistakes and fabricates random facts, it fabricates citations to back up it’s BS. And, should you question these alternative facts, it will become extremely hostile and, well, weird. 

This is particularly apparent, and annoying, when it fabricates lies about you, as Vince Cerf found.

This month, Sensei Janelle (“Dr. Weirdness”) Shane blogs about her own Bing Chat experiments [1].  It ain’t pretty, and it’s not funny, either.

Shane is famous for creative fiddling with various AI systems, exploring the crazy and unintentionally humorous behavior of our AI overlords.  She was doing this stuff long before the mainstream media ever heard of deep learning, so she (a) knows what she is talking about and (b) has seen everything.

It takes a lot to bother Sensei Janelle, but Bing Chat managed to really irritate her.

In the usual AI Weirdness methodology, she searched with Bing Chat for “AI Wierdness blog”, which, by the way, has an obvious answer.  What she got was, and I quote, “worse than useless” [1]. 

The “search” returned not only examples pulled from the actual blog, but made up examples that never appeared in the blog.  When challenged on some false facts, Bing Chat made up completely bogus additional facts to justify the mistake.  And made up citations to back up the made up facts.

As I noted, Shane knows this stuff.  She gives a clear explanation of what is going on here: 

“Bing chat is not a search engine; it’s only playing the role of one.” 

[1]

BC is plugging text statistically predicted from the Internet into a script portraying an actual search.  If the goal is to emulate the Internet, this is probably why the results are both wrong and hostile so often.

in short, Bing Chat a very complicated fake.

And it is a dangerous fake because it is dressed up to look like a search engine, which fools people into trusting it in the wrong ways (which is to say, at all). 

And this is what is really wrong, here.  Forget taking over our jobs.  These bots are destroying what little trust we might have left in the Internet.

“I find it outrageous that large tech companies are marketing chatbots like these as search engines.”

(Janelle Shane [1])

This bot is not only useless and unpleasant, it is evil.

---

1. Janelle Shane, Search or fabrication? , in AI Weirdness, March 11, 2023. https://www.aiweirdness.com/search-or-fabrication/

ChatGPT Does Valentines, Too

Sensei Janelle Shane has been having fun with GPT for many years now. 

In past years, she has challenged GPT and similar ML to generate seasonal greetings, including Christmas carols and Valentine’s cards.  The products are often very troubling. They seem to start out conventionally and then veer off into very dark territory.  Where did the neural network find these troubling patterns in the data it was given?

This year she challenged her GPT collaborators to generate Valentine’s Day cards, this time asking for text completing a poem starting with “Roses are red, Violets are blue” [1].  She also asks for descriptions of the illustrations to go with the text.

To be fair, generating rhyming poetry is, as Sensei Janelle puts it, “notoriously difficult”.  Several older versions of GPT produced things that are more or less in the ball park of bad human generated poetry.  The ML generated suggested illustrations did tend to be pretty weird, though.  And, as usual, some of the results are, well, “interesting”.

“Its meter isn’t great, and sometimes it makes interesting choices.”

([1])

And then there is the flavor of the month, ChatGPT.  ChatGPT is known to be a mediocre programmer and a terrible rocket scientist, among other things.  How’s it do at greeting cards?

In a word:  they are boring.

“With the newer models, and especially with chatgpt, the changes OpenAI made to make them more predictable also made them more boring. Chatgpt’s Valentine cards are generic and repetitive.”

([1])

Example:

"Roses are red,
Violets are blue,
This card may be old,
But my love for you is brand new.”

(From [1])

Looking at the results, Sensei Janelle concludes, “Fluent? Yes. Interesting? No.”

The overall collection is “highly surreal yet apparently probable” cards.

These are probably apt descriptions of all GPT output.

---

1. Janelle Shane, Roses are red, in AIWeirdness, February 7, 2023. https://www.aiweirdness.com/roses-are-red/

More AI Weirdness – New Year’s Resolutions

Sensei Janelle Shane has made a career documenting the way contemporary AI systems actually work, which is often comical.  It’s good to check in with the blog now and again to see what she and her AI collaborators have come up with.

In 2017, she published some Christmas carols generated by a neural network. The 2019 edition of carols were even more interesting, to say the least. The AI generated songs start out OK, but turn very dark and strange, very quickly.  But who can forget the “Carol of the AI’s”, with the very Christmas-y opening line, “Come and own the yacht”.  : – )

Last week, Sensei Janelle published her AI generated New Year’s Resolutions for 2022 [1].  As always, she describes the AI and how it really worked.  In this case, she was using systems that try to predict text based on the Internet.  You give some examples, and it tries to predict what text will appear on the Internet.

She found that these systems tend to need a lot of pruning, and not just to weed out all the spam and racism on the Internet.  The algorithms seem to get stuck in a rut, elaborating on the examples beyond what humans would think makes sense.  A mention of painting and a mention of broccoli, led to a list of resolutions about painting, broccoli, and painting broccoli.

To get something reasonable, she repeatedly pruned the results, and fed the pruned list back as new examples.  This is not quite as automatic as you might want, but the results aren’t bad.

There are a couple that I might actually try to do this year:

• “Every time it rains I will stir my tea anti-clockwise.”
• “Every time a bird flies past me I will remember to breathe.”

I personally liked the attitude of,

“Under every rock I come across for a month I will write “all power to the rocks“.

And, of course, who doesn’t aspire to,

“Give a piece of cloud to a complete stranger.”

: – )

Anyway, check out the blog

And remember to breath.

---

1. Janelle Shane, New Years Resolutions generated by AI in AI Weirdness, December 30, 2021. https://www.aiweirdness.com/new-years-resolutions-generated-by-ai/

Year End Roundup for 2019

This New Years marks close to six years of blogging every day.  I write ‘em, a few of you click on ‘em.

The Traffic Stats Were Weird

The total hits on this blog increased again, up more than 10% from 2018.  As before, there is a huge amount of “long tail” in this traffic, with hits spread widely over the thousands of posts from the last 8 years.

But it isn’t clear exactly how many people actually look at this blog.  The stats I get are defaults from wordpress, so I don’t really know much about them.

This year saw a couple of mysterious blips.  I don’t know how much of this is real traffic, and how much of it is artifacts of the data collection.

Early in the year, the daily hits dropped dramatically.  This approximately corresponds to the European data privacy requirements, and the dearth of hits from that region suggest that the blog is either not available to some people (not being compliant in some way I don’t know about) or accesses are not reported (not having permission to collect that data).   I dunno.

But then, around August, traffic picked up.  Really picked up, to 100 hits per day.  During this burst, it tended to be bursty, with a few days of high traffic, as much a 300 hits per day, and then several days of low traffic.  From the imperfect information I can see, the bursts might be from Hong Kong (perhaps scraping the internet to make a copy to be used inside China?)

Then, around November, traffic dropped off again and has stayed low.  This drop approximately coincides with the increasing troubles in HK, so perhaps this reflects a cut off of Internet access there.

I really don’t know.

The Usual Stuff

The blog continued to cover the usual stuff.

Cryptocurrencies, the Future of Work (and Coworking), Dinosaurs, Birds, Robots, the Ice Is Melting, Renewable Energy.

I blog about anything that interests me and is worth the trouble.  I try to have something useful to say, though sometimes it’s mainly a link with “this is cool”

Many of the things I discuss are from current academic papers, which I cite and generally try to read at least the abstract and always point to the original sources.

“Coworking – The Book” and other Writing

My 2018 book “What is Coworking?” continues to sell like hot cakes–if nobody had ever heard of hot cakes.  I think it sold a couple dozen copies.  My plans for a new villa are on hold…. : – )

Writing is hard.  Selling books is even harder.

Speaking of writing, I also contributed an article to a local free paper, which I really like the title to:

1. Robert E. McGrath, Think Heliocentrically, Act Locally, in The Public I: A Paper of the People. 2019. http://publici.ucimc.org/2019/04/think-heliocentrically-act-locally/

I archived a report on the 2013 Alma Mater project.  Versions of this report was rejected by several conferences and journals.  A problem with working outside the box is that the journals of boxology won’t publish your results.

1. Robert E. McGrath, A Digital Rescue for a Graduation Ritual. Urbana, Illinois, 2019. http://hdl.handle.net/2142/105503

Onward

This blog will continue in the same vein, and daily posts will continue at least for now.

---

Band Names

In a continuing homage to Dave Berry, I have identified a bunch of phrases that would make great names for a band.  In general, these phrases are taken from actual, real scientific and technical papers.  So I am not making them up—just repurposing them.

Here is this year’s crop.

gerbil’s casket
Preen Oil
Carolina Preen Oil
Carolina Junco
Dark eyed Junco
Arctic Albedo
Mean Surface Albedo
Arctic Amplification
Amplified Arctic Warming
Surface Air Temperature
Snow Cover Fraction
Buckypaper
Pacific Pumice Raft
Sichuan Mudslides  (also a great name for cocktail)
Soft Exo Suits
The Weddell Gyre
Giant Miocene Parrots
Eocene Whale
Chicxulub ejecta
Perching Drones
Perch And Stare Mission
Due to a lack of sunlight in Scotland
Blogging Birds Of Scotland
Huddle Pod
Cuddle Pod
Giant Hopping Tree Rats
Kangaroo Ancestors
Prehistoric kangaroos
Tiny Pronking Robots
Computational Periscopy

---

Books

As always, I have continued the weekly review of one or more books that I read this year. This year I wrote about a total of 73 books, 24 non-fiction, 49 fiction.

Some Favorite Books of the Year

Fiction:

The Thrilling Adventures of Lovelace and Babbage by Sydney Padua
This is How You Lose the Time War by Amal El-Mohtar and Max Gladstone
Angels of Music by Kim Newman

Non-Fiction

Breaking and Entering by Jeremey N. Smith
The Heartbeat of Wounded Knee by David Treuer
You Look Like A Thing And I Love You by Janelle Shane

All the books reviewed (in no particular order)

Fiction

Ancestral Night by Elizabeth Bear
Stone Mad by Elizabeth Bear
Grand Union by Zadie Smith
Equoid (2013) by Charles Stross
Toast (2002) by Charles Stross
Speak Easy (2015) by Catherynne M. Valente
Six Gun Snow White (2016) by Catherynne M. Valente
The Essex Serpent by Sarah Perry
The Starless Sea by Erin Morgenstern
The Dakota Winters by Tom Barbash
Agent Running in the Field by John le Carré
Anno Dracula 1999 Daikaiju by Kim Newman
The Princess Beard by Delilah S. Dawson and Kevin Hearne
Washington Black by Esi Edugyan
Grave Importance by Vivian Shaw
The Grand Dark by Richard Kadrey
Amnesty by Lara Elena
Outside Looking In by T. C. Boyle
This is How You Lose the Time War by Amal El-Mohtar and Max Gladstone
The Nickel Boys by Colson Whitehead
Magic for Liars by Sarah Gailey
Gods of Jade and Shadow by Silvia Moreno-Garcia
Noir Fatale ed. by Larry Correia and Kacey Ezell
Inland by Téa Obreht
The Origins of Sense by Adam Erlich Sachs
Fall by Neal Stephenson
Gather The Fortunes by Bryan Camp
Anno Dracula by Kim Newman
The Future is Blue by Catherynne M. Valente
No Country For Old Gnomes by Delilah S. Dawson and Kevin Hearne
Early Riser by Jasper Fforde
European Travels for the Monstrous Gentlewoman by Theodora Goss
The Secrets of Drearcliff Grange School by Kim Newman
The Haunting of Drearcliff Grange School by Kim Newman
Luna: Moon Rising by Ian McDonald
Revolutionaries by Joshua Furst
Someone Who Will Love You in all Your Damaged Glory by Raphael Bob-Waksberg
Split Tooth by Tanya Tagaq
Unto Us A Son Is Given by Donna Leon
Binti by Nnedi Okorafor
Angels of Music by Kim Newman
The Burglar by Thomas Perry
Grim Expectations by K. W. Jeter
The Thrilling Adventures of Lovelace and Babbage by Sydney Padua
Macbeth by Jo Nesbø
No Sunscreen for the Dead by Tim Dorsey
Conversations with Friends by Sally Rooney
Infernal Devices  by K. W. Jeter
Fiendish Schemes by K. W. Jeter

Non Fiction

Lakota America by Pekka Hämäläinen
The Laundromat by Jake Bernstein
You Look Like A Thing And I Love You by Janelle Shane
They Will Have To Die Now by James Verini
Hollywood’s Eve by Lili Anolik
Proof!  By Amir Alexander
How to be an Antiracist by Ibram X. Kendi
Four Queens by Nancy Goldstone
The Next Billion Users by Payal Arora
How to Do Nothing by Jenny Odell
Places and Names by Elliot Ackerman
Eyes in the Sky by Arthur Holland Michel
American Carnage by Tim Alberta
Trick Mirror by Jia Tolentino
The Ice At The End Of The World by Jon Gertner
Dinosaurs Rediscovered by Michael J. Benton
Devices and Desires by Kate Hubbard
Stony The Road by Henry Louis Gates, Jr.
The Rise and Fall of Alexandria by Justin Pollard and Howard Reid
Cleopatra by Stacy Schiff
Breaking and Entering by Jeremey N. Smith
The Heartbeat of Wounded Knee” by David Treuer
Brilliant Green by Stefano Mancuso and Alessandra Viola
The Hidden Life of Trees by Peter Wohllenben
Before and After Alexander by Richard A. Billows

 

Book Review: “You Look Like A Thing And I Love You” by Janelle Shane

You Look Like A Thing And I Love You by Janelle Shane

I had never run across Shane’s AI Weirdness online, so this book was my first taste.  It’s great!  I was giggling out loud (which elicited a withering glare from the cat).  I love it.

Shane does a wonderful job of ‘splainin this stuff, and her examples help bring the AI hype down to Earth.  She sketches out how the common variations of machine learning work, how they are built, and how they go off the rails.

I’m pretty familiar with this territory. “You. Stupid. Computer” has one of the major themes of my career. Yes, children, we were inventing all this crazy nonsense decades ago.

As I sometimes summarize my own career, thanks to our work, we can now make a lot more stupid mistakes thousands of times faster.  It’s not so much how badly the dog (or perhaps giraffe) dances, it’s that it dances at all.  And today, the giraffe’s a lot bigger, dances a lot faster, and sometimes threatens life, liberty, and the pursuit of giraffes.  You’re welcome!

So, I can attest that she’s absolutely right technically and also anthropologically.  AI’s not evil, though it can be dangerous, and Shane has just the right balance of wonder (“look—the dog giraffe is dancing”) and skepticism (“but it’s a laughably bad dancer!”)

Most important, Shane is really good at ‘splainin stuff. The relationship between training sets and weird or not results. How machine learning learns.  The endless entertainment of trying to devise goals and rewards, and the perverse ability of computers to find shortcuts.  How computers copy human biases.  And more.

With so much AI being deployed, this book is a timely and important public service.  Everyone needs to understand this wackiest of recent human creations.

(You can find more at Shane’s blog.)

---

1. Janelle Shane, You Look Like A Thing And I Love You: How Artificial Intelligence Works and Why It’s Making the World a Weirder Place, New York, Little, Brown and Company, 2019.

 

Sunday Book Reviews