AI’s Achilles’ heel: typewriter technology?

by | Dec 29, 2022 | artificial intelligence | 7 comments

OK, I know there are some enemies of AI among my readers. And I’d call myself … a frenemy? I’m fascinated by its recent advances, but also very wary.

I tried asking ChatGPT some technical questions about typewriters. It produced some authoritative-sounding nonsense, which goes to show that you must be careful around these systems. 

First, some background. In case you didn’t know, the name of Loose Dog Press is a little joke that alludes to a piece in a typewriter’s escapement (the mechanism that allows the carriage to advance step by step). The loose dog waggles back and forth as it and the rigid dog alternate engaging the teeth of a starwheel or toothed rack underneath the carriage. There are many ways in which the mechanism can be designed more specifically. Here are some examples from the Underwood, Remington, and L.C. Smith escapements, as pictured in the invaluable Ames Mechanical Training Manual of 1945.

So, could ChatGPT simulate an intelligent conversation about this topic? First I asked:

Can you explain how a loose dog functions in a typewriter?

The system’s initial answer was that there was no connection:

I’m sorry, but it is not possible for a loose dog to function in a typewriter. Typewriters are mechanical or electronic devices that are used to input text by typing on a keyboard, which in turn activates levers or electronic switches that cause a printing element (such as a hammer or a laser) to strike a ribbon or ink cartridge and print the corresponding character on a piece of paper.

A loose dog, on the other hand, is a domestic animal that is not confined or controlled in any way. It is not capable of operating a typewriter or any other type of mechanical or electronic device.

It is not clear where the idea of a loose dog functioning in a typewriter comes from, but it seems to be a nonsensical and unrealistic concept.

I explained:

It is not nonsensical, because a loose dog is also part of an escapement.


I apologize for misunderstanding your question. An escapement is a mechanical device that controls the movement of a clock or watch by releasing the stored energy in the mainspring at regular intervals. In a clock or watch, the escapement converts the rotational motion of the gear train into a back-and-forth oscillating motion, which drives the hands of the clock or watch and allows it to keep time.

A loose dog, as you mentioned, is not a part of an escapement. It is a domestic animal that is not related to mechanical devices or the functioning of a clock or watch. It is not possible for a loose dog to be a part of an escapement or to have any effect on its operation. 

I hope this clears up any misunderstanding. Please let me know if you have any further questions. 

Note that the system is limiting escapements to timepieces—an overly narrow definition—and responding as if I’d denied that loose dogs were part of escapements. I tried again:

“Loose dog” is a term for part of a typewriter escapement.

Now things get interesting:

I apologize for the confusion. You are correct that “loose dog” is a term used to refer to a part of a typewriter escapement.

In a typewriter, the escapement is a mechanical device that controls the movement of the typebars, which are the metal arms that hold the type elements (such as letters and symbols) and strike the ink ribbon to print the corresponding character on a piece of paper. The escapement in a typewriter consists of a series of levers and pivots that allow the typebars to move up and down and lock into place when they are not in use.

One of the parts of the escapement in a typewriter is the loose dog, also known as the typebar dog or the escapement dog. The loose dog is a small projection on the side of the typebar that engages with the escapement mechanism to control the movement of the typebar. When the typebar is not in use, the loose dog locks into place and prevents the typebar from moving. When a key on the keyboard is pressed, the typebar is released from the escapement and can move up and down to strike the ink ribbon and print the corresponding character.

I hope this information helps clarify the role of the loose dog in a typewriter escapement. Please let me know if you have any further questions. 

This sounds so confident that it almost makes me doubt my own knowledge of typewriter mechanisms. I’m not a professionally trained typewriter tech, after all. But to the best of my knowledge, this is nonsense! The machine has invented an account of the loose dog and the escapement that is simply wrong. (The term “typebar dog” appears nowhere on the Internet, per Google, and was just made up by ChatGPT.)

One more example: I asked whether all typewriters have keyboards. Response:

While the vast majority of typewriters do have keyboards, there are some exceptions.

One example of a typewriter that does not have a keyboard is the Velotype, a machine invented in the late 19th century that used a system of perforated paper tape to input text. The Velotype used a punch card system similar to that of the Jacquard loom, with a series of punched holes in the tape representing the characters to be typed. The tape was fed through the Velotype and read by a set of sensors, which activated the typebars and printed the corresponding characters on a piece of paper.

Another example of a typewriter that does not have a keyboard is the Stenotype, a machine used by court reporters and other professionals to take shorthand notes. The Stenotype uses a specialized keyboard with a smaller number of keys, each of which represents a combination of letters or syllables. The Stenotype records the keystrokes in a shorthand notation and can later be transcribed into standard text.

While these examples are relatively rare, they demonstrate that it is possible for a typewriter to function without a keyboard. However, most typewriters do have keyboards as a primary means of inputting text.

In fact, the Velotype is a chording keyboard invented in 1982 (not to be confused with the Velograph and Virotyp, which are index typewriters and have no keyboard). The much-better-known Stenotype also uses a chording keyboard. Note that this is a kind of keyboard, as the chatbot itself says—so these are not examples of keyboardless typewriters! Furthermore, the chatbot’s description of the Velotype seems to confuse it with an automatic typewriter such as the Friden Flexowriter.

This all suggests that a way to foil a chatbot (make it fail the Turing test) is to ask it questions about a specialized field in which you have some firsthand knowledge. (This is also the best way to find errors in Wikipedia—forgive me for linking to it above.) 

I almost find myself getting mad at this stupid machine! I have to remind myself that it’s neither stupid nor intelligent, because it is not a mind—it is only an algorithm invented by clever humans.

Moral: Don’t let AI befriend you too easily—you just may be talking to a digital con artist.


  1. Bill M

    I always say the A part of Artificial is true, it is artificial. Intelligent is a misnomer as there are no digital devices or anything except a human that is intelligent (ok, some people seem unintelligent and some animals seem intelligent). So AI is artificial but it is far from intelligent.

  2. Richard P

    I agree. I think it should be called SI, simulated intelligence.

  3. Anonymous

    I am reminded of several episodes of the 1960's Star Trek, where Kirk saves the day by talking all-powerful computers into committing seppuku by confusing them with illogic. :D

  4. Anonymous

    I'm glad to see this post! I just read an interesting book about the dangers of AI: THE LOOP by Jacob Ward. Worth the read. Also, I listened to an interesting podcast about the limits of AI and ChatGPT (I can't remember which podcast). The example given was that ChatGPT was asked to write a love story. The story seemed amazing at first, but really it was just a random story cobbled together using information from stories that had been fed to GPT. It can produce a story only from the information it has been given. So the characters were standard fairytale characters, not BIPOC characters, for example.

  5. Richard P

    I wonder what would happen if one asked ChatGPT to write a story with certain types of characters. I suppose it all depends on its available database—but that database must be pretty huge. Thanks for the tip on THE LOOP.

  6. Anonymous

    I also played a bit with AI this way. I asked what it knew about the Adler 9. I got a lot of general nonsense as an answer. Worse was that it accepted my corrections as the truth and used it confidently in its answers. For example, I made up that the Adler 9 had a calculation mechanism. AI believed it and told me this as if it were true. I was really disappointed, also because it doesn't want to reveal specific sources, so you can't check anything. Also, AI can't rhyme.

  7. Richard P

    Thanks for sharing your experience. One would like to believe that such examples of spurious information would make the general public less naive about their use of computers—but that belief would be naive.


Submit a Comment

Your email address will not be published. Required fields are marked *


typewriter revolutionary factory logo




Dept. of Philosophy
Xavier University
3800 Victory Pkwy.
Cincinnati, OH. 45207