Anthropomorphic Software

25 points by abhin4v


peniblec

If you are aware of any more examples, I would be glad to list them here.

Years ago while skimming dmesg, I chanced on:

Dazed and confused, but trying to continue

I have no memory of what I was trying to debug that day, but I think about that message often.

"Mood", I suppose. Hang in there Tux đź«‚

adamshaylor

I agree completely with the substance of the article, but I’m not so sure about the label anthropomorphic software. I think maybe the author of the article senses a problem with it, too:

LLM agents are the ultimate version of anthropomorphic software. But agents do not have to sound human. I force my agents to not speak like humans for my own sanity. Agents might become the default paradigm of computer user interface in the future. I will miss anthropomorphic software then.

When I encounter ideal versions of anthropomorphic software as in the examples in the article, I don’t feel like I’m having a conversation with the software, I feel like I’m having a conversation with the human that wrote the software. For example, when the Elm parser prints a message saying, “I got stuck...” I take the word “I” to mean the person who wrote the parser (presumably Evan Czaplicki)—someone I can actually relate to and has a particularly human set of values based on human existence and experience.

Author’s voice could be a better label. LLMs can write prose and code. Setting aside how well or poorly they do it, they don’t have a singular voice and it’s debatable whether they can even be considered an author.

doug-moen

When I see these kinds of messages, I read them as the voice of the software author, who in these examples is a human, since all of the cited software is old enough. I do not anthropomorphize either the software, or the hardware that is outputting the messages, just as I don't personify a paper book.

LLMs are in a different category for me. There is no human behind the utterances, and I perceive that as an abomination.

fleebee

Nice read. I appreciate occasional moments of whimsy like these in software as long as it's not overdone.

LLM agents are the ultimate version of anthropomorphic software. But agents do not have to sound human. I force my agents to not speak like humans for my own sanity.

Indeed, I imagine by now most people equate LLMs with conversational agents. But it need not be so! I wish it was easier to interface with an LLM without it gaslighting me about having human qualities and in so much prose (instead of e.g. single word responses). I believe issues such as sycophancy and chatbot psychosis are exacerbated because of this behavior which was put in there deliberately.

kevinc

Despite the usability cost, I was delighted when I tabbed away and the title changed.

Screwtape

Apparently early versions of the C compiler shipped with the Macintosh Programmer's Workshop (the SDK for the original 68k Macintosh) could produce error messages in this category.

Call me paranoid but finding /* inside this comment makes me suspicious

a typedef name was a complete surprise to me at this point in your program

type in (cast) must be scalar; ANSI 3.3.4; page 39, lines 10-11 (I know you don't care, I'm just trying to annoy you)

theblacklounge

LLM agents are the ultimate version of anthropomorphic software. But agents do not have to sound human. I force my agents to not speak like humans for my own sanity.

What's the prompt/skill for that?

valdemar

A bit of a odd example of this is the configure script that Perl uses when it is being built. Here is a random run of it in their GitHub test suite: https://github.com/Perl/perl5/actions/runs/43767022285/job/132976918972#step:5:1

It has loads of lines like

Checking to see how big your integers are...

Which was just really fun to me the first time I had to build it.

gavinmorrow

I usually don’t like this style of program. When used sparingly/as an easter egg almost it is fun, but for eg compiler error messages I find it always distracts me. I usually think of software that I’m using as a tool, so it saying “I” bugs me as somewhat uncanny/wrong.

gcupc

I have frequently reused the phrase "Cowardly refusing to... " in my own error messages.

theblacklounge
Comment removed by its author
coby

Most usage of this effect is purely for whimsy and isn't intended to improve the user experience

I disagree. I think it's a reasonable way to avoid passive voice, e.g. "X could not be found." In complex cases it can really simplify the structure of the sentence.

Using Elm's error messages, which are world-class IMHO, as an example of what not to do is kinda wild to me. I guess it's a preference thing, but I've always found passive voice with complex constructions way more annoying.