I’m mostly a descriptivist when it comes to the meaning of words. If everyone uses “literally” to mean “figuratively”, that’s fine by me. Same if there is no likelihood of confusion, as with “aggravate” for “annoy” as well as “exacerbate”. But that still leaves me plenty of room to gripe about the way in which words are used, so it’s time for a good old whinge.
What got me started on this was talk about “demand destruction” in oil markets, often claimed to be a term used by economists. This piece of idiocy conflates reductions of consumption in direct response to price changes (movements along the demand curve), reductions for other reasons like the arrival of electric cars (shifts in the demand curve) and permanent shifts away from oil as a response to (presumably temporary) price shocks arising from events like the US-Iran war.
Next up is “algorithm”. Until recently an algorithm was a defined procedure, guaranteed to yield a correct answer to an appropriate class of problems given the right starting conditions and sufficient time. The classic example is Euclid’s algorithm for finding the greatest common divisor of two numbers.
Now the term is used to mean the undefined and incomprehensible way in which a variety of decisions (from the content of social media feeds to the selection of bombing targets) are derived from the inputs fed into complex models[1]. The term carries over the connotations of precision from the original definition, often for people who would deride any inference described as coming from a “model”. This battle is pretty much lost, but I’m not giving up quite yet.
On the other side of this coin are spurious attempts at sophistication, rather like children putting on lipstick. My favourite is “methodology” (the philosophical study of scientific method) used as a fancy synonym for “method”.
Runner up: contorted attempts to use “data” as a Latin-derived plural (presumably of “datum”[2]) rather than as a mass noun like “water”. To see how wrong this is, compare “a large amount of data” with the clunky “a large number of data”. In the rare case when we need a singular/plural distinction, correct is “data point(s)”, with the associated collective term “data set”.
Finally, my favorite peeve is “begs the question”. Its original form, meaning “assuming the conclusion” is an absurdly literal translation of the Latin “petitio principii”. The most common modern use, meaning “raises the question”, just marks the user as unaware. Best to avoid it completely.
As I said at the start, there is no way of setting the language in stone. But some innovations make for more clarity or expressiveness, while others collapse useful distinctions (like methodology for method) or are inherently dishonest (like algorithm for model).
[1] There must be an algorithm somewhere inside the model, maximising an objective function of some kind. But since we don’t know what the function is, that’s not very helpful.
[2] “Datum” is a technical term used by surveyors with the ordinary plural “datums”.
{ 35 comments }
AndrewMcK 09.29.26 at 8:10 am
“As I said at the start, there is no way of setting the language in stone. But some innovations make for more clarity or expressiveness, while others collapse useful distinctions (like methodology for method) or are …inherently dishonest (like algorithm for model).”
Or my pet peeve: reform for change
Alex SL 09.29.26 at 8:53 am
I was not aware of the correct or original use of some of these, e.g., methodology and demand destruction, so thanks. I also wince when I see begging the question being misused.
However, I would in good faith disagree with model. That word has long been been very broad, with a lot of applications. There are mathematical models, a 1:50 model of a building, model airplanes, my mental model of the world, and even fashion models. I find it difficult to argue that the current use by the AI bros is a serious misappropriation before that history. How they use words like classification, label, and skill, on the other hand… I have a gripe about those.
MisterMr 09.29.26 at 9:14 am
I like the data/datum thingie! Is mass media/mass medium used?
I also think the correct form should be “omosexual” instead than “homosexual”.
marcel proust 09.29.26 at 12:27 pm
I’m with MisterMr about the data/datum thingie; I quite like “a large number of datums”!
I also appreciate Alex SL’s use of the word “before”. I would have used “after” in that sentence (e.g., I find it difficult to argue that the current use by the AI bros is a serious misappropriation after that history.), indicating “following”. Using “before” instead suggests the history as a backdrop which I find a much more colorful image.
I too had been unaware* of the history of begs the question. Thank you.
*”Unaware” is so much kinder (to me) than “ignorant”, thank you very much!
Scott P. 09.29.26 at 3:11 pm
Data is the plural of datum. In archaeology, we use both.
John Q 09.29.26 at 6:38 pm
Alex SL. I wasn’t clear enough about “models”. What I meant to say is that there are lot of people (climate deniers, for example) who sneer at models, but that many of the same people are impressed by “algorithms”
MP Yes, unaware is kinder. I’ve edited that way
Scott P. A quick search suggests that “datum point(s)” is used by archeologists in the surveying sense I mentioned, and that “data” is used in the ordinary way to refer to information generally. Am I wrong in this/
JPL 09.29.26 at 11:39 pm
The peeving would be more effective if you were to give examples of the uses you find objectionable, preferably from written sources or from journalists on the telly. My pet peeve is frequently exemplified by our current generation of college students, namely “based off of”, instead of “based on”. With “based off of”, which expresses “movement away from” some location, and which sometimes seems to be used as equivalent to something like “on the analogy of”, you lose contact with the very powerful relation of logical support, or logical dependency which is widely expressed by the word “on” in a variety of contexts. In “The judgment that noncitizen voting is not a significant problem is based on the fact that there is very little evidence of actual cases of noncitizens voting.” the truth of the judgment is logically dependent on the truth of the premise (“the fact that …”), just as the fact that my computer is “on the table” and not on the floor is dependent on the fact that a table is supporting it and preventing it from moving further to the ground. When you say “based off of” you lose all that. “The claim that there is a widespread phenomenon of noncitizens voting is based on nothing” (and therefore can not be accepted as true) seems more effective than saying “… is based off of nothing”. The attempt to make the further conclusion would then seem to lack necessity.
Moz in Oz 09.30.26 at 2:27 am
Until recently an algorithm was a defined procedure, guaranteed to yield a correct answer to an appropriate class of problems given the right starting conditions and sufficient time.
There’s a whole class of ‘chaos theory’ algorithms without definitive answers, ones that aren’t guaranteed to complete in finite time (the halting problem), the mathematics of limits and a few other issues that push your “until recently” back quite a bit. We have algorithms for calculating the value of pi, for example, that don’t fit the halting problem because they are guaranteed never to complete.
spurious attempts at sophistication
Sophistication is one of those words too. Once upon a time it mean “unnecessary complexity, especially when used for show” and I think it’s reasonable to argue that your use refers to that meaning of the word :) “Enabling us to utilise our learnings on a going forward basis” as the modern managers would no doubt put it.
Tm 09.30.26 at 9:32 am
Moz: I think this is a misunderstanding. The halting problem is the observation that given a certain set of computer programs, there is no algorithm that can accurately determine which of these programs will stop in a finite time.
Given a specific algorithm, it is possible to prove whether it will stop in finite time. An algorithm for calculating Pi that doesn’t stop is … not very useful (of course the result will always be an approximation). What are you alluding to?
nonrenormalizable 09.30.26 at 2:50 pm
I would include the change in “enormity”, previously used to describe something vastly monstrous, to become the abstract noun for the quality of being “enormous”. See also “prodigal”.
There is also the creep of business-speak: one is now expected to extract the “learnings” from an experience, rather than take a “lesson”.
An older one that I keep getting wrong and believe should be retired from use is the difference between being a “spendthrift” and being “thrifty”, which are antonyms yet give the impression of being close in meaning.
nonrenormalizable 09.30.26 at 2:53 pm
Btw @JohnQ, given your views on “algorithm” and “model”, any thoughts about this debate on anthropomorphising language when describing the behaviour of AI agents?
Moz in Oz 09.30.26 at 7:55 pm
My understanding is that the algorithms for calculating pi all involve infinite sequences. They’re useful because people can run them for a finite number of steps and decide that the result is accurate enough. But those are variations of an algorithm that starts with “for i = 1 to infinity”.
I disagree that it is always possible to prove that a specific algorithm will (or will not) stop in finite time. For example “watch this proton. Halt when it decays”. Will it stop? You could write an algorithm that halts only if P=NP if you want a “pure maths” version. AFAIK both are currently unknown.
Tm 10.01.26 at 4:53 pm
“My understanding is that the algorithms for calculating pi all involve infinite sequences. They’re useful because people can run them for a finite number of steps and decide that the result is accurate enough. But those are variations of an algorithm that starts with “for i = 1 to infinity”.”
So what happens if you let an algorithm run to infinity? You’ll never get a result. An algorithm must have a termination condition otherwise it’s useless. Of course many algorithms calculate approximations. They stop when the approximation is close enough by some predefined criterion. But they stop eventually.
“I disagree that it is always possible to prove that a specific algorithm will (or will not) stop in finite time. For example “watch this proton. Halt when it decays””.
That’s not an algorithm. “Wait until I come back” is also not an algorithm.
I wasn’t very precise about the Halting Problem. “In computability theory, the halting problem is the decision problem of, given an arbitrary computer program and an input, determining whether said program will eventually finish running and halt, or will continue to run forever. Alan Turing proved in 1937 that the halting problem is undecidable, meaning that no *general algorithm* exists that can correctly solve the problem *for all possible program–input pairs*.”
https://en.wikipedia.org/wiki/Algorithm
https://en.wikipedia.org/wiki/Halting_problem
Moz in Oz 10.01.26 at 9:33 pm
Tim, I think it’s worth noting that the algorithm definition requires that a set of instructions must provably terminate after a finite number of steps, but the halting problem is mostly interesting for finite sets of instructions that might not fulfill that criteria. In that sense your objections are tautological.
https://www.merriam-webster.com/dictionary/algorithm this points out another modern use of the term, “the algorithm” that picks content for a social media feed. But also “a set of step-by-step instructions used by a computer system to achieve a particular result” that does not include the “provably finite time” requirements from wikipedia and the “until an indefinite condition is met doesn’t count” of your definition.
If we accept the wikipedia definition we need a new term or terms for things like infinite sequences and indeterminate ones. And another for computer programs/collections/subsets, since a lot of useful ones don’t provably terminate in predictable ways (operating systems, for example, especially if you exclude “crash when the unix date rolls over” as an algorithmic condition)
https://en.wikipedia.org/wiki/Computer_program for example, describes a Turing machine as “machine can move the tape back and forth, changing its contents as it performs an algorithm” when we’ve established above that the set of instructions is not interesting if it describes an algorithm,
MisterMr 10.02.26 at 9:55 am
@TM 13
“Wait until I come back” is also not an algorithm.
ac1 wait 5 minutes
ac2 check if TM is here; if no goto ac1
ac3 start nucular armageddon
is actually an algorithm, that shows that the term is just too broad.
In common use though, “algorithm” is used as a shorthand for “complex and therefore obscure set of choices and steps done by an automated system that somehow have influence on my life but I don’t know how it works”.
So this is more a problem of a new derived meaning that is actually way more specific (but also more subjective) than the original one.
Thomas P 10.02.26 at 10:06 am
People using ´epicenter´instead of ´center´ just to sound fancy is one thing that annoys me.
John Q 10.02.26 at 10:57 am
non-r @11. I just had a piece in the Guardian arguing against anthropomorphising and in favour of holding AI companies respongable for malfuctioning programs aka “rogue agents”. Coming soon
Tm 10.02.26 at 3:01 pm
Folks, as JQ already pointed out, algorithm is a scientific term with a specific meaning. If you are not familiar with it, don’t use it. Algorithm isn’t coextensive with computer program. An algorithm may be implemented in a computer program, but not any set of computer instructions is an algorithm. Just use computer program if that’s what you mean.
A Turing machine (TM) is a precisely defined mathematical abstraction. It’s also not the same as an algorithm, it’s a mathematical model for a computer that can implement algorithms. Part of the definition is that a TM has an input and can generate an output. But the output is only defined when the machine terminates. That of course is true for any computer program. If it runs into an infinite loop, it’s useless. Why is this so hard to understand?
Infinite sequences are mathematical objects. An infinite sequence may converge, for example on pi. But any computer program used to approximate pi has to stop after a finite number of steps. Of course it does!
JimV 10.02.26 at 11:47 pm
“I was humbled to receive [some award]”, which I have heard from many athletes and recently from a well-known mathematician, is my peevest pet. In my day, being humbled meant being grabbed by the back of he neck and having your nose rubbed in the dirt, not being honored by an award for achievement.
Mark 10.03.26 at 12:29 pm
I used to work in a support role at a management consultancy and I noticed the arrival a few years ago of “an unlock” as a fancy way of talking about the breakthroughs that the consultants promised to achieve. I knew it wouldn’t be long coming, and sure enough, not long afterwards I spotted my first “key unlock” in the wild.
oldster 10.03.26 at 7:25 pm
I’m not a pedant, but I find it unforgivable to speak of “the metric system” when referring to the International System of Units (SI).
Of course the SI is * a * metric system. But the Imperial System of units is also a metric system. Thumbs and feet are also metric systems. A metric system is any system used for measurement. Talking about “the metric system” as though it were a proper name is simply displaying one’s ignorance for all to see.
Mind you, though, I’m not a pedant. I simply insist on clarity in speech.
John Q 10.03.26 at 7:48 pm
Archimedes (version of the) method of exhaustion provides an algorithm to calculate the first N digits of pi , for any finite N. But there isn’t, and can’t be, an algorithm to calculate pi.
MisterMr 10.05.26 at 12:28 pm
@TM 18
I do in fact believe that in scientific terms all computer programs are algorithms, though not all algorithms are computer programs of course.
Basically all that can be described by a flowchart is an agorithm AFAIK.
John Q 10.05.26 at 10:16 pm
MisterMr @23 See footnote 1
John Q 10.05.26 at 10:18 pm
AndrewMcK @1. I’ve given up on the idea that “reform” must mean “change for the better”. Now I just use it to mean “programmatic change, for good or ill”
Tm 10.06.26 at 12:54 pm
MisterMr: No, not any flowchart describes an “algorithm”. There’s a difference between process and algorithm. For example HR has a process for the steps taken when an employee is hired and this can be represented in a flowchart. It makes no sense to refer to the HR’s “hiring algorithm”. An algorithm is a defined set of instructions for solving a computational problem.
And of course there’s a difference between computer program and algorithm. A computer program may implement an algorithm but it’s not an algorithm. Nobody says “I just wrote an algorithm” when they mean “I wrote a computer program”, although writing a computer program may, as a preparatory step, require writing algorithms. Take it up with the computer scientists if you disagree with their terminology.
David, a Bostonian in Tokyo 10.09.26 at 8:06 am
“Take it up with the computer scientists if you disagree with their terminology.”
You called? (If an SB and MS in CS from a previous century count as “computer scientist”.)
Yes. The _halting problem_ isn’t about algorithms at all. It’s about whether a particular computer in a particular state will ever halt if set running. The time you have to wait to see (as a function of the size of the program/computer) turns out to grow faster than any computable function, that is, computers are the most complex things possible*. Very nasty function, that.
*: And an obvious corrollary** is that the human mind is either a computer or it’s magic, and since there’s no such thing as magic, humans are computers. Get over it.
**: This depends on the point that the word “analog” is a lie: the mathematics used to describe analog signals and circuits implicitly assumes that every analog signal can hold an infinite amount of information, but in real life, every analog signal has noise, and that noise limits the amount of information (the value of) that analog signal can represent. So any actual real-life analog signal or system can be simulated by a digital signal or system with enough bits. But I digress (as usual).
J-D 10.10.26 at 3:29 am
Is a network of computers not more complex than a single computer? Is a robot which incorporates a computer not more complex than a computer? Is the whole cosmos not more complex than a single computer?
MisterMr 10.11.26 at 2:10 pm
@TM 26
thanks for the clarification, so the difference (in strict sense) is that an algorithm as an exact ending point, and that point is computational, not an action.
engels 10.12.26 at 10:18 pm
#include
int main() {
for (;;) roll(boulder, Uphill);
// boulder rolls down hill
}
Tm 10.13.26 at 7:38 am
MisterMr: An algorithm, contrary to a computer program, is concerned with a *single, specific computational problem*. And conceptually, it’s an abstract entity, not to be mixed up with a specific implementation in a specific programming language.
engels 10.13.26 at 9:26 am
not all algorithms are computer programs
https://plato.stanford.edu/entries/church-turing/
engels 10.13.26 at 11:26 am
I don’t think there’s a standard definition of algorithm. I think it refers to a defined procedure that isn’t a formal computer program (≈a recipe). People talk about “Google’s algorithm” rather than “Google’s software” because they want to keep the discussion at a higher level of abstraction than implementation, resource use, bugs, etc. Iana computer scientist but i think it’s legitimate to speak of non-terminating algorithms when discussing the design of operating system processes etc.
somebody who knows that Algorithms on internet websites are a...holes 10.13.26 at 12:51 pm
Be-cause Algorithms on internet platforms frequently amplify hate speech, extreme content, and toxic political discourse to maximize user watch time and advertising revenue.
How the Mechanism Works
• Profit-Driven Design: Platforms use machine learning models trained to prioritize content that triggers strong emotional reactions—such as anger, outrage, or fear—because these emotions keep users scrolling longer.
• Amplification of Extremes: Studies published in Science and Nature show that algorithms systematically favor partisan, hostile, or activist content over neutral or traditional news sources, driving up engagement.
• Feedback Loops: When users interact with sensational or divisive material, the algorithm interprets that action as a preference and feeds them more of the same, creating insular echo chambers.
Impact on Society and Polarization
• Affective Polarization: Research indicates that exposure to hostile content increases emotional hostility toward opposing political or social groups (known as affective polarization).
• Normalization of Hate: Anti-defamation and civil society groups, including leaders from the Anti-Defamation League, note that automated feeds can turn fringe bigotry or targeted harassment into mainstream conversation by funneling it to massive audiences.
Diverted Attention: Beyond political divides, algorithmic feeds act as a constant diversion mechanism, pulling public attention away from complex, nuanced issues toward fast-burning cultural conflicts.
Potential Mitigations
• Alternative Ranking Systems: Independent experiments (such as browser-based re-ranking tools tested during recent elections) show that downranking antidemocratic or hostile posts can significantly lower political hostility without removing content.
• User Agency and Regulation: Researchers and policy groups emphasize the need for greater algorithmic transparency, options for chronological or user-controlled feeds, and potential regulatory “engagement taxes” to alter platform incentives.
Tm 10.13.26 at 9:00 pm
Engels, No doubt Google‘s software uses algorithms. But what makes you think they are “non-terminating” and what would that even mean? They never produce a result?
Comments on this entry are closed.