I know it's petty, but describing things as machine learning when they existed for a long time as part of statistics is a red flag for me. k-means is machine learning, regression is machine learning, t-tests are machine learning, everything is machine learning. - ThreadSky

chanret.bsky.social • 24 days ago

I know it's petty, but describing things as machine learning when they existed for a long time as part of statistics is a red flag for me. k-means is machine learning, regression is machine learning, t-tests are machine learning, everything is machine learning.

Reposted from Tim Bale

Strikes me this bit from the TBI's new report is redolent of a wider, ongoing tendency to de-emphasise (and therefore normalise) the essential ideology of the far right, but hey-ho. (Source: institute.global/insights/pol...)

Comments

hetanshah.bsky.social•24 days ago

I’ve taken to actively complimenting people for not describing simple statistics as AI to help shift the incentive structure! https://bsky.app/profile/hetanshah.bsky.social/post/3lgwv3otxkc2b

johanzvrskovec.bsky.social•24 days ago

Probably a language shift. Lots of people aware of this fact that just want to communicate, get their results across. Clustering may seem exotic to social scientists and will probably be called machine learning just because of that, as opposed to regressions etc.

psurridge.bsky.social•24 days ago

I wholeheartedly agree. The same was also true in the 'big data' era.

danielvulkan.bsky.social•24 days ago

Ooh - I used k-means in my masters dissertation 11 years ago - can I now put "experienced in machine learning techniques" on my CV?

mariosrichards.bsky.social•24 days ago

The thing that got me was the actual content - they appear to have clustered all the parties (N>~20?) at once so the US Republican Party (ass == "Better-off supporters of Donald Trump") has a whole category almost to itself (CSU+Can-Con) ...

... and the UK has no "Established Right" party.

psurridge.bsky.social•24 days ago

Obviously the machines fault, I think that's why you call it machine learning so you can blame the machine 😀

mariosrichards.bsky.social•24 days ago

Silly to niggle about text labels but weird to choose to label "Established Right" as the *smallest* category (in public support/number of parties) when it just cries out "maybe rerun this one country at a time / normalise input dimensions per country?".

policysketch.bsky.social•24 days ago

Regression isn't machine learning, but machine learning *is* just curve fitting... I guess the difference is machine learning is curve fitting where we have no idea what the program is actually doing?

chanret.bsky.social•24 days ago

I've got no interest in characterising machine learning properly; I think that with the right nudge it probably extends to human learning too

adamlbrown.bsky.social•23 days ago

I used excel to draw a line. This is ai. Pls give me £45m

cjterry.bsky.social•24 days ago

Give them time, it'll be AI soon

jack-bailey.co.uk•24 days ago

Also hilarious that they had to use k-means to come up with those labels!

tombarton.bsky.social•24 days ago

Reminds of when I was applying for my PhD on the advanced quantitative methods steer the ERSC had, one condition was to use big data, turns out big data ≈ have statistical power at the aggregate level

t0nyyates.bsky.social•24 days ago

https://x.com/MaartenvSmeden/status/1083832145779535872

regularfry.bsky.social•24 days ago

K-means *is* machine learning, though. It has been for decades. It's an unsupervised learning algorithm. I don't think it's possible to come up with a reasonable definition of "machine learning" that excludes it.

chanret.bsky.social•24 days ago

I'm generally not that interested in giving a positive characterisation to machine learning; I do think that people have been drawing Voronoi diagrams around scatterplots for longer than machines have been automating that process, and so k-means is not exclusive to machine learning

regularfry.bsky.social•24 days ago

The thing we call k-means came out of signal processing in the 60's. It's always been a mechanised process. People might have done it manually as well, but that's incidental.

regularfry.bsky.social•24 days ago

Besides, if you're not going to allow for a definition of "machine learning", then I don't see how it's reasonable to object to any use of it as a label *at all*. "Everything is machine learning" comes directly from that lack of definition, not from anyone else's use of the term.

chanret.bsky.social•24 days ago

I allow for it, but I'm not interested (personally) in working towards one. Is there a particular definition that you like, or find useful?

regularfry.bsky.social•24 days ago

I'm generally ok with "machine performance against some evaluation function improves with some experience". It makes the term extremely broad, but intentionally so. And yes, that does make regression machine learning *when done by a machine*.

maryindevon.bsky.social•24 days ago

No petty - important

And I'm old enough to remember when we human machines had to calculate some of that!

eduardoramirez.bsky.social•24 days ago

I can’t find the original picture, but a great slide I saw once:

If you randomly guess at the answer over and over you’re called a hack.
If you program a computer to guess 50k times per second it’s called Machine Learning and pays 4x your salary.

peacedayresistance.bsky.social•24 days ago

I am confused because I remember having to do regression and t-tests with a pencil and paper and a million annoying computations. I think there was some idiot at my school who felt like calculators were the devil. But those concepts were invented before the computer.

zabong69.bsky.social•19 days ago

Did you also do neural nets by hand?

zabong69.bsky.social•19 days ago

I give statistics all hypothesis testing, but that’s about it. Any method that doesn’t start with strong assumptions on underlying distributions is machine learning, it’s not the ML worlds fault that we had to wait for the invention of the computer to make them useful.

greyskiesthinking.bsky.social•24 days ago

It's not petty when people are pretending stuff that's been around for decades is cutting edge.

skeptical-lemming.bsky.social•21 days ago

The roots of ML are based in prob & stats, but it adds the CS side of computational complexity/efficiency and algorithmic improvement to its problems. Add 'black box' algos, integrating ideas from cog sci. And only more recent has headway (usu by Bayesianists) been made into their explainability.

caprosser.com•24 days ago

Can’t charge the big bucks if you say ‘we use a method invented in the 1950s’!

colmpm.bsky.social•24 days ago

Pure idle speculation here, but I wonder if it's because the word "algorithm" has taken on more negative associations in some quarters. "Machine learning" has yet to do so.

I suppose it might just be "it sounds modern and fresh, even if it isn't"

chanret.bsky.social•24 days ago

b/c of the grading fiasco? I had assumed it was more your second possibility

colmpm.bsky.social•24 days ago

Yeah I guess I was thinking of that, plus a steady drip drip of other stories. But you're probably right. Occam's razor etc

mariosrichards.bsky.social•24 days ago

It's kind of fun - the person who was asked to sit down and write the report obviously knows nothing of "machine learnering" - and the person who was asked to sit down and knock out a model in python obviously knows nothing of political science ...

mariosrichards.bsky.social•24 days ago

... I'm pro people Just Having A Go, I just hope they're more familiar with Taking Constructive Criticism On The Chin.

And if the TBI is up for throwing money at people with no relevant qualifications - my inbox is open!

profafinlayson.bsky.social•24 days ago

It’s not just the history of statistics they ignore. The politics they describe draws on ideological traditions been forgotten by many but hardly new. I’m also bewildered by the claim that class loyalty is no longer part of politics when we’ve just seen some of the richest class buy a government.

Comments

Posting Rules

Reply