Nic Fox

Outlook pop-up incorrectly suggests “your” in the drafted phrase “or whenever you’re free”

So helpful 😳

When a company in an industry built on hype tells you that a use case is a bad idea—and actually dangerous—that means it’s a *catastrophically* bad idea.

The Enshittificator Digital products and services keep getting worse. In the new report Breaking Free: Pathways to a fair technological future, the Norwegian Consumer Council has delved…

Criminally funny, but also I'm crying inside

Norway comes for the #TechBros

www.forbrukerradet.no/breakingfree

vimeo.com/1168468796?f...

“I don’t think it’s extreme or radical to think working hard should get you a nice life”

Children in care denied school places, Lincolnshire home warns Councils seek powers to require more schools to take children as heads say funding would meet needs.

“40 [/]49 providers…273 homes…difficulties securing school placements for children in their care.
Some…waited 6 months[+] to get a child into school”
“struggles to regulate his emotions, after experiencing trauma…tends to run away from situations he finds overwhelming”

www.bbc.co.uk/news/article...

Former boss of Halifax Skircoat Lodge care home abused children for decades A jury hears children were preyed upon by Malcolm Phillips and his assistant for more than 20 years.

“would often run away after being…abused”

“She began skipping school, hoping it would stop... "…I loved school but if… you started running off or skiving school or shoplifting, they were scared of us being picked up by the police & opening our mouths"”

www.bbc.co.uk/news/article...

Fellow sewist & data person here 🙋🏻‍♀️ Handmade wardrobe & mini quilt wall hangings.

Nice work on the sweatshirt alteration 👌

Sounds good. Hope “Failure to engage will have serious consequences, such as possible criminal charges+time behind bars” doesn’t negatively impact neurodivergent children, those w/ trauma or any other children who may find it more difficult to engage w/ the hopefully-not-one-size-fits-all process.

Crafting and pets. Absolute bliss 😃

“we should hold on to our […]rough edges, our lack of fitting in […]the things that make us resist compliance with pressures of power.
Instead of feeling guilty or out of place, feeling responsible for fitting in, we should hold ourselves (&each other) responsible for resisting the forces of harm” ❤️

Thanks 🙂

(Cont’d…) Is the difference whether we’re commenting on an individual person/observation or the variables A and B generally?

Thanks for this post and your paper. I’ll have a look. I’m still getting to grips with DAGs & causal thinking. If not a feedback loop / bidirectional between A & B, what is the terminology for “I did B because A1 happened” then another instance of A happened (A2) & reason for that was “due to B”?

SATISFY EASY LADDERS: accessible machine learning to facilitate analysis of larger sensitive text data samples in criminology Large-scale data analysis has advanced criminology, but context-rich text is largely analysed manually in small samples. Machine learning (ML) with natural language processing (NLP) allows scalable...

9/n
Paper: doi.org/10.1080/3067...

Code & suggested resources: github.com/foxnic/acces...

8/n
Why It Matters

This framework will hopefully help more criminologists, police & others to conduct reliable, secure, scalable text analysis, to uncover important mechanisms & contexts (in bigger samples = better evidence base) to inform improved safeguarding of vulnerable children/adults.

7/n

Impact:

- ML flagged 60% of documents for school exclusion; human review corrected this to 28%. Shows how crucial human oversight is - without it conclusions could be wrong. Big diff between majority & minority!

- Accurate data analysis is vital for policy, resource allocation & safeguarding

6/n
Did It Miss Anything?

At the sentence level, there were some false negatives but the associated documents still had other correctly identified sentences so no documents were misclassified. Precision was the main challenge, especially for SEND and exclusion, but human review fixed these issues.

5/n
How Well Did It Work?

- ML beat keyword searches on multiple metrics (F1 scores up to 0.924)

- Decision trees (a simple ML algorithm) delivered best results

- Human review of ML-labelled sentences corrected mistakes made by the ML models

- Only 5% of ML-labelled sentences needed checking

4/n
The framework was tested on 193 child safeguarding practice review reports (prev. known as serious case reviews). It reliably identified issues like exploitation, missing incidents, school exclusion, & special educational needs (SEND), outperforming basic keyword searches.

3/n
Few ML/NLP resources address these crime text data constraints. This new framework (+demo) addresses this using simple, explainable ML designed for small, sensitive datasets. It’s for ppl w/out extensive ML expertise or time to learn, but who want to learn a bit or collaborate with ML experts.

2/n
Many "AI" tools e.g ChatGPT are off-limits for sensitive crime data (privacy risks + unreliable). Fancy ML methods require resources/skills that underfunded agencies/researchers lack + complex models are hard to trust or explain their results, & aren't always better than simpler algorithms.

1/n
Criminologists use police reports, court records, & other text documents to understand crime. But manually analysing these is slow, limits how much data can be studied, & delays insights. Machine learning (ML) could help, but many tools need huge training datasets and deep technical expertise.

Brief Report SATISFY EASY LADDERS: accessible machine learning to facilitate analysis of larger sensitive text data samples in criminology Nicola Fox ABSTRACT Large-scale data analysis has advanced criminology, but context-rich text is largely analysed manually in small samples. Machine learning (ML) with natural language processing (NLP) allows scalable analysis, but criminological use remains limited...

The 2nd* paper from my PhD has been published as an open access Brief Report! :)

Topics: SEND, exclusion, exploitation, missing children, machine learning ("AI")

Paper: doi.org/10.1080/3067...

The thread below attempts to explain what its about in a way that's accessible

*first solo author paper

Beyond data ceilings: rethinking how we understand missing persons Missing persons research has expanded substantially over the years, yet its analytic focus has changed quite little. This essay introduces the concept of data ceilings to examine how what is known ...

www.tandfonline.com/doi/full/10....

AI has supercharged scientists—but may have shrunk science Analysis of 41 million papers finds that although AI expands individual impact, it narrows collective scientific exploration

This analysis of papers from 1980 to 2025 goes beyond LLMs, but notes one challenge that arises when research questions rely on the same tools and datasets: “the adoption of AI seems to induce authors to converge on the same solutions to known problems rather than create new ones.”

"The hardest single part of building a software system is deciding precisely what to build. [...] No other part of the work so cripples the resulting system if done wrong. No other part is more difficult to rectify later."

Fred Brooks, "No Silver Bullet"

'The Effect of Counting Rules on Cross-National Comparisons of Homicide', by @davidbuil.bsky.social (who came up with the idea & did all of the data analysis and most of the write up), myself & @marcelo-f-aebi.bsky.social, just out at Social Indicators Research (link.springer.com/article/10.1...).

'Chatty Cafes help tackle the loneliness epidemic', says founder A mum from Oldham set up Chatty Cafes after herself feeling lonely while on maternity leave in 2016.

Does this fit the definition of analog social media?

www.bbc.co.uk/news/article...

Congratulations! 🥳