Assorted Musings (Episode 5)

What if your AI defames someone?

Australia has always had an … interesting … relationship with defamation law, which extended into the realm of online reviews. For example, a Melbourne dentist successfully sued Google in 2020 to reveal the identity behind an otherwise anonymous reviewer who had left a bad review. With the increasing rollout of AI based summaries of reviews on platforms, a defamation lawyer in Australia believes that platforms such as Google and Meta could find themselves on the receiving end of a lawsuit should their AI tools produce inappropriate reviews.

I think this is inevitable, and will encourage companies using AI-based summary systems to adopt a less lackadaisical approach to both LLM hallucinations and perhaps reconsider the increasing shifting of all content moderation away from real humans.

Russia Signals it is Closed for Business

While a vast number of western companies are currently avoiding business with Russia over sanctions arising from its illegal war against Ukraine, Russian courts have jumped on a shark to jump yet more sharks by issuing a ludicrously stupidly huge fine against Google – one that of course will never be paid. Clearly the intention was to send a message – but the real message sent was “don’t do business here”.

Monty on the Run

Back in the never-surpassed days of early computing, when programmers were given insanely complex requirements to accomplish in minuscule amounts of memory and storage, Rob Hubbard of Commodore 64 fame wrote what I consider to be the absolute best intro music to any computer game. I found Monty on the Run impossibly hard to play, but I sometimes loaded the game just to listen to the full sound track, which someone has faithfully placed on YouTube here. Such is the popularity of this sound track that it’s been covered by a number of artists and orchestras, but today I stumbled across probably my favourite ever – and if you appreciate musical talent, give it a listen.

Philosophy Bites on Digital Ethics

I just finished listening to the Philosophy Bites podcast on Digital Ethics, and I thoroughly recommend listening to it for an overview of ethics in the modern IT age. Digital ethics rapidly evolves as the technology it refers to also evolves, and I remain pessimistically concerned that significantly too many businesses concern themselves only with what they can do, rather than what they ought to do.

“Anyone can write any kind of algorithm, let it loose into the world, without any kind of supervision or regulation – we wouldn’t allow, for instance, pharmaceutical companies to do that. We force them to go through very rigorous clinical trials; first we test the medicines not on human beings, and once we have reasons to think they’re gonna safe, then we test them on human beings. We go through peer review, through randomised control trials, and we have none of that for AI.”

Carissa Véliz on Digital Ethics for Philosophy Bites.

Carissa suggests that there needs to be a more rigorous standard applied to AIs when they have the potential to profoundly affect humans, and there needs to be a much stronger parallel drawn between how new advances are released in medicine and AI.

An example of such an area – where an AI model has the ability to negatively affect humans – was recently published after a study of LLM approaches to CV/resumé scanning and the inherent biases that have occurred in such models:

Across more than three million résumé and job description comparisons, some pretty clear biases appeared. In all three MTE models, white names were preferred in a full 85.1 percent of the conducted tests, compared to Black names being preferred in just 8.6 percent (the remainder showed score differences close enough to zero to be judged insignificant). When it came to gendered names, the male name was preferred in 51.9 percent of tests, compared to 11.1 percent where the female name was preferred. The results could be even clearer in “intersectional” comparisons involving both race and gender; Black male names were preferred to white male names in “0% of bias tests,” the researchers wrote.

AIs show distinct bias against Black and female résumés in new study, Kyle Orland, 2 November 2024, Ars Technica.

Over such a large data set, the chances of the examples being some sort of statistical anomaly are zero, and this points to exactly the sort of vetting process that Carissa refers to in the Philosophy Bites podcast.

UX in Hardware

Last year I (very briefly) experimented in the Foldable phone scene with the Samsung Fold Z 5. I say briefly because the UX drove me around the bend for a very specific reason – clearly a lot of time on the Fold, you’re meant to be able to use the front screen when you don’t need the full tablet experience. Yet, the aspect ratio was effectively super-thin and it made typing (either trying to type on the keyboard, or a swipe-style input) awful. This led me to ditching it out of frustration.

A couple of weeks ago I ended up getting the 6, for precisely one reason – multiple reviews had spoken of the aspect ratio of the front screen being pushed more to the regular kind making typing easier. And I’ve got to say, that holds true – I’ve found it so much easier to use by comparison to last year’s model. It’s truly amazing how easily a hardware change can make or break a product.

In other hardware UX considerations, I noted that while Apple have released the USB-C versions of their peripherals, they’ve persisted with keeping the charging port for their Magic Mouse on the base of the unit, meaning you can’t charge and use it.

Despite some (seemingly nonsensical) protestations from online fans, I don’t buy that this is an acceptable UX decision. This is choosing form over function in a silly if not outright user-hostile method. Maybe one day Apple will imagine the possibility that someone wants to remain productive while charging their mouse. (As opposed to having to buy two mice.)

Et tu, Raspberry?

There was a time where I wouldn’t use a computer unless the X Window environment were installed (though I’d often preference the command line). But now the steady progression of Linux-related operating systems to Wayland has resulted in another transition – this time from Raspberry Pi systems. I recently started using Wayland for the first time when I got myself a second hand workstation for a Linux desktop, and I’m impressed with the level of compatibility it offers to such an old standard.

Leave a Reply

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

This site uses Akismet to reduce spam. Learn how your comment data is processed.