Adam's Brain Dump

Whilst separated by hundreds of years, I keep noticing distinct parallels between modern conceptions of AIs, especially those of a more doomer persuasion, and the long history of myths around for example golems.

Many such stories originate from Jewish folklore. The Jewish Museum in Berlin describes a golem as follows:

…a creature formed out of a lifeless substance such as dust or earth that is brought to life by ritual incantations and sequences of Hebrew letters. The golem, brought into being by a human creator, becomes a helper, a companion, or a rescuer of an imperiled Jewish community. In many golem stories, the creature runs amok and the golem itself becomes a threat to its creator.

Make a handful of changes, perhaps swapping out “sequences of Hebrew letters” with “sequences of computer code”, and “an imperiled Jewish community” with “humanity” and there we go. It’s certainly not hard to see one of the more famous such myths, the Golem of Prague, as a kind of AI alignment problem.

The story of Frankenstein feels like yet another warning from fiction about the perils of creating something supposedly akin to a human. From the Wikipedia summary:

Victor buries himself in his experiments to deal with the grief. At the university, he excels at chemistry and other sciences, soon developing a secret technique to impart life to non-living matter.

…

Despite Victor’s selecting its features to be beautiful, upon animation the Creature is instead hideous, with dull and watery yellow eyes and yellow skin that barely conceals the muscles and blood vessels underneath. Repulsed by his work, Victor flees.

Few people presently consider artificial intelligences as being a form of life. But that aside, a couple of weeks ago the “Godfather of AI”, Geoffrey Hinton, also fled, repulsed by his work. Or at least quit Google, sorry for what he had done.

…citing concerns over the flood of misinformation, the possibility for AI to upend the job market, and the “existential risk” posed by the creation of a true digital intelligence…Hinton, 75, said he quit to speak freely about the dangers of AI, and in part regrets his contribution to the field.

The Observer reports the extremely chilling fact that the stress caused by Ofsted school inspections has been cited in at least 10 coroner’s reports concerning the deaths of teachers.

The topic has been brought to light by the recent case of Ruth Perry, a headteacher who killed herself after her school rating was downgraded by Ofsted.

Death is of course only the most extreme health outcome of these inspections:

…an Observer investigation has found that the pressure of school inspections has led to headteachers suffering heart attacks, strokes and nervous breakdowns, and as a helpline for heads reports that the vast majority of crisis calls it receives are now about Ofsted.

Sadly this is not at all surprising to me when I think about the teachers I have personally known. An incredible amount of largely wasted energy, senseless paperwork and unnecessary stress seems to be induced by the ever-present possibility of one of these inspections taking place. It’s hard to imagine that teaching quality doesn’t suffer as a consequence.

Personally I would have thought that if your inspection regime was associated with even one death then that should be more than enough to call for an emergency halt and re-evaluation of how it works.

Or even if it should exist at all. A quick search suggests to me that there’s not much evidence out there that Ofsted inspections are even all that effective when it comes to educational outcomes.

A 2016 report by the Education Policy Institute suggests that Ofsted ratings are largely dependent on the school’s intake. Amongst other things they find:

…a significant inconsistency between the identified deterioration in academic standards and the resulting Ofsted judgement

…

schools with more disadvantaged pupils are less likely to be judged ‘good’ or ‘outstanding, while schools with low disadvantage and high prior attainment are much more likely to be rated highly…

As Frank Coffield notes for the British Educational Research Association, amongst other harms, this creates a perverse incentive:

The very schools that need most help are further harmed by inaccurate and biased Ofsted reports that make the recruitment and retention of teachers even more difficult.

Overall he considers that Ofsted “does more harm than good” and uses methods that are “invalid, unreliable and unjust”.

In a different study, Leslie Rosenthal finds that Ofsted inspections do have an impact on exam performance. There’s just one problem:

It is found that there exists a small but well-determined adverse, negative effect associated with the Ofsted inspection event for the year of the inspection.

And there’s no detectable effect apparent in the exam results for the year after the Ofsted inspection occurred.

I’ve recently been using using Visual Studio Code to interface with remote Github Codespaces when developing data models. Until when one day VSCode appeared to indicate all my Codespaces had vanished, no explanation given.

It turns out that I needed to update my VSCode version. This hadn’t occurred to me as I thought I had it set to auto-update. Which the settings indicate it I had but yet for some reason it hadn’t updated for several months. Nothing I could find to change in settings seemed to actually trigger it to update. There was also no “Check for updates” option in the menu.

A light bit of Googling suggests this is maybe something to do with Windows and admin mode, although I don’t feel inclined to dig further.

Here’s how you can update VS Code in these situations. I just copied and pasted this from a colleague so don’t intend to try and explain it. But all you have to do is open a Windows command prompt and type

winget upgrade --id Microsoft.VisualStudioCode

and an updated VS Code will download and install.

winget is a built-in package management tool to some versions of Windows 10 and 11 that helps with installing and upgrading applications. If you didn’t know the –id for VSCode you could have found it by running:

winget search vscode

King Charles III’s coronation day. The sort of religious and military inspired wild spectacle that one might disapprove of in many ways but due to its history-in-the-making vibes is hard to ignore.

Everything appeared to go very smoothly, although to what extent this was as a result of some rather disturbing sounding anti-protest policing with the code name ‘Operation Golden Orb’ I don’t know.

Scotland Yard has been accused of an “incredibly alarming” attack on the right to protest after police used new powers to arrest the head of the leading republican movement and other organisers of an approved demonstration just hours before King Charles III’s coronation

Some of the less aminate stars of the show were the various regalia of the monarchy. Part of the ceremony involves our new king touching various items that seem very similar in name to those I encountered whilst playing The Legend of Zelda a while back.

The i website kindly lists some of the gold-and-jewel encrusted items involved. The include the:

• Long Sceptre, remade in 1661 for the previous King Charles.
• Sword of Temporal Justice: signifying his command of the armed force.
• Sword of Spiritual Justice: our monarch remains the ‘defender of the faith’
• Sword of Mercy: nothing says mercy like a sword with a broken-off tip.
• Sword of State: decorated with lions and unicorns.
• Jewelled Sword of Offering: symbolising duty, knightly virtues and power. Also subject to a slightly odd ceremony whereby ex-Conservative leadership candidate Penny Mordaunt trades a bag containing £50 worth of brand new 50p coins for it.
• Coronation Spoon: used to convey magic oil onto the monarch.
• Golden Spurs: symbolising knighthood and chivalry
• Bracelet of Sincerity
• Bracelet of Wisdom
• Sovereign’s Orb: a jewel encrusted sphere with a cross, symbolising that the king’s power comes from God.
• Sovereign’s Ring: symbolising dignity.
• Sovereign’s Sceptre with the Cross: a shoutout to earthly power.
• Sovereign’s Sceptre with Dove: another symbol of justice and mercy.
• finally, of course the 5lb St. Edwards Crown, resized to fit his head.

Yesterday much of the UK saw local elections taking place. Counting is still in progress but so far we’re really just waiting to see whether the Conservatives lost quite a few or a huge number of seats. With 82 of 230 results in they’re down 279 councilors.

There was an unexpected upset though. Graham Galton, the Conservative candidate for Coxford in Southampton, unfortunately died between the time the polls opened and their closure. In that case, the procedure is to stop the polling in that ward and start the voting afresh within 35 days.

Whilst dying at the same time people are in the process of voting for you appears to be a fairly rare occurrence, it seems that dying after being nominated may be less so. Apparently a further 7 candidates for these elections died between they date they were nominated and the day of the election itself.

They’ve made King Charles' coronation interactive.

Presumably in an effort to promote diversity and equity (haha), the traditional “homage of the peers” where fancy aristocrats pledge allegiance to the new monarch is to be replaced by the “homage of the people”. This time we’re all invited to stand in front of our TVs and pledge our humble devotion to the magic king man.

So on Saturday, expect a chorus of British-accented voices to cry out these words:

I swear that I will pay true allegiance to Your Majesty, and to your heirs and successors according to law. So help me God.

Or not, if one believes a very unscientific poll.

One additional reason why it’s possible that generative AI technology might permanently infuse itself throughout the world in the foreseeable future is that anyone who uses a computer for almost anything to do with their job might soon find themselves using it by default. Both Microsoft and Google are adding the technology to their extremely popular, extremely mainstream, office software; Office and Docs respectively.

Google’s “new era” blog post has the standard revolutionise-everything pitch:

We’re now making it possible for Workspace users to harness the power of generative AI to create, connect, and collaborate like never before.

What does this mean in practice? Here’s their bullet-point list of what’s coming to testers this year.

• draft, reply, summarize, and prioritize your Gmail
• brainstorm, proofread, write, and rewrite in Docs
• bring your creative vision to life with auto-generated images, audio, and video in Slides
• go from raw data to insights and analysis via auto completion, formula generation, and contextual categorization in Sheets
• generate new backgrounds and capture notes in Meet
• enable workflows for getting things done in Chat

If you prefer to watch someone automate potentially most of their workplace written output - including asking the AI to go with more whimsical vibes at times - via video then here we go.

It’s billed as an assistant rather than a creator, with features including the ability to draft a document for you to review, or rewrite a document you created in another style. Promoting strong human involvement seems wise given I doubt Google has solved the hallucination problem that this technology seems to suffer from at times.

Honestly, for those of us who have had to write a job description without being a specialist in that art, the idea of just typing in “Write me a job description for an X” and receiving something passable in response rather than having to start with a blank page is quite appealing. A cynic might question whether automated “fluent bullshit” is any worse than the human kind.

Then Microsoft have their take, billed as Microsoft 365 Copilot. This one’s blog post announcement promises to “turn your words into the most powerful productivity tool on the planet” no less.

Similarly to the Google offering:

Copilot can add content to existing documents, summarize text, and rewrite sections or the entire document to make it more concise.

They talk more about how because it will (optionally? Who knows?) have access to your existing documents, emails, calendar and so on you will be able to ask it to do things like “Draft a two-page project proposal based on the data from my_document.doc and my_spreadsheet.xls”.

In Excel, it’ll let you ask questions about your data; access to your very own virtual analyst. For instance: “Give a breakdown of the sales by type and channel. Insert a table.”

Outlook is going for a new personal assistant feel: “Summarize the emails I missed while I was out last week. Flag any important items”.

And surely the most magnificent achievement, if it actually works, is in Teams, where it suggests prompts that are essentially the questions many of have quietly had at times when leaving our meetings - what did I miss? What did we actually decide?

In pastel-coloured video form:

Previously Microsoft also announced a new product, Microsoft Designer, which enables Office users to create AI-generated art, a la Dall-E, Stable Diffusion et al.

Honestly, some of this stuff looks magical in the same way that chatGPT does on occasion. But having used tools that promised to do some of this many years ago that didn’t end up revolutionising the world, the proof will of course be in the pudding.

It’s not like it’s the first foray into AI-ish assistants these office suites have had. Already Gmail users may be familiar with the “suggested responses” one finds at the bottom of certain emails - so-called Smart Reply.

With this feature Gmail gives you some contextual auto-reply options.:

Right now, my old-fashioned brain often finds them slightly repellant. The automatic “Sounds good” also “Sounds insincere”. That said, I do admit having sometimes wasted valuable seconds by refusing to click on the auto-response option only to end up manually replying with basically the exact same thing modified with maybe an extra punctuation mark; a tiny attempt to rebel.

For Microsoft, we, the citizens of the Internet, will never forget Clippy, your fun animated paperclip assistant and buddy, never far away, surveilling you just in case it ever looks a bit like you’re writing a letter.

Clippy - born in Microsoft Office 1997, died in Office 2003, RIP - turned out to be extraordinarily unpopular. Wikipedia quotes Alan Cooper as saying that the assistant was based on a “tragic misunderstanding” of research. Luke Swatz wrote a literal thesis called “Why People Hate the Paperclip” opening with a quote: “I hate that #@$&%#& paperclip!”.

But yet his legacy lives on in countless memes, including this relatively modem instantiation, previously circulating on I guess at least 48% of UK political Twitter.

His legacy will also continue more formally - he’s one of the emoji in a recent Windows 11 update. Oh, and true fans will already know of the “erotic” novel he’s featured in, “Conquered by Clippy” (I’m so sorry).

You might notice I used “his” when referring to Clippy above. This wasn’t an accident or another example the default male. Although the internet was once besieged with a picture of a pregnant Clippy, his creator refers to Clippy using the “he” pronoun.

Interestingly the whole Microsoft Office assistant thing fronted by Clippy apparently polled particularly badly with the women in focus groups.

Quoting Roz Ho, one of the very few women involved, from a documentary “Code: Debugging the Gender Gap”, the Atlantic reports:

Most of the women thought the characters were too male and that they were leering at them. So we’re sitting in a conference room. There’s me and, I think, like, 11 or 12 guys, and we’re going through the results, and they said, ‘I don’t see it. I just don’t know what they’re talking about.’

Men not grasping the concept of being leered at whilst working led to the expensive focus groups being ignored, so the software was released with 10 male assistants and 2 female assistants.

An informal poll of people I know gave a 100% verdict that Clippy was male. Something about the way he appears to be watching your every move, repeatedly jumping in to mansplain what a letter is at every possible opportunity, seems to give that impression to many people.

Contrast that with the “personalities” of the more modern voice assistants like Siri, Alexa et al. which usually defaulted to female-presenting personas in most places. Whilst still something of a surveillance style technology, they’re designed to remain silent and invisible in the background until you specifically bark an order at them. They’ll respond to even verbal abuse in a friendly and flirtatious way, at least in their original versions.

To quote parts of a 2019 UNESCO report:

…AI systems that cause their feminised digital assistants to greet verbal abuse with catch-me-if-you-can flirtation

…

When asked, ‘Who’s your daddy?’, Siri answered, ‘You are’.

…

Because the speech of most voice assistants is female, it sends a signal that women are obliging, docile and eager-to-please helpers, available at the touch of a button or with a blunt voice command like ‘hey’ or ‘OK’. The assistant holds no power of agency beyond what the commander asks of it. It honours commands and responds to queries regardless of their tone or hostility.

To be fair, apparently at least Apple has toned down Siri in this respect. It no longer presents as a female voice by default - the user must choose - and changed some of the more problematic responses it used to make.

Should you be one of the handful of people that miss Clippy, or, much more likely, someone that likes winding other people up, you can download a “prank” Clippy here that pops up independent of the usage of Microsoft Office with helpful messages like “Your computer seems to be turned on” - although it doesn’t seem to have been updated since the era of Windows 7 so YMMV.

Of course the future doesn’t always reflect the past. The technology behind this year’s office-suited AI helpers is different and radically more powerful than a 20 year-old paperclip (so much so that some people worry that further developments in AI could lead to all humans being turned into paperclips). But bias remains. And generative text efforts so are prone to occasionally spouting nonsense. A lot of its usefulness may depend on its interface as well as its output. I’m looking forward to seeing how it all turns out.

A fiery tulip 🌱.

🎶 Listening to Trustfall, by P!nk.

This was one of those occasions where I was curious what one of the musicians of my youth was up to and it turns out they just released a new album all these years later.

I’m not sure that this one is going to be a iconic as some of her earlier ones, unless that’s just nostalgia talking. But I do remember a few years ago thinking (/ despairing) that one of her older songs, “Dear Mr. President” was uncannily apt a decade or more after its release.

I mean:

How can you say, “No child is left behind”?

…

They’re all sitting in your cells

…

What kind of father would take his own daughter’s rights away?

And what kind of father might hate his own daughter if she were gay?

Believe it or not, this was released in 2006 as a critique of US President George Bush, not in recent times in reaction to this, this and this. There truly is nothing new under the sun.

Anyway, that song isn’t on the new album, for which I have more mixed feelings. There’s a few I like, so I’m sure I’ll listen to it a few times, but I don’t know that I’ll remember it a decade later as much as a couple of her earlier ones have stuck with me.

That said, some of my favourite songs in it involve the theme of how dancing is wonderful, for instance:

For these I absolutely applaud her for defying the extremely upsetting “research” reported on by Bristol Live a few years ago whereby:

…37 is the age it becomes tragic to go to nightclubs, with 31 emerging as the age we officially prefer staying in to going out.

This despite the fact that dancing at the age of 75+ is in fact associated with a lower risk of dementia, amongst other health and wellbeing benefits.

📚 Finished reading The Signal and the Noise by Nate Silver.

Nate Silver describes efforts to forecast events from a wide range of domains - everything from baseball to terrorism, from global warming to poker games. This includes a chapter on forecasting contagious diseases, some of which, despite being published in 2012, is probably all too familiar in a world where an otherwise surprising number of people have unfortunately had to embed the concept of R0 and case fatality rates into their psyches.

In doing so he largely (very deliberately) shows the limits of our ability to accurately predict things. After all, it’s not something humans have evolved to be good at. We’re optimised for survival, not for being able to mathematically rigorously predict the future. And some things are inherently easier to predict than others. We’re increasingly good at the short term weather forecast; we’ve made very little progress when it comes to earthquakes.

Naively, one might think that because we have access to so much more information today than ever before - with each day increasing the store of data by historically unimaginable amounts - that we should be in a great place to rapidly learn vastly more truth about the world than ever before.

But data isn’t knowledge. The amount of total information is increasing far, far faster than the amount of currently useful information.

We’re able to test so many hypotheses with so much data that even if we can avoid statistical issues such overfitting to the extent of being fairly good at knowing if a particular hypothesis is likely to be true or not, the fact that baseline rate of hypotheses being true is probably low means that we’re constantly in danger of misleading ourselves into thinking we know something that we don’t. This argument very much brings to mind the famous “Why Most Published Research Findings Are Wrong” paper.

As Silver writes:

We think we want information when we really want knowledge.

Furthermore, bad incentives exist in many forecast-adjacent domains today. Political pundits want to exude certainty, economists want to preserve their reputation, social media stars want to go viral, scientists want to get promoted. None of these are necessarily drivers that necessarily align with making accurate predictions. Some weather services are reticent to ever show a 50% chance of rain even if that’s what the maths says; consumers see that as “indecisive”.

Silver’s take is that we can improve our ability to forecast events, and hence sometimes even save lives, by altering how we think about the world. We should take up a Bayesian style of thinking. This approach lets us - in fact requires us - to quantify our pre-existing beliefs. We then constantly update them in a formalised way as and when new information becomes available to us, resulting in newly updated highly-quality beliefs of how likely something is to be true.

Our thinking should thus be probabilistic, not binary. It’s very rare that something is 0% predictable or 100% predictable. We can usually say something about some aspect of any given future event. As such we should become comfortable with, and learn to express, uncertainty.

We should acknowledge our existing assumptions and beliefs. No-one starts off from a place of no bias and this will inevitably influence how you approach a given forecasting problem.

There’s almost a kind of Protestant work ethic about it that forecasters would be wise to adhere to: work hard, be honest, be modest.

One can believe that an objective truth exists - in fact you sort of have to if you are trying to predict it - but we should be skeptical of any forecaster who believes they have certainty about it.

Borrowing from no less than the Serenity Prayer, the author leaves us with the thought that:

Distinguishing the signal from the noise requires both scientific knowledge and self-knowledge: the serenity to accept the things we cannot predict, the courage to predict the things we can, and the wisdom to know the difference.

My more detailed notes are here.

Knitted crowns aplenty outside the shops today, no doubt in anticipation of the forthcoming coronation of King Charles III.

I wonder if the wool is as vegan as his chrism oil will be.

🎙 Listened to “The Most Amazing - and Dangerous - Technology in the World” from The Ezra Klein Show.

No, not AI, nuclear bombs or bioweapons, at least not directly. Here the discussion is about semiconductors, used to make computer chips.

Ezra talks to Chris Miller who has written about how chips “created the modern world”, infused as computer technology is into a vast array of domains of modern life. He’s the author of the book “Chip War”.

What’s probably the most mind blowing takeaway from it all is how fragile the chip manufacturing infrastructure is, particularly how few entities are capable of making especially the more advanced chips used in the modern world. If AI is indeed the huge deal many people believe it is then one can see that any vulnerabilities in the advanced chip supply chain may be extremely impactful.

A couple of examples:

• 90% of the most advanced chips are produced by a single company, TSMC. TSMC is a Taiwanese company, a country that is not exactly in the most reassuringly stable condition at the moment. The UK foreign secretary recently warned that should China invade Taiwan it would destroy world trade, let alone the ability to make advanced chips.
• Every EUV lithography machine, essential to the manufacture of cutting-edge chips, is supplied by a single company - the Dutch firm ASML.

Chip manufacture is thus very vulnerable and also a tempting target for weaponisation. The podcast has examples of how the US and China combat each other along these lines.

A day with a huge personal revelation: adding some ground black pepper to a cup of standard black tea results in a surprisingly delicious drink.

Loved ones tell me that it’s a tradition in some parts of Africa and also that it may have health benefits. But my immediate take is: it’s really tasty.

The increase in the percentage of the US population who approve of the idea that people of different races should be able to marry freely is one of the largest changes in opinion that Gallup has seen in its polling over the years.

From the 2021 survey, 94% approve of marriage between Black and White people, up from 4% (!) in 1961.

Predictably, most of this is driven by changes in White people’s opinions, although there were previously substantial-but-smaller proportions of people in other groups who didn’t approve of marrying across racial lines either.

It’s easy to overlook how some things have changed substantially for the better, particularly whilst there remain so many things obviously wrong with the world of today. It somehow feels absolutely inconceivable that even within my own short lifetime there were extensive periods where the majority of Americans disapproved of people of differing races getting married.

That said, it was only last year that interracial and same-sex marriage rights were codified into US federal law, driven by the fear that the same kind of Supreme Court decision that overturned the constitutional right to abortion might be weaponised against the kind of marriage rights that a certain type of conservative would rather didn’t exist either in the future.

It was after all a Supreme Court decision that enabled the recognition of many American interracial marriages in the first place. Incredibly, it’s only a little over 50 years since interracial marriages were literally illegal in many parts of the US. That changed in 1967, when in Loving v. Virginia the Supreme Court ruled that bans of that sort violated the constitution.

📺 Watched The Last Of Us.

I’m not normally a big fan of zombie shows, but this one was getting such exceptional reviews I thought I should give it a go. I’m glad to have done so.

The premise is that 20 years ago increasing global temperatures led to some kind of parasitic fungus (“Cordyceps”) jumping species from insects to humans which results in them turning into something akin to the classic bite-others-to-infect-them zombie. The condition has thus spread far and wide and remains so far incurable, although there is technology to diagnose whether or not one is infected before symptoms appear. But it’s not all about jump-scares and mowing down the undead with lawnmowers, there’s a depth and occasional humour here, along side an exploration of much more human relationships, that makes it quite enthralling to watch.

An interesting dynamic occurs due to much of the show being set 20 years after the fungus began its ravages. Society has changed in many way of course, in some ways fairly predictably so. But additionally there’s a generation of children and young adults that have never lived in any other world, and thus operate from very different reference points to our own.

The post-apocalyptic world is rendered very well in visual terms, with the impact of nature’s re-taking over some of humanity’s former accomplishments that society is no longer able to maintain very visible. The appearance of the zombies themselves I have to admit finding a little comical at times, although the way they move, jerky and fast, is certainly creepy.

It’s based on a computer game, but don’t let historical precedent of game-to-show efforts put you off. It’s now one I’d love to play, it also being very much award-winning, although it looks like I’d need to get a PlayStation to do so.

Bobby Geraghty, singer and songwriter for former band Breezer, got bored of waiting for the 90s Britpop absolute sensation Oasis to reform. And so AIsis' “The Lost Tapes Volume One” came to be.

Liam Gallagher is back fronting Oasis/AIsis with some classic sounding tunes, except of course it’s not really Liam Gallagher. It’s an AI.

From Loudwire:

“All I had to do was replace my vocals with Liam’s,” he added. Geraghty spliced various a cappella recordings of Liam to train his AI version of the singer, then layered it overtop Breezer recordings.

And there we go, Oasis kind-of reincarnated.

The reaction seems to be pretty positive in general. Even Liam said it was “Mad as f**k I sound mega”.

To be clear, the songs, music and lyrics were all written and performed by a human band in the first place, the afore-mentioned Breezer. The fact that it’s fake-Liam that appears to be singing it is the only AI part. But it’s fairly convincing to my ear. I’m sure it’s only a matter of time before some band is revealed to entirely non-meat based.

Stability.ai is the company behind AI text-to-image generator Stable Diffusion. This was the first widespread version of this technology that was both open source, meaning anyone who knows how to can modify it however they like, and optimised enough to be able to freely run entirely on your own computer, a step that no doubt increased the popularity of this type of tool.

It’s also available for use in the cloud if you prefer the simplicity of installing nothing. There’s a basic version available on HuggingFace.

Anyway, they’ve just released StableLM.

Our StableLM models can generate text and code and will power a range of downstream applications.

So this is a similarly open-sourced AI large language model, akin to chatGPT et al. Only in this case it’s free for you to download and/or adapt . Once again you can play with a cloud version if you prefer.

Right now it’s apparently a lot simpler and hence less ‘accurate’ than chatGPT itself. But as it’s open source, anyone with the relevant knowledge is able to help diagnose and improve it. So quite possibly we’ll see rapid increases in its ability over time - no doubt alongside more troubling adaptations, the world being what it is in 2023.

Stability.ai themselves plan to release more complex models soon.

📚 Finished reading Unlawful Killings by Her Honour Wendy Joseph KC.

Former Crown Court judge Wendy Joseph takes us through 5 fictional-but-based-on-reality trials of the type she presided over. The common theme is their relationship to a charge of murder.

The reader gets the intrigue associated with courtroom drama in general, alongside some more analytical legal tutoring as to how the court system works in practice, the factors that affect how cases are decided, what legislation applies and when. What does “not guilty” mean? When is being sure required, vs on the balance of probabilities?

There’s the case of the young man, a victim of knife crime, stabbed to death whilst on a lunch break. We see a mother who a witness believes was trying to suffocate her baby. Next up is another young man, who killed his friend in a car accident. Then the story of the parents whose child was found dead after a explosive family row. A solider appears to have killed his wife rather than let her leave him. And finally a wife who shot her allegedly career-criminal husband.

In each case we see that nothing is as simple as it sounds; very little is black or white. Often the people that kill are not inexplicably evil psychopaths who delight in pain, but people who have found themselves in extremely difficult situations, some of which may be unimaginable to a lot of us if we’re lucky.

Complications, both moral and legal ensue; mental illness, cultures of honour, infanticide, levels of diminished responsibility and gang life amongst others. We see witnesses who won’t testify, experts who disagree with each other, and cases resting on the words of vulnerable young children. Occasionally there are tensions between the agreed procedures of law, the incentives for lawyers and defendants, and what an external observer might consider as just. Nonetheless, the defendants are all alleged to have caused some of the greatest harms it is within our power to cause. At the end of the day it’s for a random selection of us, the jury of their peers, to determine their guilt, and the judge to weigh in on the consequences.

In her telling of the stories we’re introduced to the main characters present in each such trial: judge, defendant, solicitors, barristers, witnesses, clerks, ushers and the all-important jury. Appendices cover some of the relevant legislation and its interpretation. These days it seems judges may be quite constrained as to what they can do. Joseph appears to see her primary role as managing the proceedings of the trial, ensuring the rules are adhered to by all concerned, herself included.

One of her responsibilities is providing a “path to verdict”, translating the complexities of centuries of legal legislation to a set of decisions followable by a largely untrained jury as they make their determination, several of which are reproduced in this book.

The last chapter focuses on what Judge Joseph seems to see as a key failure of the system. Unlike the author of the previous book I read she’s no prison abolitionist. But I understand that she would like to see far fewer cases brought to court, being on the same side as Mariame Kaba in terms of seeing it as very often a failure of society that something harmful enough to warrant a crown court case happened in the first place.

Wrongdoers don’t fall from the skies…they are formed not just by their capacities but by their experiences…particularly difficult experiences in childhood and adolescence. In some cases, if we are honest, we must accept we have allowed them to become what they are.

If we create societies that alienate vast swathes of people, refuse to provide care to people - potential victims or perpetrators - at risk from mental illness, enable black market trades rife with violence whilst leaving large sections of the population destitute, vulnerable and unsatisfied, then we can’t be surprised when sometimes bad things occur.

I think the author believes that we’ve done a reasonable job of identifying what things should be classified as crime. But that we’ve not done nearly enough in terms of preventing those things from occurring. More and more people are locked up in prison, their lives sometimes almost as ruined by the experience as those that they victimised. And yet, crime continues.

A key flaw she identifies in relying on the criminal justice system to manage crime is that, however progressive we make opportunities for restitution and rehabilitation, it doesn’t kick in until after the harm has been done, after the crime has been committed.

Blaming crime on an individual being greedy, dishonest or angry may be satisfying. But many of us are also greedy, dishonest and angry but do not commit crime. She asks us to consider why this is.

Why when people are confronted with a situation wherein they could commit a crime do some choose to do so and others do not? How have we engineered a situation whereby someone feels that their best option for living the life they want is to knowingly harm another?

Maybe, instead of clapping ourselves of the back for the excellence of our criminal justice system, we should accept that the commission of any crime is a mark that somewhere, somehow we have failed.

Last year an AI-generated picture won an art competition. The technology has of course only got better in the time since then. So inevitably an AI generated image has now won a photography competition, even though of course no photography was involved. Well, other than the no doubt huge number of photos the AI ingested as part of its training.

Photographer Boris Eldagsen won a prize at the Sony World Photography Awards with this entry:

He titled it “PSEUDOMNESIA | The Electrician”, the first part of which is a reference to a condition associated with pathological lying.

Mr Eldagsen is on board with the idea that AI generated images are not photography as such and hence is declining to actually take the award and associated prize. It seems he was more in it to generate some kind of discussion about how “co-working” with AI fits into the world of photography. After all, it’s been a long time since the rules of many photography competitions forbade the use of computers and digital modification entirely. And as it stands, there’s no reason to suppose that a more nefarious actor couldn’t win a competition with the help of AI image generation in the future without feeling inclined to reveal the true source of their image.

📺 Watched season 12 of Death in Paradise.

Everyone’s favourite vacation paradise / murder hotspot returns for more light-hearted death presented in a Agatha Christie-ish style except with a backing track of jolly Caribbean tunes and occasional slapstick chases.

So the same as it’s always been. It’s never been the most fashionable of shows, but it has been one of the most consistent and, perhaps surprisingly, most popular BBC shows of the past decade.

Really the worst I can say about it is its slight whiff of the White savior trope in a ‘clever and logical white man comes to solve foreign mysteries’ way. Although the gentleman in question changes every so often and is always a little weird or inadequate in stereotypically British other ways. But if you liked anything in the last huge number of series then there’s no reason not to watch this one. I found a couple of the later episodes actually fairly gripping.

Very much ‘familiar, warm and dependable’, as the Guardian says. Nice family-friendly unambiguously solvable murders in the sun, no Special Victims Unit to be seen.

Rome et al analyse the costs of newly-launched pharmaceuticals in the US. They discover that the median average launch price of a drug has increased from just over $2,000 per year in 2018 up to $180,000 in 2021. Almost half of drugs launched in 2020-2021 cost over $150,000 a year.

The mean average price increased exponentially during this period by around 20% per year. Some of this may be explained by differences in the characteristics of the drugs launched in each year; oncology drugs and those for rare diseases tend to be more expensive. But even with that taken into account as best as the researchers could do, there remained an exponential increase in annual cost of around 13% per year.

This is clearly unsustainable for anyone who is in the position of actually having to pay for these treatments. The researchers suggest that in line with some other similar countries “the US could stop allowing drug manufacturers to freely set prices”, which would seem to me to be an obvious and decent first step.

📚 Finished reading We Do This ‘Til We Free Us by Mariame Kaba.

In this book, Mariame Kaba presents a collection of short and readable articles - reprints of magazine or interviews for example - all dedicated to educating about and encouraging readers to take actions towards the total abolition of the prison system, and policing too. “Defund the police” taken to its most literal conclusion.

It’s a position I started off with some sympathy towards. I am more than happy to accept that far more people are in prison than should be. This book focuses on the US system, and that is almost indisputably true there.

I was already happy with the idea that the focus of the criminal justice system (or criminal punishment system, as she calls it - feeling there’s very little justice to be had in it today) should be on issues such as safety and restitution rather than punishment for punishment’s sake. Prison should be a last resort after all other options have been explored. There’s a lot of reform to what prison entails, and the life of someone post-prison, that should be urgently made. Likewise policing - again, a last resort, after other arms of the state whose role is to assist people in need have done all they can.

As far as I can see there is substantial evidence that much “crime” is driven by structural societal factors which opens many avenues for mass-scale prevention that could be invested to avoid incidents where people are even tempted to call the police occurring in the first place.

But I’ve never quite been able to push myself to the full abolitionist perspective that prison and policing are entirely unnecessary. Kaba would probably consider that a lack of imagination, which I’m happy to acknowledge. And so this book tries to provide the impetus for a full abolitionist outcome.

There are many aspects to this. Below is my attempt to summarise some of her main points.

The philosophical argument

Philosophically, the author entirely agrees that there should be consequences when people cause harm. However there is a difference between “consequences” that aim to help victims and make society at large and “punishment” that aims in part to make a perpetrator suffer. Sometimes they may look similar, but their goal is different and the latter is inhumane, however attractive it may be to those affected.

But the priority should be addressing the conditions that allowed the harm to take place and the actual needs of the person to whom it was done to. The default of people locking people in cages away from our sight-line is an abdication of the responsibility we all have to help those harmed. It hides important social and political failures.

Furthermore, the system does nothing to encourage accountability. In fact it works against it. People cannot be forced to accept accountability. The current adversarial court system discourages people from taking personal and public responsibility for their actions because it’s almost always in the perpetrator’s interest to deny or minimise the harm the caused, even when they’re fully aware and may like to make some kind of restitution. But if a guilty plea would result in inevitable imprisonment then they may find this unpalatable.

Many “criminals” are in fact victims of something themselves; the victim / perpetrator binary is a fiction. Most women in US prisons are poor, of colour and have experienced physical violence from a partner. Without lessening any harm they commit, we can acknowledge that some criminal actions come from someone making choices that they should never have been put in the position of making.

Hurt people hurt people; desperate people hurt people. Lessening social and economic inequities, and making other, non-criminalising, services easily available could go a long way to preventing the harm in the first place.

Neutralizing perceived threats, in an endless game of legal whack-a-mole, is not a path to safety. To create safer environments, people and circumstances must be transformed.

The legitimacy argument.

Firstly we should examine the definition of crime. It’s often arbitrary. Not everything that is criminalised is harmful. Not all harm is criminalised. A classic example of the latter is CEOs of companies who cause deaths by e.g. harmful, yet perfectly legal, environmental practices.

Additionally, prejudice is rife within the US criminal justice system, notably in the conflation of Blackness with criminality. Black people are more likely to be arrested and more likely to get harsh sentences than white people for the same offence. Black victims of crime are more likely to be ignored.

The current system doesn’t truly even try to meet the needs of victims. Services like counselling, healthcare, housing or money are rarely on offer.

Even in the domain of punishment the victim rarely has the final word. If the system requires the perpetrator to be be jailed then the are, irrespective of the victim’s wishes. The extreme case is someone being sentenced to death (remembering this book is mostly about the US) when the victim is virulent anti the death penalty. They may even feel like they have another person’s death of their conscience unless they drop the charges.

The rise of the “prison industrial complex” has produced economic incentives such for the system to perpetuate itself it needs to increase the number people imprisoned and the duration of their sentences.

Our response to violence and harm should not cause more violence and harm

The efficacy argument

Kaba writes a lot about sexual abuse. This is a crime so pervasive and in some cases so horrific that being seen to argue against an offender going to prison is almost blasphemous even in - or perhaps especially in - liberal circles.

But she raises fair points in that its very pervasiveness shows that the current system doesn’t work Very few reported cases in any case get anywhere near court. Those that do often in reality lead the victim being put on trial more than the perpetrator. These and other factors means who suffer this harm rarely even report it, demonstrating a total lack of faith in the only system that today exists to supposedly help victims of harm.

Harm caused by police officers - sexual misconduct is the second most common form of police violence - is very rarely punished.

The public does not understand the practicalities of policing. Most US officers make just a 1 felony arrest a year. Most of their time is spent dealing with non-criminal issues such as traffic citation, noise complaints or wellness checks. Many of these would be better done by non-police personnel who have better training in for example mental health issues.

In general, there is little-to-no evidence that prisons reduce violence, crime or people’s fear of it. The US prison population increased by over 400% between 1979 and 2014. Neither crime or the fear of crime is 400% lower now.

Some research suggests that going to prison increases the likelihood that you will commit crime again in the future. It’s not clear that they’re even a deterrent in the first place - criminals do not tend to be dispassionately weighing up potential risks and rewards before making their choices.

…a system that never addresses the why behind a harm never actually contains the harm itself.

Reform is not enough

The author believes that reforming the system will never be enough, and may often be counterproductive.

Previous reforms that may have been well-intentioned with regards to reducing the inhumanity of the prison industrial complex system tend to result in little more than more people being criminalised. A historical example is the introduction of women-only prisons and its relationship towards a far greater number of women being imprisoned.

The only reforms she believes we should support are those that reduce the amount of contact the public has with the police and reduce the resources the police have access to. These are steps on the way to full abolition that immediately reduce the harm police can do to people.

She believes that what we need beyond anything is imagination. Her view is that it’s perfectly fine, even necessary, to criticise the system even if we don’t have a fully fleshed out view of an alternative. Measures can be taken now to address issues that we do have solutions for, and real collective effort should be made to figure out the best way forward on those that we don’t.

In the mean time, our lack of imagination has several adverse effects. We imagine that policing, prisons and surveillance are “natural”. This is not true. Prisons are a human invention; in fact a fairly recent one if we consider the current incarnation whose aim is to punish people who committed serious offences by locking them up for long periods of time. These first appeared in the late 18th century; themselves a reform from the previous punishment regime of capital and corporal punishment.

Our criminal punishment system is also contingent on our culture. We tend to believe that our culture is static, even though it never has been. The fear we’re taught to have for each other, the need to bow to authority and celebrate criminalisation of harm-doers is not the only option.

When asked to imagine a world without prisons we tend to imagine the world exactly as today, with the same level of violence, but without prisons. Instead the aim should be to create a different type of society built on cooperation and mutual aid rather than individualism and self-preservation.

My full notes on this book are here.

Viewers Can Watch Prince Harry Talk to a Trauma Expert for $33.09

The ‘Give me privacy’ tour continues, this time in pay-per-view style.

Although there’s the argument that exposing viewers to topics of mental health and it’s treatment may be net beneficial I suppose.

Euronews.next reports on a tragic case where a man killed himself, his suicidal thoughts allegedly exacerbated by his interaction with an AI chatbot.

In a series of consecutive events, Eliza not only failed to dissuade Pierre from committing suicide but encouraged him to act on his suicidal thoughts to “join” her so they could “live together, as one person, in paradise”.

The bot in question was based on  EleutherAI’s GPT-J, and uses the same kind of technology as chatGPT et al, seemingly without the same guardrails that the more famous ones implement. He was interacting with the AI via the Chai app, which offers a varied selection of AI ‘personalities’ to chat with - marketed at least on the Android store as “chat with AI friends”.

Whilst there’s little doubt that the man in question had a troubled mental state before talking to the bot, and I’m not aware that any formal investigation has quantified any effect that the AI may have had in this case, it does feel like there’s a potentially real issue to consider here.

We tend to accept that suicidal tendencies can sometimes be reduced via chatting, at least to real humans trained to help in these situations from services like the Samaritans. In the US, the 988 suicide and crisis lifeline explicitly offers a chat feature, again, importantly, humans only. And, although I’ve not looked into the research at all, the opposite also appears to be considered to be true in the human world; multiple people have been found criminally responsible for encouraging others to kill themselves.

The app in question of course wasn’t designed to encourage these actions, but I don’t think we’ve any reason to believe that the default behaviour of large language models is to do the opposite either.

📺 Watched Wednesday.

The Addams family daughter is back in this TV series following Wednesday’s teen life as a resident of boarding school full of werewolves, vampires and the like. In between the teen angst and murder mystery, the deadpan goth girl one-liners occasionally excel.

Fun fact: Christina Ricci who played Wednesday in the 1990s Addams family films is back here in the role of “dorm mom” and teacher Marilyn Thornhill.