As part of the series of Uranus in Gemini posts I thought it might be fun to explore the AI tools that are available on my MacBook Air. If I’m going to criticise AI, I should at least give it a try to see what all the fuss is about. Apple Intelligence includes writing tools and image generators, plus integration with ChatGPT and improvements to Siri. It’s integrated into your device which provides greater privacy protections and your data is never stored to the cloud.

Here’s a basic list of what you can do with Apple AI on compatible devices:

• Writing tools
• Summarise notifications and emails
• Manage notifications to improve focus
• Use it to reply to emails
• Transcribe audio recordings and summarise
• Create Genmoji images for Messaging
• Image Playground for creating AI images from photos and descriptions
• Create images from a sketch using Image Wand
• Edit photographs and create memory movies
• Use ChatGPT with Writing Tools and Siri
• Do a bunch of extra stuff with Siri

That all sounds great, but does it work? So far, I’ve only used the Writing tools, email stuff, and Image Playground. The AI is easy to access and works seamlessly with any compatible app. But the software is still in beta so it doesn’t always work as advertised. It can be temperamental and a bit shit at times. Sometimes it’s very shit.

In Mail you can summarise the email previews instead of showing the first lines, but I switched this off because it wasn’t helpful. The ability to summarise long emails is more useful and I’ve used it with emails I couldn’t immediately understand because they were poorly written. The AI was able to pull out the salient facts and give me a sense of what the person was trying to say.

Image Playground is silly and can be fun to play around with but the results are mixed and it often creates mad and impossible things which is entertaining. It provides suggestions for creating images and you can add text descriptions, upload or take photographs or use one of their person animations. The images are rendered in an animation style or you can choose the illustration style which is less cartoonish.

You can see some examples of basic prompts in the header image above. I’ve also used it to create images of myself from photographs. The one on the About page is in the illustration style and the eyes are a bit wonky (although maybe that’s me!). When you render this image in the animation style it makes me look absurdly young. Here are the images side by side – I changed the location for the animation:

The source photo for these images is 10 years old so I tried to make ‘myself’ look older by taking a current photo, hoping that the streaks of grey and white hair would produce the right result. The hair turned out fine, along with the much-depleted eyebrows, but it’s still too young, and the animated version is ridiculous:

There weren’t any better versions of this animation in the selection that came up – they all looked pre-pubescent! And none of the images really look like me – some are closer than others. And some turn out bizarre. Here are some examples of anomalous results, including random bits of writing, malfunctioning sunglasses, a useless notebook, and what happens when you select the ‘Hiker’ prompt without also selecting a person:

So much for Image Playground. What about the Writing Tools?

The Writing Tools include grammar and proofreading, rewriting, summarising, making the text more friendly, professional, or concise, creating a key points summary, making a list or a table, and composing using ChatGPT. I’ve used it in Word and it works quite well, although I’ve noticed grammar errors in some of the rewrites.

The ‘make friendly’ or ‘professional’ options just seem to use a thesaurus to change a few words, which I can do myself. The professional option mostly adds extra verbiage, probably in an attempt to appear smarter. Here’s an example from a paragraph in the Uranus Education post (click to read):

Sometimes it comes up with the message: “Writing Tools aren’t designed to work with this type of content.”But if you click ‘Continue’, it does the rewrite anyway. The first time this happened I thought it was due to the length of the article I was working on. Then I discovered something that borders on censorship.

Apparently, this warning comes up when it comes across ‘untrained words’ or anything that might be controversial or upsetting, such as rude words or references to drugs or illegal activity. When I tried to summarise the Uranus Education post the warning came up but I couldn’t see what the problem was. There were no naughty words, although I suspected the word ‘psyops’ might be triggering it.

So like a good little Virgo I systematically went through the article to discover exactly what was causing the AI to have a panic attack. The post is absolutely littered with ‘problematic’ phrases and words. According to Apple you can’t suggest that people may be stupid or getting dumber, or are lazy or cheating, or that they might be destroying themselves. Here are some of the phrases that triggered the schoolmarm:

• clearly ignored
• completely failed
• teachers have failed to do their jobs
• functionally illiterate
• destroying your own mind
• hamstring themselves
• lose our humanity
• variety of psyops

It didn’t like mention of the totalitarian state but was fine with the zombie apocalypse even though these are kind of the same thing. I managed to get it to rewrite ‘clearly ignored’ and ‘completely failed’ by taking out the adverbs. ‘Ignored’ became:

“it seems like the message was overlooked”

And ‘failed’ became:

“I’m feeling a bit discouraged right now”

So you should, you silly AI!

This is just rewriting a phrase out of context so it doesn’t work in the article but it does illustrate the problem. The next phrase makes it even clearer. I got it to rewrite the phrase ‘teachers have failed to do their jobs’ by changing it to ‘teachers may have failed to do their jobs’ which didn’t flag the warning. For the friendly rewrite it produced:

“Teachers may have stumbled, but they’re still learning and growing.”

The professional rewrite of that phrase was simpler (for a change):

“Teachers may have made mistakes”

It seems to want you to soften what’s being said, make it less frank, less direct, as if it doesn’t want to be too critical in case it hurts somebody’s feelings. It ends up pulling its punches and being evasive. But it’s not really censorship. It will happily rewrite the ‘problematic’ words and use them in its rewrite. It didn’t like the phrase ‘functionally illiterate’ but rewrote that paragraph and used that phrase (click to read):

The rewrite doesn’t improve the selected text because there was nothing wrong with it to begin with. It just swops out a few words. I also got it to rewrite just the word ‘illiterate’ and it finally admitted the truth!

If you’ve got a big block of text, it can be helpful to summarise the main points to give an overview of what’s being said. Summarising text is difficult because you have to make value judgements about what to include and what to leave out. You need to compress what’s there without destroying the meaning. Obviously, AI is going to struggle with this because it doesn’t understand the meaning of the text.

I’ve tested the summarise function on a few things and the results are usually pretty good. The following is the AI summary of the Uranus Education post which is 2300-ish words long, reduced to 141 words:

---

“Artificial intelligence (AI) lacks true intelligence as it lacks understanding and consciousness. This misunderstanding of AI’s capabilities reflects a societal decline in critical thinking and reasoning skills, evident in declining literacy and numeracy rates. This decline is attributed to passive internet use and a failing education system, impacting students’ ability to think clearly and comprehend complex information.

“A study on Dickens readers found that many lacked comprehension despite receiving BAs in English. This highlights the disparity between basic literacy and deeper understanding, potentially exacerbated by distractions and a lack of critical thinking skills. The author warns that over-reliance on AI tools like ChatGPT, while potentially beneficial for learning, could lead to a decline in critical thinking and intellectual growth if used as a crutch rather than a supplement.

“The real danger is not AI sentience, but human stupidity leading to self-destruction.”

---

This isn’t bad considering the trouble it had with all those problematic phrases. I also tried to summarise the full post using the ‘make concise’ option but it failed and crashed. It had obviously had too much for one day!

Another option is to use key points to summarise the text. This usually provides a good breakdown of the main points mentioned but organises the information poorly. There are often too many bullets and many could be lumped together into the same point. It clearly doesn’t understand what it’s doing.

The summarise tool might help you to save time if you need to provide a shorter version of something you’ve written. But it’s not that difficult to do yourself, and it is good practice to think about what you need to include – what’s relevant in order to make your point. Doing that thinking helps you to understand your ideas better and should help to improve your writing.

My main issue with the rewriting tools is the way the AI changes my intent. It can alter the meaning of what I’ve written by taking out the correct word that I chose for a reason and swopping it for something less precise. In other words, this is a tool for non-writers or those who struggle to organise their thoughts or construct grammatical sentences.

However, I did try it on a bit of rough text consisting of notes I’d thrown together that included some full sentences plus half thoughts and random phrases. It kept most of what I already had and just completed what was unfinished, but it wasn’t what I would have chosen to write. It wasn’t in my voice.

This is one of the dangers of using AI to polish your writing or make it more concise. It might seem like a good idea but you could lose your style and end up sounding like everyone else – like a machine with no soul.

At the moment it’s usually possible to tell if something has been produced by a machine. Maybe that will change as the AI improves (if it improves). But you’re still stuck with the essential problem. Eventually it will become obvious that you can’t construct a coherent sentence, can’t spell, don’t understand grammar, and don’t know what you’re talking about. And the longer you rely on AI, the worse the situation will become.

On his excellent blog Chris Rollyson says that writing produced by AI is inconsiderate because it has been stripped of its human context and devalues the relationship between the writer and the reader. He calls AI-generated writing the equivalent of “frankenfood” that doesn’t nourish because it contains no real insight or passion.

With that in mind, I won’t be using these Writing Tools on a regular basis, despite them being so handy and easy to use. The results simply aren’t good enough and I don’t need them – and I’d like to keep it that way by hanging on to what’s left of my ability to think for myself. I won’t be joining the robot party any time soon but if the cyborgs catch up with me, this is what I’ll look like:

Weirdly, this is actually a better likeness for me because I look suitably grumpy!

More posts on Uranus in Gemini

Thanks for reading! If you enjoy this blog and would like to support my work, please donate below 🍵. Thanks in advance! 🙏❤️