For the sake of humanity, I’m turning off spellhceck

If artificial intelligence is so smart, how come the internet remains so dumb? For a start, there’s its habit of bombarding you with advertisements for products you have already purchased.

Last week, perhaps, you were in the market for a pram, but having purchased a pram, you no longer need a pram. Why does the internet – now supposedly powered by the brilliance of AI – not realise this simple fact of human reproduction?

Ditto the holiday in Greece, the bird-watching binoculars and the new e-bike: all things that, once purchased, are unlikely to be purchased again, at least for a while.

After spending so long proving I’m not a robot, I’m starting to feel like one. Credit: Getty Images

Instead, the internet imagines that having ordered one e-bike, you’ll instantly want 20 more. For the next 10 weeks you’ll see nothing but advertisements for e-bikes – on Facebook, Instagram and even your online newspaper. Frankly, they’d be better off trying to sell you a first aid kit, perfect for when you fall off the bloody thing.

It’s so annoying, or do I mean “its so annoying”? Actually, I mean the first, but picking the difference between “its” and “it’s” appears beyond the skills of spellcheck. Half the time, spellcheck blue-lines both of them like some sort of soothsayer who sees trouble ahead but is unsure of its nature.

This is often the case with my computer. It offers to be helpful despite being unequal to the task at hand. “Have you considered writing this in a more concise way?” “No, because I like it as it is, now bugger off.” It reminds me of the customer who hovers around the mechanic or the plumber, offering a constant stream of ill-informed advice.

This is typical of the world of tech. Every innovation appears to make the world a little dumber.

Here’s another example: whenever I receive emails at work, the email program gives me a “suggested response” – this to save me the tedium of typing out as many as five words in reply.

It provides this suggested response by scanning the incoming email and noting that it contains a proposed meeting with a nominated time and place.

Hence its suggested response: “Sure, let’s do it! See you then!”

There are a couple of problems with this. For a start, the tone is so perky and American it makes you feel slightly nauseous. Second, the program hasn’t really “read” the incoming email; it’s merely recorded the fact that a meeting has been proposed.

Someone should send an email that says: “Given your secret and duplicitous affair with my wife Samantha, I challenge you to a bloody duel in which we will fight to the death, armed with only machetes, pitchforks and flaming torches. How about Parramatta Park at noon?”

Suggested response: “Sure, let’s do it! See you then!”

This is typical of the world of tech. Every innovation appears to make the world a little dumber.

The latest “improvement” is ordering at a restaurant from your table using the provided QR code. This was a sensible precaution during COVID but has now spread in popularity as a way of reducing staff numbers.

Is the curry mild or hot? Is there coriander in the guacamole because I’m allergic? How many salmon rissoles are in a serve? There’s no one to ask, so you just have to take your chances. You are then asked to nominate the percentage tip you’d like to offer for the service that is yet to occur.

When using such websites you might also need to prove you’re not a robot. This is achieved by identifying bridges, motorbikes or traffic lights amid grainy, tiny photos of Texas. The task is so dull, laborious and mindless that it could only be done without complaint by some sort of compliant robot.

By finally ticking the box “I am not a robot”, you are effectively admitting you are a robot.

The sharemarket still believes AI will solve every problem, transform every workplace, and make billions for the companies that control its secrets.

I have my doubts. My favourite example, cited by the Australian AI specialist Toby Walsh, involves the spelling of the word “banana”. Unless specially instructed, any Large Language Model will be guided by the questions people have asked in the past. So, ask how many ‘b’s are in the word banana, and – unless special instructions have been added – you’ll be told there are two or three ‘b’s in banana.

That’s stupid, of course, but the training data around the spelling of “banana” is full of people querying how many ‘a’s or ‘n’s the word contains. No one has ever asked how many ’b’s are in “banana” as the answer is obvious – well, obvious to everyone except to the sophisticated system called AI.

Maybe the only way out is to turn off spellcheck and swear off AI. I’d go online and buy a print dictionary if not for the way the internet would then spend the next five months bombarding me with advertisements for 10 more print dictionaries.

Better, I think, to visit a brick-and-mortar bookstore and invest in a thesaurus. That way, I could look up the word “internet” and discover it’s a synonym for “annoying”.

Liam Mannix’s Examine newsletter explains and analyses science with a rigorous focus on the evidence. Sign up to get it each week.