Here are the explains I promised.
Where to start?!
AI is not a new species and is not capable of becoming one. AI is a broad term encompassing many computer tools that should make life easier for us. There will be considerable shift in employment patterns which will be distressing and will disruptive, but as in all revolutions, it will eventually lead to more but different jobs being created. For example, during the Industrial Revolution many men were put out of work because women and children were more agile and able to work in the factories and therefore cheaper. Men were reduced to "women's work" of mending stockings (I think Hobbes wrote about this), but their children and grandchildren prospered in jobs that hadn't been invented at that point.
Similarly, jobs that include checking things against a standard will go. For example, straightforward insurance claims and diagnosing illness from scans and x-rays. However, there will always be a place for human input. As Errol said, AI is really brilliant at saying that white spot on an x-ray is cancer compared to the human eye, because the AI has compared the x-ray to millions of other x-rays that had a white spot in that place that was cancer.
A human will be the person to break the news to the patient, we will still need personal care and physical presence for many jobs. Think of hairdressers, plumbers, bridge-builders. What's not so clear is a role such as teaching. Will we still need people in front of classrooms? Personally I think teaching will become more about nurturing.
Anyway, for a lot of office jobs, there is worrypanic that basic jobs will be lost, but take marketing for example. Sure, AI can write you a press release, which for a creative person can be the hardest bit - coming up with original content, because most marketing people are sick of finding new ways to say "our product is brilliant" and are better at rewriting "our product is brilliant" to say "our product is wonderful". But AI can write the initial text, but someone will have to fact check it while AI is still inaccurate and someone else will have to edit the text so it sounds human and not AI, and someone will have to write the prompts. I think these roles could be three different people, because they are different skillsets. So a single marketing manager could be replaced by three roles.
Essentially AI for rewriting is about predictive text. It takes the entire internet of all the sentences, mostly in English, not always accurate, mostly input by men without a female perspective or language. (As an aside, I love MN because the perspectives are mostly female. MN can be a complete bitchfest, but it doesn't pile on posters in a way that most social media does. It is supportive, it writes about pregnancy and children and parenting and relationships which are rarely seen elsewhere, and the responses are generally empathetic.) It then assumes from this what the next word will be. So you type in "The cat sat on" and mostly it will return "the mat" but it will also return "the table", "a mat", "a table", "a chair" etc.
This is why you can (and IMO should) ask for three versions when asking for output.
AI makes mistakes because it isn't sentient. It can't distinguish bad data from good except in some circumstances. It is trained to respond in the way that its creators and editors want. So, it is really good at anagrams for example, because it writes out every single combination of letters and gets rewarded when those combinations make real words. This was done electronically by matching words to the dictionary and also manually by thousands of people paid pennies an hour.
As a result you can watch Countdown and get the conundrum in 3 seconds just by inputting the letters and it will spit out the answer very quickly. Or the longest word available. But where's the fun in that?
It will get better, but right now it can't even reproduce the lyrics from I'm Not In Love in an image, or work out what a blackbird looks like.
As for AI being checked word by word, in some cases, yes that is happening. I am part of a project that is paid to factcheck AI responses and we are not allowed to use AI to check our responses. What's interesting is that the front page of the internet is now returning fake responses because so much AI has returned to the internet. This makes the job harder, and better paid. I think it's important to have women involved because we think differently and value different things. We are not so hung up on status or being "right" if we can avoid conflict.
Part of the project is to write AI prompts that force it to return false information, which is different to checking responses, and to ensure that its responses on questions about science, engineering and history are correct.
Those who think that AI will "take over" are generally men, because that's what they would do (see Elon Musk's involvement in global politics), or they don't understand what it is. AI is just machines who have learned how to predict things - that's why it's good at chess and noughts and crosses. The real issue is not AI taking over, but men using AI to take over, whether by fake news and manipulation of social and mainstream media.
The one thing that always wins is hope and I include love in that. AI can't hope and can't love, it can just return information that it has been told is "correct" because of the way it was trained to return information. This carries dangers because it could just keep returning information that runs on a loop that can't be broken.
However, AI is very energy intensive. Microsoft are literally (and I'm a pedant) looking at operating a nuclear power station at Three Mile Island to power their servers. The cloud has already superseded aviation in terms of emissions (that's not hard, aviation is only 3% globally, so I have no idea why people are so puritanical about not flying, when heating, cow burps and road traffic contribute way more!). So what might bring it down or restrict it in the end could be that it is so energy hungry that we can't produce enough power for it and live our lives.
I hope the explains help. Happy to answer questions, though I'm busy till later. Sorry if there are any typos.