This article is about AI and interviews were AI-transcribed but it’s human-written...
In April, one of egg’s biggest ever conversations kicked off about AI. Lots of people have very strong views. Some pointed out how it could change their lives, helping them overcome dyslexia or streamline running a small business alone. Others were worried about job losses, AI replacing creativity, and the ethics of using AI when it takes up so much energy and water.
It feels like about ten minutes ago I was sitting in an overheated high school classroom being told that we weren’t allowed to use CD-ROM-based encyclopaedias to do our homework. The teacher would know, and we weren’t going to have an encyclopaedia in our pocket in the real world, were we?
And here we are. Ten minutes later. Half my feed is telling me AI is going to revolutionise my work, world and life; the other half is telling me AI will destroy everything. People have very strong opinions on this. So, in the first of an egg Investigates series, here’s a brief introduction to AI. Let us know which topics resonate most strongly with you, and we can explore them in more depth in future articles
Can AI work for you?
I’ll admit I’m a slow adopter of AI. By the time I carefully craft a prompt for AI to write my email, I could have written three myself.
Good egg Julie Begbie is founder of the Happy Business Company, and it was one of her business posts that started the debate. I asked her whether AI could realistically help me. She told me the trick is to keep an honest log of where you spend your time, and then work out which tasks drain your energy.
“Because the humanity is what will make the difference. As AI comes out, people have more and more of a craving for human connection, properly human interaction. But if we can speed up other things, there’s more time for that.”
So, copywriters should write, but AI could filter their emails.
Another important aspect is human responsibility. Critics would point out that humans also make errors of judgement, but the difference is that we can be held responsible for our mistakes. If AI messes up and no human stops it, there’s no accountability, and it can be very hard to work out how to prevent things going wrong again.
That makes sense to me. When I had to handwrite notes from interviews, I was looking at my notebook. And transcribing recordings took hours. Using AI means I can focus on really listening, watching body language and picking up cues. Not transcribing means I can spend more time researching or looking for other perspectives.
But AI can’t do it alone; it’s the work-experience kid, not the editor. It still makes too many mistakes, and sometimes it misses nuance, bad questions or sarcasm. If I was making decisions based on that alone, I’d be making bad decisions.
Will AI destroy work?
Will AI steal our jobs, and if it does, will it be any good at them? LinkedIn is full of people complaining about “AI slop”, poor-quality work churned out by AI systems based on prompts, often with an allegedly distinctive style - short, choppy sentences, the “it’s not this, it’s that” format and overuse of em dashes, the long hyphens you can see, er, well, all over my work.
There are also concerns that some platforms still “hallucinate”. Programmes like ChatGPT are not designed to look for correct answers; they’re designed to look for high probabilities. And they want to please. Those two factors together mean they can pick up fake stories. It’s now taken for granted that the entirely planted story that Arthur Conan Doyle drank in The Salisbury Arms is true. They can also make up sources of information unless carefully told not to.
But these flaws are reducing over time, and some packages are designed to get around them. NotebookLM will only use the specific sources you ask it to look at. Perplexity names its sources and Consensus.ai is designed to look only at certain types of evidence, often formal reports and scientific papers.
Everyone I spoke to stressed that AI lacks judgement. It can do simple, repetitive tasks very well, but humans still need to take responsibility for decisions. Julie points out that when Excel first became common in homes and offices, people assumed no one would need an accountant anymore. What happened in real life was that accountants became more analytical and their jobs more interesting, but they still had to oversee the computers to make sure mistakes were caught.
The King’s Trust (formerly the Prince’s Trust) reports that around 60% of jobs will change due to AI, with 10% likely to be replaced. That doesn’t sound too bad, except that the jobs most likely to disappear are disproportionately the ones young people and career changers can take: routine jobs where you learn the critical thinking needed to catch machine errors. That’s still a chasm we’ll have to cross.
This all supports the cliché that you won’t be replaced by AI, but there is a real risk of being replaced by someone who uses it.
Is AI watching you?
I spoke to a LinkedIn connection who is an expert in cyber security, Magda de Jager, who is also a flamenco dancer and understands the concerns artists have about the misuse of AI. She repeats the adage that if you are not paying for a platform, what that platform is actually selling is you.
Anything on the internet can be assumed to be semi-public; anything that touches the internet can be hacked. And almost none of us have our AI systems “airlocked”, kept offline and working locally on our computers. Even when you tell a programme not to use your information for training, both prompt and answer are still held in logs.
But is that a problem? Magda says it depends on your risk tolerance and what you’re using AI for. For example, she says she wouldn’t recommend updating your CV on your work Co-pilot. The odds of a bored IT clerk trawling those logs is small, but the consequences big enough, and the alternative of doing it at home easy enough, where it’s not a risk you need to take. It is a bit like keeping a paper notebook with your passwords written on it; your Netflix password in a notebook in your bookshelf is low risk, but you wouldn’t carry the passwords for all your bank accounts in the same handbag as your phone and cards.
AI and education
Another area that gets a lot of attention is AI’s role in education. We’re used to hearing that in the negative. Students using AI to avoid writing their own essays, and learning nothing in the process or academics at risk of losing their job. I’ve seen the analogy that using AI to write your essays is like taking a forklift truck to the gym - you’ll lift loads, but you won’t get stronger.
Laurie Blair is another egg, who uses AI to support lecturers designing university modules and says that’s misrepresenting how you could use AI to learn. Students can use their own brains to do the learning, but AI to generate study materials, infographics and identify sources.
Academics who have embraced AI are using it to structure their thinking more clearly and produce early drafts of teaching materials. Doing that frees up more time, not less, for editing, improving, and finding more up-to-date references for classes. It can also help create better materials for people with different learning styles.
AI can also help iron out issues like students struggling to understand strong accents, not feeling confident speaking in large group settings, or needing subtitles and image descriptions.
Mind you, my note-taker can’t handle spoken Scots, so there are still limitations to the technology.
The environmental costs
AI is literally hungry, for water and electricity. There are figures suggesting that the Google data centre outside The Dalles, Oregon, used 25% of the city’s water as early as 2023. The impact of that could be terrifying. Many AI providers are working on efficiency, as much from cost containment as being environmentally friendly. New chips are being developed, and data centres are looking at switching to renewables. But the problem is that however efficient they are, more and more people are using them, and the scale is increasing faster than any current work to contain the electricity usage.
But we aren’t starting from zero. Our searches use water and electricity too, and so does every email and photo you didn’t delete.
It doesn’t look as though the “glass of water for every ChatGPT search is true. The amount of water varies, with complex searches and images more costly than simple ones. But a ChatGPT search tends to use much more water than a standard google search; standard google searches use a third of a ml of water, and ChatGPT between 3ml and 40ml. In terms of power, AI uses 10-15 times more electricity than an old fashioned search.
Those figures are for simple queries, though. If one search on an AI system replaces one Google, you are using more water and power. But if that prompt saves you an hour of running your laptop and 20 or 30 searches, you’re saving power. Buses use more power than cars, but you can fit a lot more people in them, so they’re still greener.
AI cynics will be thinking “yes, but it doesn’t need 20 searches to find out when the school inset days are”, and they’re right. But perhaps it’s a matter of being more thoughtful. Maybe I can save the city term dates somewhere safe and not have to look them up every time, and save the LLM searches for when I really do need to do something more complex. Planning prompts well, using specific platforms that do the task the best and using search engines like DuckDuckGo which allow you to switch off AI summaries when they aren’t needed will all reduce the waste.
So… should I use AI?
In the sort of answer that both sides of the debate will hate me for, I still don’t think there’s a yes or no answer. But for me, I’m filing it with driving. If I only need to get to the co-op for milk, I’m fit and healthy and should walk. If I have to get to the Fort to buy something heavy, I can maybe justify taking the car. So, yes to AI note takers, and yes to complex searches. But no to letting it replace my judgement or using it for every little thing.
In fact, I’ve made myself some journalling prompts – just for me, with my brain, on paper:
- What risks am I already exposing myself to? Am I getting enough good back for those risks to be worthwhile?
- What can AI do for me that frees me up to be more human elsewhere?
- Does this task need human judgement?
- Are there better, more specific tools for these uses than a general AI?
After all, my teachers’ insistence on avoiding CD-ROMs wasn’t really about the technology. It was about the need to keep thinking. And that’s not a skill I can see us losing the need for.
