There has been so much buzz recently about AI and the revolution which it is supposedly ushering in. This craze has now been so many years in the making that it’s probably appropriate to start talking about different ages of AI, even within the past decade. So is now the time to start worrying about it? Is AI really this terrifying, mystifying force that’s about to sweep your job out from under your feet?
Firstly, an aside that the graphics used in these kinds of articles are one of the parts of this “revolution” that have always tickled me. Whenever I see an article from a financial institution or another corporation about AI, there always seems to be a robot looking at a human or a robot arm touching fingers with a human. This bears no resemblance to reality, where even if you’re working in a factory, on an assembly line, or in a packaging facility, you’re extremely unlikely to come across such a thing. The reason why these pictures are used is because it’s actually relatively difficult to summarise what AI is doing in a nutshell, especially in a picture.
If we were to sum up in one sentence what ALL of AI does, we could say: “AI helps to estimate approximate solutions to problems which are otherwise difficult (or otherwise impossible) to solve.”. This may not be the most accurate description but it should give some solace to the worried reader. Instead of some organic, self-sustaining, growing sci-fi monster, or a terminator-esque soulless robot, AI is generally giving estimates of solutions to problems which otherwise remain poorly solved, or unsolved. This has moved frontiers for sure, problems which used to have complicated and sophisticated solutions are now solved eloquently. For example:
LLMs can translate between languages accurately and quickly
estimates about which video or tv show you may or may not like to watch can be calculated from what you’ve viewed previously
neural networks can help classify objects or even recognise people
you can transcribe whole meetings and provide a summaries on the fly
Notice that AI is often being used in places where other models were being used previously, or in creating new insights into things that weren’t possible before. There are a few places in which processes which rely on AI have replaced what was being done previously by a human, but often these were the manual and menial parts of a job.
For instance, how many of the meetings you’ve participated in have someone taking notes? Of those, how many people actually enjoy taking the notes, summarising and sending them out later? Using this example, I’ve noticed how online meeting services like Co-Pilot in Teams or Gemini in Gmeet take all the worry away. It’s been eye opening to me how much of a meeting I thought I’d recollected vs what the AI generated transcript said, and what the actions it took were.
If I look closely, however, I often catch places where the context was misunderstood and the result was something nonsensical, or where several distinct areas were blended together to form an awkwardly impossible construction. Beyond this, I’d choose an AI note-taker over a human 100 times out of 100 (especially if that human was me!). The proviso is that I need to review the notes before sending them out. It follows that AI solutions which firms choose to use will likely decide how to utilise: “Human in the loop”
Human in the loop is exactly as it sounds; whilst a system or solution powered by AI can help make decisions, it can’t act upon those decisions and needs a human to execute them. This would appear to be the best of both worlds: the intelligence that an AI system can bring and the insight of a human whether to act upon it or not. You could start to imagine your job where instead of bashing your head against a brick wall of mundane tasks, you have greater oversight of the job you are doing, and are providing the piece of oversight or insight which makes the different systems powered by AI effective.
What’s been replaced so far? Ask AI what jobs it has actually replaced so far (try it!) and you’ll get a laundry list from cashier to physician. But dig deeper and most of it doesn’t hold up. Let me show you two examples I know well which were suggested when I used AI to give me a list:
“Stock trader” - Whilst there has been a shift in jobs on the trading floor away from manual trading and towards automated trading, this shift wasn’t brought about by AI! There used to be a large number of jobs involving manual trading which reduced by about 90% with the advent of algorithmic trading. The algorithms were created in the early to mid 2000’s, and generally revolved around models created by quants, using the available financial formulas at the time. Whilst there have been improvements made by using AI, the entire job shift happened over more than a decade previous. It also created a large number of jobs that were dedicated towards running and improving algorithmic trading. Further, the traders that adapted and remained spend less time constantly trading tiny segments of a client order, and spend more time understanding market movements and news in the stocks they cover.
“Computer programmer” - With the advent of LLMs there has certainly been a step change in the ability to create code from very little. The term vibe coding (using LLMs to create your code for you) exists to describe this exact phenomenon. There are scare stories suggesting that computer programmers will disappear completely as LLMs will write your entire code base. To those that think this possible, I would suggest trying it now! Once you move beyond straightforward scripts that you can run locally, and start to move towards larger, complex systems, the idea that this can happen is laughable. LLMs can help an experienced programmer write code more quickly for sure, but that’s exactly what’s needed: someone who actually knows what they’re doing. A complex system still needs an overall design and purpose, a clear setup and a proper environment. A LLM needs to be cajoled into writing precisely what’s needed every step of the way. As always seems to be the case, a lot of dross is far less useful than a small amount of sensible setup and code.
The list suggested to me showed it’s patently obvious that this “AI replacement” simply hasn’t happened. A sweeping statement might say that some jobs have had many of their menial tasks replaced by programs, which may or may not include AI solutions.
So what next? Now is the ideal time to start putting yourself in that seat where you have AI working for you, rather than instead of you. The more you understand where AI can and cannot help, the easier it will be to start changing the daily tasks you do to become more efficient.
Just as a thought, here are some of the things that I’ve happily handed over to an AI assistant:
Used meeting transcription features
Used LLMs to summarise the meeting and action points
Used LLMs for insight (e.g. to see whether I’d interrupted or was considered rude)
Crafted prompts to tidy up content
The point isn’t whether AI will take your job, but which parts of your job you’d gladly give away. So what’s the first task you would delegate?
