AI is a bit crap
In the past 12 months or so, I’ve been trialing and attempting to use AI Tools at work (mostly ChatGPT, but also in-built AI addons to software & tools like Salesforce, LinkedIn and WhatsApp/Facebook Messenger).
Although initially I was excited at the possibilities from using these tools, I’ve concluded that as it stands, AI is a nice summarization and word/spreadsheet editing tool, but that its capabilities and potential for the near future are massively overhyped.
More bluntly, at the moment it’s a bit crap. I imagine it’ll remain a bit crap for the near-ish future given the first public ChatGPT model was released over two years ago.
‘You should’ve done more Prompt Engineering, mate’
A lot of defenders claim it’s the problem of the user, the cretin who hasn’t put in hours of ‘prompt engineering’ training. I’m not so sure. Last summer I put my money where my mouth was and subscribed for the premium edition, ChatGPT 4 or Plus or whatever grandiose title they’ve given it now.
As part of my work, I wanted to research some of the best Business Sustainability conferences to attend in Canada in 2025. I was very clear in my prompt, asking my friendly robot for a list of 8–10 sustainability-focused business conferences taking place in Canada next year, with details of their location and exact dates.
Seconds later, a list materializes before my eyes. Wunderbar! The list included ideal sounding conferences such as the ‘Sustainable Business Summit Canada 2025’, taking place in Toronto in June, focusing on Sustainable Investment & Corporate Sustainability.
Exciting! I threw open a Google Chrome tab to learn more about this gift from the Gods. Except, it didn’t exist. There was no ‘Sustainable Business Summit Canada 2025’. Nor for that matter was there the ‘Green Building Festival 2025’ in September. Hold on Santa, you’re tell me the kids won’t be able to go to ‘Canada’s CleanTech Leadership Summit 2025’?
The above is just one example where I frequently have to double, triple and quadruple check whatever comes from the mouth of the beast. When I ask it to do more simple tasks, such as removing duplicate entries from an Excel file or a Word doc, more often than not it just spits out an empty document or the same document that I uploaded. When I tell it that’s not what I asked for and to take another look, it apologizes and then just does the exact same thing again.
AI Garbage
These examples are when I’ve actively taken the initiative to use AI. More often though, in the course of our daily lives we’re just passively coming across its mediocrity and limitations. A recent Wired article estimates over 50% of English-language content on LinkedIn is AI-generated
Anyone who’s spent more than a minute on the corporate cesspool of the Internet will be familiar with this kind of post. A punchy opener, some corporate ‘problems’ and a few generic sounding proposed solutions from a self-styled Guru — ‘companies need builders, sellers, and enablers who can execute’. Execute? Who are we executing?!
Presumably, the ghost in the machine has crunched this WSJ article (available on MSN here) and come up with this LinkedIn Post. The statistics are confusing and most don’t seem to accurately reflect the content of the article. ‘McKinsey cutting MBA hires by over 50% at top schools’ is taken from one example from Chicago Booth School of Business.
LinkedIn and other social platforms are swamped with this kind of badly written nonsense. While nobody is comparing ChatGPT to James Joyce, even its abilities to summarize and draw conclusions from articles are just frequently really bad.
Why can a six-year old draw better than AI?
When you ask DALL-E (the ChatGPT for generating images) to produce a picture from a description, the results vary wildly. Last year, I asked DALL-E to create an image of how technology can help humanity achieve the Sustainable Development Goals for a conference my organization was presenting at. I prompted it to produce an idealized image of drone technology giving a farmer insight into soil conditions and more data on his crops. The picture produced a nightmare picture of a swarm of drones circling over a lone black farmer bent over in a field. I went back to the drawing board.
I initially thought DALL-E had done a pretty good job with this image, when I asked it to produce an image representing how AIpowered recruiting tools could fall victim to bias built-in from their production and training data. In this case, I just needed DALL-E to fix the massive spelling error in the top right of the computer screen. I went back and forward over five times asking it to change the ‘U’ in ‘BIUS’ to an ‘A’. Each time it apologized and spat out the same image, noting helpfully ‘here’s the image you asked for’. Eventually, I just gave up and used MSPaint to put a big blue square over the word.
There’s tons of examples like this, where you get your hopes up with something that seems useable but you get so frustrated trying to fix it and work with the machine that you just give up and find an image on Shutterstock.
AI — The Moonshot Technology
In the last few years, I’ve even seen lines in articles that claim AI could even help solve the climate-crisis. We’ve seen this before, when people started to get excited about carboncapture technology. As David Wallis-Wells states in his brilliant summation of the climate crisis, ‘The Uninhabitable Earth’, to stay below 2* warming, we need to open new full-scale carbon capture plants at the pace of one and a half per day every day for the next seventy years. In 2018, the world had eighteen of them, total.
AI as a solution to the climate crisis is a classic example of a Deus ex machina, where a seemingly unsolvable problem in a story is suddenly solved by something coming out of nowhere to totally solve the issue. Will there even be enough computing power in the world to get AI to a point where it can at least fix the spelling error in my picture? Some experts think Moore’s Law is breaking down, the trend which saw computing power double, and its cost halve, every couple of years (why your iPad is far smaller and can do far more than your old family desktop PC). So can we even get to the stage where AI can start really making a difference, or am I doomed to travel to fictional conferences in Saskatchewan my whole life?’
I always think of self-driving cars as a classic illustration of why AI isn’t coming to save us any time soon. Weren’t autonomous cars meant to be chauffeuring us around to our every appointment and social engagement by now? Well, where are they?
Probably where they were 10 years ago, on a couple of streets in San Francisco.
