Whirligigs & AI
The dangers of AI myths
Humans are naturally cautious of the unknown. It’s a useful instinct that has a tendency to keep us alive. But sometimes we get over cautious.
Take the early days of railways. Back in the early 1800s, medical professionals wrote about the risk of suffocation due to the air rushing past too fast. Or permanent brain damage from the motion. A pamphlet opposing the London to Birmingham line in 1831 claimed “A velocity of fifteen miles an hour is in itself a great source of danger.”
Around the same time someone wrote: “Twenty miles an hour, sir! … It will encourage flightiness of intellect. Veracious people will turn into the most immeasurable liars… Upon the whole, sir, it is a pestilential, topsy-turvy, harum-scarum whirligig.”
Turns out the writer held shares in canal companies. Perhaps a little biased? Nonetheless I admire their choice of words. I really need to use whirligig more often…
But the lesson is new technology can be scary. And fear mongering is not new.
Is AI scary?
Something similar is playing out with AI. It is new and it can seem scary. Just as some Victorians considered the idea of travelling at more than 20mph frightening, some of us consider AI equally terrifying.
There are prophecies of AI replacing all humans. We have paperclip maximizers that are going to turn all matter into paperclips. Or the "AI overlord" scenario where an AI decides humans are inefficient and unpredictable, so it creates a seemingly perfect but controlled society where we're given just enough comfort and distraction to keep us docile. Our own personal Truman show.
AI is amazing. But I’ve written before about the current limitations. Models have a limited context window. They don’t understand time’s arrow. They are unable to understand the physical world. Despite the hype we are a long way from the dystopian futures some are predicting.
Others tell stories of AI consuming vast quantities of water and energy. Even reputable news sources help feed this fear.
Yes, AI uses power. A 500-token query uses about 0.3Wh. Small, but not zero. However, you’ve got to make the correct comparisons. Driving 1 mile in a petrol car takes about 1.5kWh; roughly the same as 5,000 queries. A ten minute shower is 4,500 queries. An hour on TikTok (!) is 2,000 queries.
Look at it another way. How many queries do you make a day? If it’s 100 then a years worth of AI is the same as one car trip of 7.5 miles. Those queries consume ~10kWh per year. That’s about 0.1% of the average energy used to heat one house for one year in the UK.
As ever, heating and transport drive the majority of our energy usage.
So what about water?
Again, yes, AI uses water for cooling. Much of it is recycled, but some is lost. Somewhere between 20-500mL per prompt. That seems a lot. But then consider that a hamburger requires 2500L of water. That’s somewhere between 5,000 and 125,000 queries.
Perhaps we should be focusing on Tiktok-ers who eat hamburgers while driving?
The danger
But the real danger isn't AI consumption - it's the myths that fuel distrust of AI. While we’re debating whether AI will destroy us, we’re missing the benefits it offers right now. It creates an AI scepticism tax. A tax paid by those who hesitate.
For individuals, AI sceptics are essentially choosing to work harder for the same results. They are insisting on digging holes with their hands, well after the invention of the shovel. I've seen colleagues spend hours researching a topic that Claude could summarise in minutes. Or struggle with a technical problem, when AI could offer ideas and suggestions to get them unblocked. This isn't about AI replacing human work, it's about augmenting it. The person who uses AI effectively isn't doing less work; they're doing more, better work. Focusing their energy where it matters most.
For organisations, the cost is steeper. Companies that delay adoption are at a competitive disadvantage which compounds over time. While cautious organisations debate potential risks, their competitors are working out how to use AI effectively - and then implementing it. These early adopters aren't just gaining a temporary advantage. They're building an experience gap that widens every month.
At the national level, the contrast is becoming stark. China has taken a fundamentally different approach to AI adoption. Despite not having the most advanced models (the US still leads there), China is implementing AI broadly - and rapidly - across its economy. Their "use first, regulate second" approach contrasts sharply with the West's "regulate first, use cautiously" philosophy.
Chinese universities are embedding AI across their curriculum - not just in computer science departments. Their small businesses are using AI for everything from customer service to inventory management. Their government is deploying AI in public services at scale. While Western media focuses on surveillance concerns, China is building practical AI literacy across its population.
The result? Despite using models that are 6 months behind the cutting edge, China is on track to gain a significant advantage. What they lack in model sophistication, they're making up for in implementation expertise and cultural acceptance. And China is rapidly making up ground with model sophistication too - it’s entirely possible they will soon lead in both areas.
I'm not suggesting we ignore ethical concerns or sensible regulation. Of course not. But the myths feed scepticism which feeds inaction. Every day we vacillate is another day where our competitors pull ahead. Where other nations pull ahead. Every regulation born from fear rather than understanding adds friction. Slows innovation. Holds us all back.
It's a bit like those canals versus railways. The railways transformed society despite all the dire warnings. Those who embraced them early gained advantages the canal businesses never recovered from.
We're at the same inflection point with AI. The cost of excessive caution might well be greater than the cost of thoughtful adoption. Far greater.
Are we going to be the canal owners publishing pamphlets about the perils of whirligigs? Or are we going to ride the rails into the future?