Of Punch Cards and AI
Many decades ago, when computing was still in its infancy punch cards were used for looms, to design patterns in English mills. They were then modified to store census data and to help measure the US population. In the process they saved the US government weeks of work and five million dollars at the time. This is according to my memory of reading The Big Switch and coming across this information two or three days ago.
The idea of punch cards was that you design cards with holes. Each hole gives a true or false value. Instead of writing “Billy Bob lives in a house with five children, two cats, three geese with an income of 500 USD per month you could punch the relevant hole on a tabulating machine. “Two contestants required 44.5 hours and 55.5 hours. Hollerith astounded Census Bureau officials by completing the task in just 5.5 hours!”
Fast forward to last year or the year before and you see that the world become super-excited about AI and how it could process big data, but also how it could intuite what we wanted to write by hoovering petabytes of data. Plenty of people are saying “This isn’t AI, this is machine learning”.
I am currently reading I, Robot by Isaac Asimov and we see that even today the same problems are occuring. (spoiler alert coming up). They fed info to one supercomputer and it broke, so another company fed the same information to another system but told it “if you find that this would kill people ignore it and return the document”. To this day AI is being taught this. “I am only an AI and I cannot answer this question” for Gemini and “I am not able to answer any questions related to events after September 2021 by ChatGPT.
If The Big Switch was written today it would be extended from “Rewiring the world from Edison to Google” to “Rewiring the World From Edison to OpenAI”. The book was published in 2008 and I’m reading it in 2024 so it will stop way before current information.
I bought the book over a decade ago, but didn’t get around to reading it until now. We are told that AI and other technology is revolutionary. Books such as the Big Switch show that in the grand scheme of things AI is evolutionary and itterative, rather than a brand new paradigm. Punch cards were as revolutionary in their day, as AI is today.
If we want to jump forward a few decades is another good example of “evolutionary progress”. We see that the technology that Google uses to help us look for flights is decades old and that Google is just a user interface built upon old code and interactions. The video I link to demonstrates how old tech needs to be modernised to help cope with crewing requirements. In the video they speak about how when everything is working it’s easy to manage. It’s once flights are cancelled and flight crews are scattered in the wrong place that chaos ensues.
During a big storm you need one system to track flights, but you need a second one to track aircrafts, and a third to track crews. You then need that system to route new routes until everything is brought back to the normal schedules. With time you could combine these three requirements to provide a flexible and optimal flight re-booking system to minimise bad weather disruptions.
And Finally
It’s easy to call something revolutionary, if you don’t look far enough back in time. If we look only at our own lifetime then plenty of things seem revolutionary because the change is new to our generation but if you expand beyond our own lives, and look back a century or two you will find that almost everything is itterative. If anything is “revolutionary” it’s that we now have enough processing power for AI to make reading “punch cards” that are petabytes of data faster than in the past.
Just search for a file on a mac with easyfind or finder and you’ll see how slow searching on a local machine is, compared the the AI optimised systems that people are playing with today. LLM runs slow on Pi and 2016 mac book pro but it runs fast on newer machines. That’s where the new opportunities lie. Machine learning, masquerading as AI is enabled by technological advancements. Just look at the rivers, lakes, and gigawatts of power that are required for “AI” to be of any use.
