The most famous building in Pisa in Italy is the Leaning Tower. The lean became obvious soon after construction began in 1173 and the builders tried to adjust for the tilt even as they built higher. Obviously that didn’t work.
Not nearly as famous as the Leaning Tower but barely any distance away, is Pisa’s cathedral. I doubt that one tourist in a hundred gives it much thought and yet it, too, has a claim to fame. It was there that Galileo Galilei discovered the secret of pendular motion.
According to legend, he was in the cathedral and he was bored. It must have been a hot day because the windows were open. He watched as one of the chandeliers was blown about by the wind.
For whatever reason, he decided to time how long it took for the chandelier to return to the vertical after each wind gust. He did this by counting with his pulse. And he discovered something odd. It didn’t matter if the chandelier was moved only a little bit, or a lot. When it swung back towards the vertical it always took the same length of time to do so.
Galileo was fascinated enough by this phenomenon that he repeated the experiment when he got home. And he got the same result. The angle through which a pendulum swings is irrelevant. The time taken is dependent only on the pendulum’s length.
According to Wikipedia, the Dutch mathematician and scientist Christiaan Huygens followed up these observations a century later. He went further than Galileo and realised that the back-and-forth of a mechanically impelled pendulum could be used as a timing device. From this insight he invented the pendulum clock.
Pendulum clocks aren’t used much anymore. They’ve been superseded in the quest for greater accuracy. To understand why, I’m going to tell my Piccadilly Circus story.
Like most people, I’ve done different jobs in my working life. One day quite a few years ago I found myself monitoring escalator number three at Piccadilly Circus Station.
I was there because one of the other escalators in the station was out of order and it was important that number three remain in service. My task was to stand near the bottom of the escalator and watch it. If it stopped, I was to get it restarted as soon as possible.
This was exactly as boring as it sounds but sometimes in life you do boring jobs. I might add that nothing went wrong with the escalator that day.
Still, as I stood there I eventually noticed that someone had stuck an Arsenal FC sticker to one of the steps. I would see it every now and then as the escalator completed another revolution.
It occurred to me that I could work out the duration of a full rotation by using the sticker as a marker. When I did this, from one sighting of the sticker to the next took something like 112.5 seconds.
But of course, my timing might have been inaccurate. I might have fractionally delayed either starting or stopping the stopwatch. I reckoned that my timing could have been off by as much as a second.
But what if I counted two rotations? Or if I let the escalator perform fully 10 rotations?
If I counted 10 rotations, my overall timing could still be wrong by one second. But I could divide this time by 10 and if the escalator had been moving at a consistent speed, my timing for a single rotation would now be accurate to 1/10 of a second. If I let it go for 100 rotations, my timing would be accurate to 1/100 of a second.
And you see the pattern? Any process that repeats with an absolutely regular pattern can be used for timekeeping. And the more repetitions, the better.
Clock-makers were well aware of this and you might think they would have preferred shorter to longer pendulums. After all, they move faster. And clockmakers would, but they had to take other factors into account. Fast-moving pendulums wear out faster.
Pendulum clocks eventually became obsolete from the 1930s onwards. Quartz crystal oscillators were developed that were capable of thousands of accurate oscillations every second.
The very best clocks nowadays are atomic clocks, which are very expensive and very rare. They use technology that I won’t pretend to understand to count the switching of caesium atoms between two different states. This happens at the rate of 9,192,631,770 times a second.
So accurate are atomic clocks that we now know that the speed of rotation of the Earth can vary slightly. They’ve also been used to test Einstein’s Theory of Relativity. Einstein predicted that time would pass more slowly for an object moving at high velocity and this was first tested in 1971 in an experiment where several atomic clocks were put in jet airliners while another remained on the ground. When the clocks were reunited the clocks showed different times, vindicating Einstein’s theory.
But what’s all this got to do with Siri, that I mentioned in the title?
Well, nothing. Except that Siri, and other products like Amazon’s Alexa, are part of the same trend towards greater accuracy. Perhaps I should explain.
Machine learning is an umbrella term for various computing techniques where data is processed to find useful information. One such technique involves the use of simulated neural networks, where linear algebra and calculus are used to try to simulate how brains process information. If that sounds unlikely it should be pointed out that this is exactly how number plate recognition systems work.
One of the holy grails of the tech industry is to produce a truly effective computerised voice recognition system. Using simulated neural networks, voice recognition software has been created that can transcribe spoken English with more than 95% accuracy. The industry goal is 99%.
The thing is, to achieve 99% will require a lot of number crunching and that in turn requires a lot of data. Just as counting more and more rotations of the escalator at Piccadilly Circus would have enabled me to give a more and more accurate rotation time, so having more and more voice data enables voice recognition systems to become ever more finely tuned.
Which is where Siri and Alexa come in. If it has ever occurred to you that Apple and Amazon have let you access their systems for an incredibly low price, it’s because they have. The big tech companies need to get voice data in huge quantities to feed and fine-tune their systems. Giving you cheap access to the systems is the most cost-effective way they’ve found of doing that.