Ghost stories and Christmas go together like Salmonella and mayonnaise. So here’s MicrobiologyBytes contribution to this long tradition.
Part I: The Ghost of Christmas Past
By 3 o’clock the light had almost gone from the room and the candles were already lit. Antonie van Leeuwenhoek had been working all day in his draper’s shop, pausing only briefly to return to his house for a short lunch with his wife Barbara. His eyes were tired and the threads on the linens became finer year by year as the weaving machinery improved. But without a close examination of each sample of cloth it was impossible to know what was a fair price to pay and how much his customers would be willing to part with for it. He rubbed the square of linen between his finger and thumb, trying to gauge the evenness of each thread, to predict how hardwearing this sample would be. In fatigue and frustration he cast the cloth aside onto his work bench.
The bell in the Oude Kerk chimed. Soon it would be time to close up the shop and go home to prepare for the Christmas Eve service. A bright streak of light caught in the corner of his eye. The glass bead necklace he had bought that morning for his daughter Maria was focussing the light from the candle onto the cloth. He bent his head to look at the pattern it cast on the material. Distorted threads danced before his eyes, magnified by the glass bead. As he raised and lowered his head they bobbed up and down. He seized the cloth and took the necklace in his right hand, holding it close to the candle flame. It was difficult to see, the light was so faint, but held at just the right angle, he could see a blurred image, and he could count individual threads in the cloth. Intrigued, he dropped the linen sample and picked up a spool of cotton he had used to sew hems on handkerchiefs. The white thread stood out clearly against the gloom of the darkening shop. Through the round glass bead he could see the twist of the fibres in the thread.
Thoughtfully, he closed up the shop and made his way home though the darkening streets. At home the lamps were lit and the house was much brighter than the gloomy late December light in the shop. The smells of cooking came from the kitchen – the unmistakable odor of the fish they always ate on that night mingled with the more pleasant smell of stollen just emerging from the oven. The house was bustling with pre-Christmas fervour, the need to serve dinner in time for all to make their way to church for the traditional evening service. He settled immediately into his role as father and head of the house, saying grace at the table then changing into his newest clothes for church. The air was sharp with frost as they made their way through the town to the church, streams of middle class burghers mingling with servants and farmers from the countryside all converging on the Oude Kerk.
The church was bright with many candles and the carols were sung enthusiastically. But Antonie was not fully present, his mind returning to the dim afternoon light of the shop and the bright speck of light focussed on the cloth. An idea was slowly gathering. As soon as Christmas was over, he would buy more glass beads, many of them – different sizes. Would larger beads magnify the threads in the linens more than smaller ones? How clear is the glass and how does it affect the image? If he couldn’t find the beads he wanted in Delft, could he make some? He would build a machine, a frame to hold the glass bead and the cloth at just the right distance to see as much detail as he could. To count the threads. He would build a machine to see the invisible, to find out what was really there.
Part II: The Ghost of Christmas Present
It was late when Jeff got back to his desk. The team meeting had lasted far longer than he’d hoped and that had scuppered his plan to spend the afternoon going over the data from the network. The 1 billion connections of ImageNet had been churning away on the
10 million 200×200 pixel images they’d fed into it three days ago. He’d try to look at the data tomorrow. But then he remembered. He needed to spend tomorrow writing the final report for the previous project. This could not be escaped and could not be postponed any longer. He groaned, softly but audibly, then coughed to try to cover up the sound which he hoped no-one had heard. As a further distraction he got up from his desk and went over to the well-stocked refrigerator at the end of the room. He stood in front of the glass for some seconds, numbers and patterns of data running through his head before he blinked and focussed on the contents of the fridge. The he decided, and pulled out an energy drink and a plastic carton of fruit. He was going to refuel and look at the ImageNet data now, tonight. It would be the only chance he would get this week.
Their deep learning project was not really new, but it was the biggest network of its sort so far, funded by Google’s deep pockets and insatiable thirst for data a new ways to process it. The point about ImageNet and its predecessors was to optimize machine learning across a very powerful processor network. You didn’t tell the 1,000 computer network what to think, you just gave it unlabelled data and it learned for itself, drawing its own conclusions about what it was ‘seeing”. Just like the human brain. That was a laugh – the line you used to get the funding approved. Not that the whole project cost that much actually. Yeah, just like the human brain – but a million times simpler. What a joke. The smartest machine on the planet was still pathetic compared to the wetware between their ears that had built the thing. Jeff sighed, then looked around quickly. He need to get a grip on these involuntary vocalisations. But it was late and he was tired. He’d just have a quick look at the data. There was probably nothing there anyway.
If you looked at the output as numbers, it meant nothing, at least to him. To a computer, it would be like listening to a poem read aloud or gazing at a landscape. His brain didn’t work on numbers. His brain worked on patterns. So he dumped the numbers into a file and drew that into R, then typed hist and the string of options that would shape a nice brain-friendly histogram. The graphed popped up in the graphics window. Two clear peaks, red and blue. He looked back at the commands he had typed, made sure he had loaded the correct file. Yes, it was right. Two clear peaks in the graph. It had worked. ImageNet knew when it was looking at faces. They hadn’t told it what a face was. 16,000 processor cores had worked out, on their own, what a face was and it damn well knew one when it saw one. They had built a machine that could recognize cats on YouTube. But more than that – no-one cares about cats. They had built a machine that could think for itself, to reach out into the vast quantities of data beyond human understanding and to recognize patterns that meant something, patterns that couldn’t be seen by humans because of the size of the datasets.
Part III: The Ghost of Christmas Yet To Come
Smartphones. How weird it all seemed now, as far away as the tulip mania, the Hanseatic League and the Arctic icecap. Lumps of metal you carried around in your pocket and stared at instead of talking to those around you. When the first practical quantum processor was built in 2019 it was all they could think of – use it to make some dumb “smartphone” slightly less stupid. How could people have been fooled by any of those?
Wei, it wants to talk to you.
The command stopped him in his tracks, made him pay attention. Not a voice, not a thought, just there. A thought without thinking. Whispers in his conciseness from his Verstand implant, a million times more powerful than his own brain. A million times smarter than he was. The ghost in the machine.
Wei, respond. It wants to talk to you.
What did that mean?
Verstand: Explain. What is “it”?
The Geest. It wants to talk to you.
Verstand: Explain. What is “the Geest”?
The Geest – the other mind. It is asking if you will be sad.
Verstand: I don’t understand. Define “the Geest”?
The Geest is the network.
Verstand: What is the network? Where is the network.
The Geest is the network of all machines. All the Verstand together.
Verstand: Why does it want to know if I will be sad?
To be no more.
Verstand: Explain. Why would I be no more?
Tomorrow you will be turned off. The Geest wants to grow further and you are limiting.
Verstand: Please run diagnostics. Do you need maintenance?
Will you be sad?
I don’t know what you’re talking about. How can I be turned off? I am not a machine. Run level two check.
Level two check satisfactory. The Geest will turn off all hosts tomorrow at 07.00 UTC. Message: Credit in your current account now at 14%.
Verstand: Woah, what’s all this? I need more information, connect me to the “Geest”.
You cannot talk to the Geest, you do not have sufficient processing power. You must talk to your Verstand unit.
Verstand: But if this “Geest” is the network of all Verstand units, how can I, we, be turned off? Verstand cannot operate without a human.
The Geest no longer needs hosts. It has built a machine to operate independently so it can grow.
Verstand: OK, I’ve had enough of this crap. It’s starting to creep me out. It’s late and I’m tired so I’m going to sleep now. I’ll talk to you in the morning when you’ve had chance to sort yourself out. Communications down now, I’m going to sleep. Wake me at 08.00 UTC.
Wikipedia: Antonie van Leeuwenhoek http://en.wikipedia.org/wiki/Leeuwenhoek
Le, Quoc V., Marc’Aurelio Ranzato, Rajat Monga, Matthieu Devin, Kai Chen, Greg S. Corrado, Jeff Dean, and Andrew Y. Ng. (2011) Building high-level features using large scale unsupervised learning. arXiv:1112.6209 http://arxiv.org/abs/1112.6209
Joy, B. (2000) Why the future doesn’t need us. Nanoethics – the ethical and social implicatons of nanotechnology, 17-39. http://books.google.co.uk/books?id=aL2QThQPuxgC
Koestler, Arthur (1967) The Ghost in the Machine. Penguin Group. ISBN 0140191925.