Ants are not intelligent creatures, they obey very simple sets of rules, yet the ants nest, with its elaborate tunnels and foraging parties, is enormously complex; it’s as if the queen ant sits in the command centre, dishing out orders. She doesn’t of course, instead we know that an ants nest is an example of an emergent system — a complex system built from simple interactions. There is nothing simple about the software robots that form RPA. But the AntWorks RPA system, known as ANTstein, tries to create order from disorder, applying fractals and RPA, by using fractal based algorithms, applying data ingestion and even has something akin to a queen ant. It claims to be solving the problem that is haunting the RPA business, namely the challenge of scaling RPA.
Fractals and RPA
The Polish mathematician Beniot Mandelbrot, a man who said he was interested in the art of roughness, and described himself as a fractalist, is the man who came up with the word fractal. Look at a coastline from a distance, it usually has a rugged appearance, coves and inlets, and beaches, perhaps. Then zoom in, look at a portion of that coast, the same ruggedness is there, while other, proportionally smaller coves and inlets appear. Then zoom in again, and again, the coastline might not look identical, but it will look similar to the original coastline. At the heart of fractals is self-similarity. Mandelbrot studied how it was possible to “measure roughness.”
Data can be pretty rough too, especially when it is unstructured. This is the essence of the AntWorks system, using fractals to create clean data. It boils down to the difference between a fractal and neural engine.
Asheesh Mehra, AntWorks co-founder and CEO, explains with a metaphor concerning an apple. He said: “If I have to train a neural engine to recognise an apple, it would need to be trained on an extra small apple, a small apple, a medium apple, a large apple, an extra large apple, in every shape, colour and form for the neural engine to be able to recognise any apple that occurs. Because fractal are fundamentally about pattern recognition, you need to train the fractal engine only on one apple, irrespective of size. That’s because the pattern of an extra small apple, or the pattern of an extra large apple, will remain the same.”
Gartner releases first-ever Magic Quadrant for RPA software
How fractals support RPA?
“Fractal science” said Mehra, is based on the premise of self-similarity and that networks of neurons carry similar but not identical signals about patterns. Fractal science is a more deterministic science, whereas neural is a more probabilistic science. Since fractal is a more deterministic science, it needs a smaller representative data set for training. Because you need a smaller representative data set for training, the downstream benefits automatically accrue. You need thinner infrastructure, lesser computing power and it becomes much easier and faster to implement.” He suggests that this contrasts with “neural engines, where you need large data sets, which then means you need larger infrastructure, more computing power, which leads to cost increasing and time increasing to deploy the solution.
“I would say neural is also very people dependent where you need large resource pools to be able to train the engine,” whereas Fractals need less people.
Ergo, goes the argument, fractal based RPA is easier to scale — at least that is the gist of what Mehra says.
So that’s the idea behind fractals and RPA.
Intelligent automation to replace RPA, suggests AntWorks
Ingesting data
Mehra suggests that there are four types of data: “structured, unstructured, image and there’s inferred data.” He says that the rationale behind AntWorks was to create a data ingestion engine that could read all four data types.
But aren’t the other RPA companies looking at data ingestion too?
Mehra concedes that they are but that this is a recent development — for example Blue Prism’s Decipher, announced earlier this year.
Mehra argues that while AntWorks was late to the RPA party, by coming in later it was not weighed down by legacy, and that its data ingestion tools are consequently in the lead. Indeed, AntWorks doesn’t even describe itself as an RPA company, rather as an intelligent automation business.
Does that mean AntWorks will do to existing RPA companies what Netflix did to Blockbusters? That is unlikely, but maybe its focus on fractals and data ingestion gives it a sufficient niche to carve itself out a slice of a market dominated by a small number of companies that between them have raised well in excess of a billion dollars.
Is RPA overhyped, scalable or a bandaid? Are decision engines next?
Is RPA like a sticking plaster? Is RPA overhyped? Are decision engines the future?
Is RPA a bandaid?
Critics argue that RPA is a short term fix to a problem that will eventually be overcome by more integrated software.
This is a charge that unsurprisingly the RPA companies disagree with. Their defence is that RPA is evolving, in fact many have stopped using the acronym RPA and are saying intelligent automation instead. For example, recently Jason Kingdon, chair at Blue Prism, told Information Age that: “Critics say that RPA is a brief moment between now and when we get — some great all singing solution, but it’s complete nonsense… an RPA tool, can do the work of between nine and 50 people.”
Mehra suggests that RPA can mutate away from being a sticking plaster by becoming intelligent automation. “RPA is sticky tape,” he says, “intelligent automation is using a nail. If you put up a painting using Blu Tack or sticky tape, it will stay up for a period of time but at some point it’s going to fall. If you use nails, that painting is going to stay up for ever and ever until an earthquake hits or something like that.”
Why RPA? Blue Prism chairperson tells Information Age why he thinks RPA is a game changer
What’s this got to do with ants?
Finally this takes us to ants and in particular the queen ant, or what AntWorks call the queen bot.
Mehra explains: “The queen bot is constantly monitoring the ant bots. If the queen bot sees that one particular bot has failed, it allocates that bot’s task on to another desktop.”
The queen bot approach also makes something called intelligent workforce management possible. For example, in the case of an insurance company, there might be three bots, one doing claims, one payroll and a third accounts payable. If the bot working on claims is struggling with the work load, but the other two are finished for the day, the queen bot may allocate the two underworked bots to assist the first bot.
It’s not exactly like an ant’s nest. The queen ant doesn’t give out orders, there is no master plan, order is created by each ant following simple rules that are hardwired into it, such as follow other ants. But if you see an ants nest as a metaphor, with the queen ant in control, workers following her orders, then that seems to be an approximation to what AntWorks and it ANTstein product is meant to be about.
Does the UK really need more robots? In RPA the UK is a leader
