A not-quite-random walk demystifies the algorithm

A not-quite-random walk demystifies the algorithm
Looking at the world through the lens of an algorithm illuminates some aspects but obscures others, says Malte Ziewitz, assistant professor of science and technology studies. Credit: Cornell University

The algorithm is having a cultural moment. Originally a math and computer science term, algorithms are now used to account for everything from military drone strikes and financial market forecasts to Google search results.

"People outside of computer science and math have come to describe these and other phenomena as 'algorithmic,' ascribing complex matters to a single, somehow magical entity that has developed a life of its own," says Malte Ziewitz, assistant professor of science and technology studies and a Mills Family Faculty Fellow.

But what's actually going on when we talk about, make sense of and rationalize the workings of computational technologies that seem so mysterious and inscrutable? And how could this be studied?

A not-quite-random walk around Oxford, England, offered Ziewitz some surprising answers.

He gave himself the task of exploring Oxford, where he was a scholar, with the help of an algorithm. The resulting stroll, and the insights that sprang from it, form the basis of his paper published Nov. 10 in Big Data & Society.

"This paper is not so much concerned with what algorithms actually are, but with what kind of work our reasoning with algorithms does," Ziewitz wrote.

In layman's terms, an algorithm is a step-by-step procedure for calculating the answer to a problem from a given set of inputs, Ziewitz says. Algorithms at the heart of Google's search engines, for example, take a search term, sort through web pages, calculate their relevance to the term and pick the top 10 pages.

Algorithms can also be simple. To structure their walk, Ziewitz and a friend wrote this one on a sheet of paper: "At any junction, take the least familiar . Take turns assessing familiarity. If all roads are equally familiar, go straight."

They could have created any number of instructions, such as "take every third left." But that would not have solved the problem they decided to tackle – a crucial requirement for algorithmic processing. "The key here is that you have to define the problem before your code can make any sense," Ziewitz said.

Off they went, consulting their sheet of paper as they went along. They walked down a busy road with buses, taxis and pedestrians. But only 60 feet from their starting point, their walk came to a halt, at a narrow alleyway.

Was it a road?

They realized they not only had to come up with a definition of "a road" but also test it in a specific situation. They decided to define a road as being wide enough on which to walk a bike, then added that line of code to their algorithm.

Looking at the world through the lens of an algorithm illuminates some aspects but obscures others, Ziewitz said. "Rather than looking at the beauty of the architecture or at other people, we were mostly focused on the road and the junctions," he said. "We tend to forget that a search for a certain term allows us to see those top 10 results but not a lot of other things."

Moving along, they found themselves at a Y-junction with two unfamiliar roads – so they added another line of code: "When all else fails, flip a coin."

"There was constant tinkering," Ziewitz said. "We had to not only redefine the rules but also redefine the world in terms of the algorithm."

That lesson applies to the idea of "relevant" search results, he said. "What is 'relevant"? Is it something of general public interest? It is something specific? Is it representative of diversity? It depends, and it's in the application of the decision rules that 'relevance' comes to mean a certain thing."

Eventually they found themselves on a paved lane that led to a parking lot. From a nearby building, a security guard shouted, "Excuse me! What are you doing here? This is private property."

They exited, quickly finding an adjacent road.

The incident highlighted another finding: It's tough to design an algorithm that will account for all possibilities. "Were we in Ithaca, you wouldn't want to be led down a gorge," Ziewitz said.

Part of his goal in writing the paper was to highlight a key challenge of the moment: how to deal with the fact that a term from computer and mathematics is all of a sudden being used across the humanities, social sciences and popular culture, Ziewitz said.

"What kind of work does the term 'algorithm' do in our reasoning?" he said. "How can we get at that? Put ourselves into a situation where we reason with the common-sense understanding of the term '' and see what happens."


Explore further

Better search engine results thanks to new method

More information: Malte Ziewitz. A not quite random walk: Experimenting with the ethnomethods of the algorithm, Big Data & Society (2017). DOI: 10.1177/2053951717738105
Provided by Cornell University
Citation: A not-quite-random walk demystifies the algorithm (2017, December 15) retrieved 21 April 2019 from https://phys.org/news/2017-12-not-quite-random-demystifies-algorithm.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
565 shares

Feedback to editors

User comments

Dec 15, 2017
It doesn't make sense. An algorithm is just some code with some function, so sure, lots of algorithms... funny that most people say... a "logarithm" and can't tell the difference.

Dec 15, 2017
In layman's terms, an algorithm is a step-by-step procedure for calculating the answer to a problem from a given set of inputs, Ziewitz says.

I'd put it more broadly:
"An algorithm is a method that minimizes an error function."
In the optimal case it leads to your error function becoming zero. However, most everyday algorithms are of the kind where 'close enough' is an acceptable solution - or where a perfect solution is not attainable because the target is either fuzzy in itself (like "navigate to my house") or not an rational number at all.

Looking at the world through the lens of an algorithm illuminates some aspects but obscures others

...or as Alfred Korzybski remarked almost 100 years ago: "The map is not the territory"


Dec 15, 2017
It depends on your education and scientific background, but it's more likely to be the reflect of the infinite movement, either predictable or not, of the universe.

KBK
Dec 16, 2017
For electronics people (technologist, technician, engineer, etc), this is programming 101.

You can't solve a problem until you understand the definition of the problem, with the stipulation that you can (and must) do some very basic programming (ie the verb part of algorithm) of the problem into the algorithmic formulation.

Nice article but it appears to be meant for the logic newbies, kinda thing.

Dec 17, 2017
I didn't quite understand, what the above article linked really wants to say: that we cannot apply fully random walk in the world filled by private properties?


That we cannot design a fixed algorithm for a complex real-world task without it failing constantly. The point is to debunk the popular myth, that when you program a computer to do something, it will automatically grant the computer human-like or superhuman powers to comprehend the situation and make choices accordingly - to solve the problem.

"There was constant tinkering," Ziewitz said. "We had to not only redefine the rules but also redefine the world in terms of the algorithm."


Think of how self-driving cars have to navigate the world - through an algorithmic understanding of what is a "road" or "junction". Well, if a human cannot always decide what it is, how can the computer? You know it when you see it.

Dec 17, 2017
I didn't quite understand, what the article really wants to say: that we cannot apply fully random walk in the world filled by private properties? This is maybe good enough subject for blog post of some teenager - but for peer-reviewed article payed from public taxes?


Not every article is for people in that particular field. This article was clearly meant to give people who DON'T write algorithms an idea of how algorithms are written, and why.

The article does a nice job of showing that writing an algorithm that works in the real world is not as simple as it looks.

Dec 18, 2017
The whole concept of artificial intelligence is, that there is no fixed algorithm.


Most implementations of AI, especially in a thing like a self-driving car, are fixed. They cannot afford the program changing itself on the go, because that would lead to unpredictable behaviour.

For example, one of the current problems of neural network emulations is catastrophic forgetting, where the network simply forgets whatever it has done before on learning new information, so on-line on-the-job training is not currently possible, or at best entirely too unreliable to implement.

That means the self-driving cars cannot modify their own algorithms on the go. They can't actually learn or adapt to anything new without going through the programmers and having their software updated manually.

The holy grail of AI is a self-constructing algorithm that modifies itself according to need, which doesn't exist. Hence, most AI is just clever wordplay on what "intelligent" means.

Dec 18, 2017
Instead of it, these algorithms serve like virtual high-dimensional landscape with hills and walleyes dynamically adjusted in such a way, every combination of input signal would find its optimal route to the combination of output signals.


That sounds like a half-understood marketing department version of fuzzy logic.

Every fixed computer program in the end is an algorithm that can be represented by a finite state machine, which means it can always be reduced to a long list of "IF x THEN y GOTO z" statements. That's all a computer program really is, as long as you don't allow any outside agent or influence to change the running code while it is running. This is John Searle's Chinese Room where John sits with the book of Chinese symbols and slavishly follows what the book tells him to.

The question with "AI" is simply, how do you come up with that list; do you program it in directly by yourself, or do you use some "training" method to come up with the algorithm.

Dec 18, 2017
Now, you might be tempted to argue that you can create a self-modifying program which alters itself, and therefore is not an algorithm, but you'd be forgetting that in doing so you must tell the computer how exactly to modify its own running code, under what circumstances and conditions.

It's like, you are God and the program is Adam, and you say "thou shall have free will", to which Adam replies, "Alright, but how? What do I do?", and so you go "Well, first you start by..."

...and so you've created just another algorithm. A meta-algorithm.

Meanwhile the real Adam wasn't programmed by God, or by evolution, but the real Adam is a happening and a continuously evolving interaction with the environment - not a separate entity you could isolate out of the world. A person driving along a road is continously defining the road by their act of driving it, by their very purpose of going somewhere, as roads have come to be by the need of people to travel efficiently.


Dec 18, 2017
The whole concept of artificial intelligence is, that there is no fixed algorithm

Quite the opposite.
The algorithms underlying AI are just at one removed from the solution of a particular problem. The algorithms employed are those that govern how individual neurons connect and excite one another. But when applied to a problem it is still the same thing: minimizing an error function (in this case deviation of the solution the AI presents from the desired result) which is fed back to make the next guess better.

Such AI algorithms are just vastly versatile (like genetic algorithms or finite element solvers are versatile in that they can be used on many different types of problem) - but they are still algorithms. They 'just' codify how to solve a problem in general.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more