Information autism X algorithm autism (autistic algorithms) ===> sense-making scarcity
Today it occurred to me to coin a term that encapsulates how applying quantitative methods and approaches to code without also incorporating quality dimensions (which include emotions, tastes, perceptions, wit, values and beliefs) leads to deficient decision-making. I’m calling it “ALGORITHM AUTISM” (© Twain 刘秋艳, 16 May 2010).
Its definition is as follows:
Algorithm autism is a state of syntax and audiovisual deficiency in the input, process and output stages of code which don’t enable the machine to interpret or relate the socio-emotional elements of content, and thereby understand its context and make sense of it.
Given my previous definition of information as “a consciousness of quantity and quality that enables differentiation and contextualization over time”, the definition of INFORMATION AUTISM is as follows:
Information autism occurs when the constituents of information are purely quantitative, objective facts or data objects which don’t carry any associated subjective contextual qualifiers such as human emotions, tastes, perceptions, wit/humor, beliefs and values.
Examples of autistic information would be most of the equations in quantum theory and that underpin risk management models (===> implications for understanding why and how the global financial crisis happened). Examples of non-autistic information could be found in psychometric and EQSQ tests such as Myers-Brigg, Alpha Assessment for Leadership, Belbin, Saville-Holdsworth and Simon Baron-Cohen etc. Simon Baron-Cohen is the Cambridge professor cousin of Sacha Baron-Cohen of ‘Bruno’ and ‘Borat’ movie fame.
Why did I choose this designation of autism?
Well, it’s well known in medical circles that autistic children are often highly intelligent (numerate savants and linguistic encyclopedia). However, they’re afflicted by an inability to read, interpret and understand the emotional states of others and nor do they have much concept of social relationships and their role in the dynamics. Their brains process mechanistic, metaphysical inputs (numbers, words) but don’t capture or process those socio-emotional codes that would make them understand the philosophical and psychological motivations underpinning another human being’s communications and interactions.
So………if we think about the neural networks of the World Wide Web and the codes which are streaming between its nodes…….It’s arguable that there’s algorithm autism. We have lots of functions that enable us to capture and interpret numbers and words (binary, probabilities, data objects). Yet none that enable us to capture and interpret socio-emotional context and thereby make sense of the whole of it. Therefore, the algorithm is itself autistic.
This has far-reaching consequences in global finance terms because it means the risk management models — which are built from increasingly complex and sophisticated mathematics (chaos, Black-Scholes etc.) — as well as economic models en masse are FAILING TO CAPTURE THE UNDERLYING PHILOSOPHICAL AND PSYCHOLOGICAL HUMAN QUALIFIERS which actually drive human engagement, intent and consumption of any piece of content, product, service, lifestyle etc. It’s been commented upon broadly in this MIT Technology Review article, even if no solutions have been proposed there.
Now, in the case of human autism there are methods to help those with the condition to deal with it. Plus there’s support for their loved ones to identify it, appreciate that it makes their autistic child special in different ways from other children and work with it. In the case of algorithm autism………..We need and are going to find methods in object-oriented programming (OOPs) in conjunction with whatever semantics technology has to offer (NLP, machine learning, AI, neural nets et al) to re-orient code pathways so that the machines do comprehend those socio-emotional elements of content that will enable it all to make sense and help us arrive at more informed decision-making.
And, no, the existing sentiment analysis algorithms do not tackle or resolve their own autism bias. They’re not able to interpret socio-emotional context with much degree of accuracy. For example, let’s take a look at Twitrratr and the search term “facebook” today and let me highlight some obvious examples of “algorithm autism”:
This last week has seen unprecedented criticism of Facebook’s privacy policy with Google trends showing that “how do I delete my facebook account” is increasing in popularity:
The Institute of Quantitative Studies at Harvard University has also pointed out how the number of words in Facebook’s privacy policy has grown over the last 5 years, from 1004 to 5830 (which makes it even longer than the Constitution).
So……..existing empirical and anecdotal evidence — Google trends and numerous examples of users threatening to leave Facebook en mass or griping about it all across the socmedia space — indicates to us that the sentiment towards Facebook is negative. Yet Twitrratr shows only 3.89% of comments as negative, 10.03% as positive and 86.08% as neutral. Superficially, this may look acceptable but when we manually sanity-check some of those comments, it becomes clear that Twitrratr’s sentiment extractions are……..AUTISTIC. They can’t read or interpret the socio-emotional context to any degree of reliable accuracy. Here are some specific examples from about 50 comments:
(1.) lied! i didn’t have lunch…i just worked. now i’m having lunch and playing a little on facebook. still listening to cool vibes — This has been bucketed into “positive” when in fact the word “cool” is in reference to the music this commenter is listening to and not specifically to facebook itself.
(2.) methinks i want to take some modeling pics. any burgeoning/talented photogs need willing subjects? xxxxxx@gmail.com, facebook or dm me — This has been bucketed into “positive” when in fact it belongs to “neutral” because it doesn’t make any sentiment about whether the commentator likes or dislikes Facebook itself.
(3.) glad to see most of the online retail partners I work with have a presence on Facebook — This has been bucketed into “neutral” when it’s clearly a positive statement.
(4.) found living kidney donor through Facebook – another perk of social media — This has been bucketed into “neutral” when it’s clearly a positive statement.
(5.) is posting 1 obama supporting link for every anti obama post he sees on his facebook gotta keep it fare — This has been bucketed into “positive” when in fact it belongs to “neutral” because it doesn’t make any sentiment about whether the commentator likes or dislikes Facebook itself. Actually, the brand in question is President Obama rather than Facebook!
(6.) having fun the toolbar from www.cooliris.com. wicked for my iphone, facebook, and surfing……and i’m not getting paid to pimp
— This has been bucketed into “negative” because the sentiment engine interpreted the word “wicked” as a negative term when it’s clearly a positive statement — albeit not for Facebook specifically but for cooliris. So Twitrratr’s natural language processing is being doubly autistic.
(7.) eathing lunch and sending out birthday greetings on facebook while reading the comics in today’s paper. adhd or wickedmulti-tasking? — This has been bucketed into “negative” because the sentiment engine interpreted the word “wicked” as a negative term when it’s clearly a positive statement — albeit not for Facebook specifically but for the commentator’s own ability to multi-task. So Twitrratr is being doubly autistic in this instance too.
Some people might argue that 10-20 percent of inaccuracy in 50 comments is not a big deal. However, there is something called compounded inaccuracy which — like compound interest rates — can accumulate to quite a sizable influence. The main issue, though, is how this socio-emotional deficiency of context then propagates and permeates throughout the rest of social media which……CONTRIBUTES TO MORE NOISE RATHER THAN TOWARDS SENSE.
Ergo, autistic algorithms are making the Web less intelligent.
How I arrived at this term of “ALGORITHM AUTISM” was the result of twaining elements from these topics: dsycalculia, DNA, Objective-C language, IQ-EQ, dexterity and discern. One day (when my book gets published), I’ll explain exactly how this twaining happened.
:*).















