Decontextualized algorithms

17

October

2018

No ratings yet.

In Weapons of Math destruction (2016) Cathy O’neil describes a system for evaluating teachers in the USA, called IMPACT.
This model predicts how well students will perform, and to compare the expected results with the actual results. IMPACT is used to monitor how well teachers perform. If students consequently score lower than they are expected to, the teacher is fired based on this algorithm.
O’neil argues a model like this are perceived as more efficient and fair than ‘regular’ evaluations submitted on paper which filled in by humans. This is because this model “not only saved time but also was marketed as fair and objective. After all, it didn’t involve prejudiced humans digging through reams of paper, just machines processing cold numbers” (O’neil, 2016, pp 3).
However, there may be problems with IMPACT and algorithms like it.
O’neil argues IMPACT is an algorithm that tries to quantify or formalize the behavior of the teacher. According to O’neil, Impact may be used inappropriately because it is applied in a context that is not suited to formalize. For this reason, O’neil calls IMPACT a Weapon of Math Destruction( WMD). WMD are systems that inappropriately use algorithms. She calls such algorithms decontextualized.
A decontextualized algorithm is a “bias that originates from the use of an algorithm that fails to treat all groups under all significant circumstances” (Friedman & Nissembaum, 1996, pp 334). The fair treatment of all groups under all circumstances, is a high goal to aim for. No system may be able to live up to this goal. But the massive use of algorithms to decide how loses her job takes false positives as a hazard of doing business.
Like O’neil describes, models portray an image of mathematical certainty and objectivity, “[n]evertheless, many of these models encoded human prejudice, misunderstanding, and bias into the software systems that managed our lives” (O’neil, 2016, pp 3). These biases emerge because algorithms are in their core statistical or mathematical models. These models function well by the virtue that data are complete and objective. This is not always the case. A teacher may for example teach a class which educational level is lower than the model assumes. Every statistical models carries the risk of false positives.
Therefore, O’neil argues WMD’s should be recognized as such and abolished.
Do you agree with Cathy O’neil?
Do you know examples of algorithms that are used in the wrong context and therefore cause problems?
Sources
Friedman, B., & Nissenbaum, H. (1996). Bias in computer systems. ACM Transactions on Information Systems (TOIS), 14(3), 330-347.

O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Broadway Books.

Please rate this

Does IT have moral values? – my philosophy thesis

16

October

2018

No ratings yet.

The choices designers and developers make when they create an app or website can significantly impact your life.
In the information strategy lecture, we learned that the turning on or off, of the anonymity function in a dating app can significantly impact the number of matches you get (Bapna et al., 2016).
We also learned that the (not) showing of the buyer ID at the Dutch Flower auction, impacted buyer behaviour significantly (Lu & Heck, 2016).
What these examples illustrate is that information technology has a profound impact on human behaviour.
Some authors argue we should attribute moral value to IT because of this impac.t
To me, this is fascinating and therefore I decided to write my philosophy thesis about the question: can IT’s have moral values?
To answer this question, we must go back to the start of the morality in technology debate:
The idea that technology could incorporate values wasn’t generally accepted and seldom or never applied to particular technologies, until the 1980’s. This was noticed and contested by Langdon Winner in his famous essay Do artifacts have politics (1980). He argued that “[i]n controversies about technology and society, there is no idea more provocative that the notion that technical things have political qualities” (Winner, 1980, pp 121).
Winner argued that artifacts do have political qualities and influence human life in profound ways impossible to account for referring merely to human actions. His view took root and inspired a new area of philosophical thinking about how material things can have values incorporated, embedded or built in them.
This view is stronger than to say that artifacts can be used for specific purposes, or that artifacts favor a certain use. Winner wants to inquire if artifacts “have been designed and built in such a way that it produces a set of consequences logically and temporally prior to any of its professed uses” (Winner, 1980, pp 125). The word prior must in this context be understood logically, not temporally. It is to say that artifacts can have observable or traceable values in them aimed at certain effects but without actually being used. This idea is what was most controversial about the essay.
The consequences of the embedded values in artifacts are that the material world has a strong impact on human behavior. In Winner’s view, built structures influence almost all aspects of human behavior, freedom and feelings, whether consciously experienced or not.
Winner’s ideas are applied to ICT’s by Lawrence Lessig, a legal scientist at Stanford University. He acknowledges many of the ideas articulated by Winner and extrapolated them to the digital world. He argues that there is a modality that shapes the virtual world in the same way as architecture shapes the built environment. By making this comparison Lessig incorporates Winner’s concern about the importance of the way we build our artifacts. He calls attention to the “architecture of cyberspace”, his way of describing “the software and hardware that make cyberspace the way it is”, designated by the simple word code (Lessig, 1999, pp 503). In other words; code is the architecture of the virtual world.
Lessig’s most important insight is that code regulates and constraints behavior effectively.
“In cyberspace technology constitutes the environment of the space and it will give us a much wider range of control over how interactions work in that space than in real space” (Lessig, 2009, Pp 15).
It is interesting to think about Lessig’s claims in the light of information strategies.
Do you believe IT and the information strategies chosen by firms have significant impact on your life?
Should we argue that IT’s have moral values in them?
Sources
Bapna, R., Ramaprasad, J., Shmueli, G., and Umyarov, A. 2016. One-way mirrors in online dating: A randomized field experiment. Management Science, 62(11), 3100-3122. Links to an external site.
Lessig, L. (1999). The law of the horse: What cyberlaw might teach. Harvard law review, 113(2), 501-549.

Lessig, L. (2009). Code: And other laws of cyberspace. ReadHowYouWant. com.

Lu, Ketter, Heck – 2016 – E XPLORING B IDDER H ETEROGENEITY IN M ULTICHANNEL S EQUENTIAL B2B A UCTIONS 1-annotated.pdf
Winner, L. (1980). Do artifacts have politics?. Daedalus, 121-136.

Please rate this

How to become a better predictor

12

October

2018

No ratings yet.

In my last blog, I explained why people are bad predicters.
If we know we are bad predictors, the question of course becomes: what can we do about it? How can we become better predictors?
Nate Silver offers a solution in his book The Signal and the Noise. And he should know! Nate Silver predicted the outcome of 49 out of 50 states correctly in the Obama-McCain election in 2008.
In order to become better predictors, Silver argues people should list their predictions, and confront themselves with the actual outcomes.
Using the outcome of past predictions is a technique used in Bayesian statistics.
According to Silver, we should become Bayesians. Here’s how:
Imagine you don’t learn for the exam, and you’ll score a good grade. You now want to predict the chance that not learning leads to getting good grades.
To calculate the conditional probability of you scoring higher if you put in more work Pr (A|B) , you need:
– the conditional probability that you put in more work given you got a higher grade Pr(B|A)
– the chance that you put in more work Pr(B)
– a prior: your initially held belief that you would get a good grade Pr(A)
You can calculate the result with the following formula:

formula

Because you use a prior in the formula, which is your initial belief, your prior knowledge is relevant in making an estimation.
This is an advantage of the Bayesian method, as it uses your initially held beliefs as a source of knowledge.
But it also serves as a way to update your beliefs. If your predictions are constantly wrong, you should adjust your beliefs.
In other words: your prior changes.
The consequence of this method is that if you are confronted with your own biases because your estimations are often wrong, you can fill in the Bayes formula multiple times, every time with an adjusted prior.
In this way, if you update your prior after every test, your beliefs will slowly diverge to the most accurate prediction possible.
So, if you want to become a good estimator, you should become a Bayesian and keep track of the result of all your past predictions!
I’m curious what you guys think: is this a good strategy for gathering truthful information?
In any case, I recommend everyone to read The Signal and the Noise as it is a great read!
Sources:
Silver, N. (2012). The signal and the noise: why so many predictions fail–but some don’t. Penguin.

Please rate this

What is information? (and what is the difference with data)

10

October

2018

No ratings yet.

We as BIM students, constantly hear about information. Information strategy, information technology, information asymmetry, etc, etc.
But what is information exactly? And what is the difference between data and information?
These questions have been asked by mathematicians and philosophers. And there is a vast body of literature answering these questions in great detail.
Because you probably don’t want to get in this area of literature yourself, I summarized the most important points for you.
First of all, it is important to know that it is hard to exactly pinpoint what information is. Luciano Floridi, a well-known information-ethicist, warns us that “information is notorious for coming in many forms and having many meanings” (Floridi, 2010, pp 1).
In fact, information can be seen both as a technical concept and a semantic concept. Both concepts have elaborated theories underlying the definition. The distinction between data and information is a subtle one.
Information in the technical sense is understood as “the probability of a signal being transmitted from device A to device B, which can be mathematically quantified” (Shannon). This technical description of information, excludes meaning. This technical description of information is defined as data. Data are nothing more than quantifiable signals, without meaning. In its most rudimentary for, data are thus bits.
To speak of information, we need something more than just data. Only when data have meaning, we speak of information. Information can thus be understood as data plus meaning.
Data should be understood as a “two-term relation” (a signal from device A to device B).
Information should be understood as a “three-term relation” (a signal from device A to device B, meaning something to C).
This distinction between data and information was recognized by the men who developed the “mother of all models”, the mathematical theory of communication, also known as the Shannon-Weaver model.
Maybe you have already seen it:

information model

In this model you can see that the word “information” is below informer and informee. This is because in the space between the informer and the informee (econding and decoding), there are only data that do not have meaning. Data only have meaning when they can be understood as saying or meaning something. This is the case when the data reach a person that can understand the data to say something. Only then, we speak of information.

Sources:
https://en.wikipedia.org/wiki/Shannon%E2%80%93Weaver_model
Floridi, L. (2010). Information: A very short introduction. New York: Oxford University Press.

Please rate this

Probably, you’re not good at predicting (and why)

2

October

2018

No ratings yet.

You probably make predictions all the time.

How many days you will need to finish your group assignment, the number goals Ajax will concede against CL opponents and what time you will be home for dinner. Have you ever asked yourself if you are any good at predicting?

Probably not as good as you think.

In Thinking Fast and Slow, behavioural economist Daniel Kahneman argues that people, are not very good at statistics and making predictions. While people making statistical claims and predictions every day, more often than not, their predictions are very far from the truth. This also goes for highly intelligent people.

In addition, Nate Silver discusses how bad humans are at predicting in his book The signal and the Noise. In his book, Silver shows how far often political polls, economic models and weather forecasts often are beside the truth.

We need to stop and admit it: we have a prediction problem. We love to predict things – and we aren’t very good at it” – Nate silver

The failure of prediction is also apparent in software cost predicting. Cohn states that “project schedule estimates are typically as far off as 60% to 160%”.

The question is: why are we so bad at predicting?

A large part of the reason is that people are way to optimistic. People overestimate their ability and underestimate the tasks they are about to do. Even when a task takes much more effort than you expect, you predict it will go better next time.

So, in a nutshell: people refuse to learn from their experiences. For example, a project manager of a software project that lasted 40% longer and costed 50% more does not predict his next project will take longer. Instead, he argues: this was a particularly vague customer, and we had a lot of bad luck. Next project, we’ll do better. So, instead of adapting expectations, people think of reasons why it went bad, and how they will do better next time.

This, among other factors, prevents people from making accurate predictions.

In my next blog, I will go into more detail on how to make good predictions and explain how Bayes rule can be used for this purpose.

For now, I’m curious: are you a good predicter? Why do you think you’re (not) a good predicter? And how relevant do you think predicting is for companies in a digitalized world?

 

Sources

Cohn, M. (2005). Agile estimating and planning. Pearson Education.

 

Kahneman, D., & Egan, P. (2011). Thinking, fast and slow (Vol. 1). New York: Farrar, Straus and Giroux.

 

Silver, N. (2012). The signal and the noise: why so many predictions fail–but some don’t. Penguin.

Please rate this

Smartphone addiction – How to manipulate a mind

1

October

2018

5/5 (1)

“Man has an almost infinite appetite for distractions”
Aldous Huxley, brave new world

You have probably never heard of BJ Fogg. But he has impacted your life more than you could imagine. He is the godfather of “behavioral design”, the study of designing human behavior. Fogg taught his students how to capture human behavior and how to create a lasting connection between a human’s mind and a product. This approach works especially well for (social media) apps.
Among Fogg’s students were many later Facebook and Google employees. These students put Fogg’s insights into practice by creating strong “hooks” in the minds of people, related to their apps. They do this by creating a triggers that are linked to rewards. For example, the vibration of a phone (trigger) to a social message (reward). According to Fogg, “when motivation is high enough, or a task easy enough, people become responsive to triggers such as the vibration of a phone or Facebook’s red dot”. Like a Pavlov-reaction. This is how the makers of apps can create a strong connection to a person’s mind and their apps.
This has resulted in the massive use of smartphones by millions of people around the world. Studies indicate that frequent users open their smartphones as often as 200 times a day! That means every five waking minutes (Falaki et. al., 2010). Checking your smartphone so often seems unhealthy and consequently, some argue that it should be indicated as an “addiction”. The creators of social media have been criticized for intentionally making people mentally dependent on their products. However, you might argue people are responsible for their own behavior.
So, I have two questions for you:
How often do you check your smartphone each day?
How often would you want to check your smartphone each day?

Sources
Bentley, F., K. Church, B. Harrison, K. Lyons and M. Rafalow, M. “Three Hours a Day:
Understanding Current Teen Practices of Smartphone Application Use.” (2015) Accessed
November 11, 2016, arXiv:1510.05192.
Falaki, H., R. Mahajan, S. Kandula, D. Lymberopoulos, R. Govindan and D. Estrin,
“Diversity in smartphone usage.” Proceedings of the 8th international conference On mobile
systems, applications, and services. ACM (2010) 179-194. Accessed November 2, 2016. doi:
10.1145/1814433.1814453
Huxley, A. (1936) Brave new world
https://www.1843magazine.com/features/the-scientists-who-make-apps-addictive
https://www.theguardian.com/technology/2017/oct/05/smartphone-addiction-silicon-valley-dystopia

Please rate this

Do transaction costs of information fall or rise in the info-economy?

24

September

2018

No ratings yet.

According to Thomas Friedman, IT and fast internet have decreased the transaction costs of information since the emergence of the information economy. While many intuitively agree with Friedman, Philip McCann argues to the contrary.
According to McCann, the information economy actually increased the transaction costs of information. While it is true that data can be transmitted over distance faster, that is not sufficient ground for claiming that the transaction costs of information have declined. To see why this is the case, spatial transmission costs must be distinguished from spatial transaction costs.

  • Spatial transmission costs are the costs of moving information across space. These costs have indeed fallen with the introduction of IT and fast internet.
  • Spatial transaction costs are “the costs associated with engaging in and coordinating activities across space”. Transaction costs have actually increased because in the information economy, information is more complex and more valuable than before.

The complexity of information arises from the amount of (big)data, the multitude of sources of information. The value of information has increased as more business rely on data for making decisions, or even sell information as business model. In other words, knowledge is so valuable that the opportunity costs of not having information have increased. Consequently, in the information economy where many people perform knowledge-intensive work, it is both important and difficult to have real-time, trustworthy and understandable information.
Managing the complexity and value of information is done by ensuring frequent face-to-face contact between workers both inside and outside the firm. Think for example of consultants who work on location at their clients, or of Silicon Valley firms that are located close to one another. Thus, contrary to what you might expect, for many firms active in the information economy, the transaction costs of information have increased because of the digitalizing world.

Sources:
Thomas Friedman, The world is Flat (2000)
Philip McCann, Globalization and economic geography: the world is curved, not flat (2008)

Please rate this

How Alexa can get into trouble by antitrust

20

September

2018

No ratings yet.

Artificial intelligence (AI) driven virtual assistants such as Alexa (Amazon) are predicted to be ground-breaking new technologies that will transform the way we work and live. However, virtual assistants face an important legal problem with possibly enormous economic consequences. This problem is competition law.
Virtual assistants potentially form a large problem for the economic freedom of consumers around the world. That is because virtual assistants give their owners all kinds of advice. You can ask Alexa what to eat, where to eat, where to do grocery shopping and how to travel. Consequently, virtual assistants have a great influence on the buying behaviour of consumers. This influence would not be a problem when big tech companies would only create technologies. However, the big tech companies are transforming into “Hubs”, as they are spreading their activities to all kinds of market. Amazon for example, is said to plan the opening of 3000 grocery stores.
Watch this:

The combination of the exploitation of a virtual assistant and grocery stores gives Amazon the possibility to recommend the users of Alexa to visit Amazon stores, whenever they ask for a shop. The transferring of market power from one industry (virtual assistants) to another (grocery shopping) is viewed as a breach of antitrust.
Google has been penalized 2,4 billion euro for a similar use of market power. In this antitrust case, Google used Google shopping to recommend its own products at the top of search results, over the products of other companies. In this way, Google used its power in the search engine industry to gain an advantage in the retail industry, which is a violation of competition laws in the EU.
So, what do you think? Will virtual assistants face similar problems as Google shopping? And should antitrust play a big role in the shaping of our digital world?

Sources:
https://www.bloomberg.com/news/articles/2018-09-19/amazon-is-said-to-plan-up-to-3-000-cashierless-stores-by-2021

Google tweaks search ads after EU shopping antitrust ruling

Please rate this

The algorithm that determines the order of your Instagram timeline

17

September

2018

No ratings yet.

The photos on your Instagram timeline are not ordered chronologically, but by an algorithm.
The algorithm takes various factors into account to determine which posts the user sees.
While hated by some users, the algorithm is essential for the long-term viability of the Facebook-owned medium’s business model.

Instagram would lose its dominant position if would not adopt an algorithmic based ordering of its timeline.
That is because a social platform that is gaining in popularity, like Instagram, witnesses a growth in both the number of users and the average posts per user. Consequently, many posts remain unseen as there are simply too many posts to show to the average user.
To maintain a healthy mix of engaging content and Business posts, Instagram needs to control user timelines by algorithmic ordering.
The exact algorithm used by Instagram is unknown, but it is known that the algorithm takes the following factors into account:
– interest of the user
– recency of the post
– relationship between user and poster
– usage of the user
– frequency of posting of the poster

The problem with algorithmic ordering of timelines is that Instagram gains more control over the timelines of its users, while users have no knowledge of the algorithm and might have different interests than the makers of Instagram.
Instagram wants is users to be engaged, spent as much time on the platform and click on ads, incentivising the makers of the platform to show users “engaging” content close to the user’s known interests.
At the same time, users might value moderate use of the platform, variation of content and connectedness to both friends and distant acquaintances.
So algorithmic timelines could place the interests of users and makers in direct contrast to each other.

What do you think? Is algorithmic ordering of your Instagram timeline better or worse than chronological ordering?

Sources:

How Instagram’s algorithm works

Please rate this