Algorithms are very powerful search tools: they have the ability to almost instantaneously sift through massive numbers of data points and order that information into a form that can help answer questions and decide courses of action. Despite their many positive uses, algorithms can generate self-supporting conclusions which, over time, can progressively narrow search outcomes and make certain types of decisions more likely, even if those decisions are not the best decisions. To counter this bias, deliberately seeking to break chains of information sources, considering outlier information and utilizing anti-algorithms can all help present alternative data which can lead to more balanced overall perspectives.
Independent and Dependent Algorithms
An algorithm, in its simplest form, is a set of instructions for carrying out a calculation. Algorithms are everywhere, but one particular place we come across them every day is at our computer when we carry out internet searches. Each time we conduct a search, sets of algorithms are used to take search queries, match them with information available on the internet, rank those results and then provide us with the outcome of those rankings. These emissaries of code descend into an impossibly large mountain of information and return with pieces of ancient history, technological innovations and, for those at a loss for lunch ideas, prize-winning guacamole recipes.
While there a great many types of algorithms, from a decision-making perspective we can divide algorithms into two large categories: independent and dependent algorithms. Independent algorithms are algorithms whose results are independent of the specific characteristics of the person who initiates the search query. For example, if I conduct a search for Socrates, an independent algorithm will return the same set of results regardless of whether I have a blog about Socrates, participate in discussion groups about Ancient Greece or buy books about Socrates on line.
Dependent algorithms, on the other hand, are search results that are directly related to my on-line profile and previous search behavior. These results might then be weighted to reflect such facts as where I am located, book purchases I have made and where I made those book purchases.
Both of these types of algorithms have many strengths. I just ran a search query for hotels in Rome and the results, generated in slightly over 1 second, were based on over 100,000,000 results. If one did not have the benefit of this ranking and spent one minute checking each source it would require 1,666,666 hours or 190 years assuming one did not take a break for eating, drinking or sleeping. Taking such breaks could well add another century to the information review task.
Dependent algorithms are also highly useful in that they return search results based on factors that are specific to us: our online profile, our previous content search history, the amount of time with spend engaged with certain types of content and the content we do not search all combine to create preferences that can skew the type of information that we receive. These algorithms, in other words, show us what we are most likely based on our profile to want to see.
The Illusory Truth Effect and Individual Search Bias
Despite the positive aspects of algorithms, they also have drawbacks in the presentation of information. With respect to independent algorithms, search rankings are a function not necessarily of the most responsive information but rather how search results are weighted. If rankings are weighted by prior site visits, the rankings can become self-supporting as high rankings support repeat site views which in turn supports even higher rankings. This creates a real risk of the Illusory Truth Effect, which is the tendency to believe false information simply because one has been exposed to it repeatedly.
A similar phenomenon can be observed with respect to dependent algorithms. As I listen to classical music on YouTube quite a bit and particularly like Bach I found that the more I listened to music by Bach the more Bach-related choices were presented to me as further listening options. Moreover these choices were progressively narrowed down until the suggeted listening choices were not only Bach pieces but variations of the same piece. Presumably if I continued this process to infinity I would be confined to listening to a single note.
The self-confirming nature of dependent algorithms means that what I am looking for can become not what is known but in essence what I already know or think I know. For example, if I am planning a trip to Sienna and I have never been there before, an algorithm-generated search runs the risk of designing an itinerary that others have planned (in the case of independent algorithms) or one that reflects my own life in Chile (in the case of dependent algorithms). In both cases, this paths may lead me to experience a Sienna that is not what it is or would be most interesting for me to see, but rather what it is to others or what I am most interested in apart from Sienna. Paradoxically, there is a tendency for this sphere of knowledge to shrink over time rather than grow.
Decision-Making and Anti-Algorithms
Inherent biases in algorithms have important decision-making implications. Good business decisions require trying to understand often highly complicated phenomena, such as market movements or potential consumer behavior. It is rarely the case these types of phenomena can be understood from a single perspective and harder still when that perspective is biased.
Because of this, good decision-making requires a structured approach for challenging not only conclusions but also the manner in which those conclusions are reached. The way to do this is to consider outlier data and, in a sense, build into decision-making patterns a type of anti-algorithm: information and search patterns which do not follow independent or dependent algorithm logic. Rather than creating analytical chaos, this means balancing traditional search query results by introducing breaks in algorithmic reasoning that highlight different possibilities and perspectives.
There are many practical ways that this can be done. To provide a few examples of anti-algorithm approaches:
- When searching for a restaurant an algorithm may highlight the views of food critics but an anti-algorithm could highlight the views of restaurant workers
- When searching for an approach on how to learn a language an algorithm may highlight the top language schools but an anti-algorithm may highlight the approach of tour guides who had to learn a language almost instantly to remain employed
- When searching for an approach on how to get around a city an algorithm may lead you to the most powerful digital mapping devices but an anti-algorithm may lead to you taxi drivers who know the city inside and out
Each of the above examples point out the reality that knowledge driven by statistic-driven algorithms is only a certain type of knowledge that may contain real knowledge gaps: assumptions that food critics know the most about food, which sounds logical but may not be true, drive the fact that food search results may be biased to reflect food critic preferences.
On the other hand, assumptions that restaurant goers must know the most about food may include the bias that the vast majority of restaurant goers are tourists and their restraurant choices may be based not on food quality but on the proximity of restaurants to tourist destinations. Regardless of the direction from one approaches opinion, the potential gaps and biases in all opinions have to be questioned in order to get at real knowledge.
Conclusion
Despite their power, algorithms can present a very skewed view of information. To ensure flexible decision-making it is useful to combine algorithmic approaches with outlier information and types of anti-algorithms that can present alternative facts and information and order information in different ways. To get at real knowledge that can help face complicated market and business situations, one needs to be prepared to take the unlikely path: turn left at the right hand sign, ask the person who never speaks what their opinion is, read a few books by unpublished authors rather than ones that are at the top of the bestseller list and perhaps, for a period of time, leave one’s own perspective behind.