Nowadays, it is impossible to imagine our lives without algorithms. They play a particularly important role in connection with the Internet and computers. They are often seen as the driving force behind digitization, because without them innovations such as the first PC, the cell phone or even self-driving cars would be impossible.
In the process, we often see a "black box" when it comes to how they actually work. Alternatively, we only talk about the negative effects, but leave out the real societal values that are added. This article aims to both highlight these added values and discuss the potential of algorithms.
What is an algorithm?
The German science journalist Christoph Drösser describes algorithms as something that "thinks" differently from us humans, which has the strength to perform many computational steps in a short time. Can we actually understand algorithms then, if they think so differently? And is that really necessary in order to use them? A precise definition of an algorithm is difficult as well, since algorithms do not exist exclusively in the digital world. Algorithms can be found everywhere and their definition must be correspondingly general. Algorithms are an unambiguous set of actions for achieving a specific solution to a problem. To make a generalized statement, algorithms can be described as instructions for solving various tasks.
So far, so good. But what makes them so brilliant?
Why our lives wouldn’t function without algorithms
To find algorithms, you often don't have to look far. Well-known and common algorithms are, for example, those of Google, Facebook, YouTube and Co. But the fact that algorithms can also be found in the analog world proves that they are difficult to separate from our everyday lives.
Even solving a simple mathematical task such as 1+1 can be described as an algorithm, as can recipes, user manuals or the weekly grocery run. Anyone familiar with how a simple linear search works will quickly recognize the parallels this algorithm has with searching the supermarket for the corresponding product on the shopping list.
In this context, algorithms influence our lives no more and no less than any other technology; rather, what is important is how we use them and what we make of them to achieve the best possible effect for us.
One (digital) example of the brilliance of an algorithm is the Spotify algorithm. It evaluates the mood of each individual song together with the most-listened-to songs of each user (including overlaps with other users). Thanks to the intensive training material, this is an artificial intelligence that can independently recognize matches between songs. Moreover, it even scours the Internet in search of articles or descriptions for a more precise interpretation and thus categorization of each song.
And although the Spotify algorithm can do a lot, it cannot recognize songs by their melody. Nevertheless, since the launch of Shazam, we no longer have to try to understand bits and pieces of song lyrics in order to later enter them into a search engine of our choice. Thanks to the algorithm behind Shazam, we get information about the title and artist of a song in seconds. Until now, fans of classical music had no way of obtaining this information using song lyrics. Audio fingerprinting is the solution here. Similar to a hash function, the audio signals are transformed into a digital code, which is compared with a database to find the song that matches this code.
One algorithm that is perhaps even more popular than these two examples is the Google algorithm. What started in a garage in 1998 has steadily evolved and still holds the largest market share of all search algorithms, with now over 200 search criteria and aspects according to which artificial intelligence evaluates websites. Why Artificial Intelligence? In November 2018, Google launched the open source program BERT, among other things, to improve linguistic data processing, better known as natural language processing, for Google search queries and optimize corresponding results.
Netflix also uses an algorithm, or more precisely an artificial intelligence, which is fed with training data from users in order to suggest new films or series that might interest them. It behaves similarly to the Spotify algorithm in that suggestions are not only made based on movie titles or genres, but also take into account actors, locations, or even the script. The data with which the recommendation system works is now made up of over 200 million users. Accordingly, the artificial intelligence can efficiently and accurately decide Netflix statements about suitable suggestions for users.
Behind some of these algorithms are artificial intelligences that ensure and drive the efficiency of all the corresponding operations, providing a personalized experience to all users.
The beauty of personalized content
Personalized content makes using a platform very enjoyable and easy. Discovering new music that we like directly increases the likelihood of returning to Spotify. The fact that Shazam helps us to recognize songs on the move and without further information, which we can also listen to directly through the connection to Apple Music, is another example of practical collaboration between technologies. As a result, we as users get exactly what we want at that moment. Whether it's new music that we like, the name of a song that's currently playing in the supermarket, or a new series that other users with similar tastes enjoyed.
The never-ending "What do we want to watch today?" - "Oh, I don't know" and the subsequent tedious search through various streaming portals is no longer necessary, which also saves an enormous amount of time. So there may be time for two episodes of this new series, instead of just one, because it was suggested directly and not searched for, or it was discovered by chance while browsing through the media library.
Of artificial intelligence and digital mirror images
Our lives are increasingly turning toward the digital age. What used to be the task of humans is increasingly being taken over by machines and algorithms. It is important to distinguish between an algorithm and artificial intelligence. Often, these two terms are used synonymously and inflationarily without distinguishing them properly. While algorithms, as already described, are instructions for solving a wide variety of tasks, one only speaks of an artificial intelligence when an attempt is made to emulate human decision-making structures by means of machine learning. For this purpose, an artificial intelligence can make use of further algorithms.
We are still a very long way from a future in which artificial intelligences take over most heavy-duty tasks and thus make our work easier or even replace us humans altogether. Nevertheless, the likelihood of such a scenario must be considered critically. As long as algorithms are programmed by humans, they will always harbor potential sources of error that must be taken into account. And as soon as we talk about an artificial intelligence that has been trained with training data and thus develops independently, the principle of the black box arises. The decisions of an AI can no longer be comprehended due to its independent further development. They would therefore have to be explained to us humans by the AI, since we can no longer understand the reasons for these decisions. Conversely, does this mean that humans can be overtaken by machines?
We already wrote an article about bias in recruiting, addressing the challenge of artificial intelligence.
Pedro Domingos, a professor of computer science and engineering at the University of Washington, has been talking about a master algorithm that uses our digital footprint to create a perfect digital mirror image of ourselves and that could take care of all our online activities for us in the future since 2015. Sending an algorithm to find information for us on the Internet is faster than going to a library and searching through all the encyclopedias for an answer. So it's not a question of man versus machine. It's about "man with machine" versus "man without machine". We are often faster when we work together with algorithms.
This could save even more time in the job search than is currently the case in the market. A master algorithm would have an exact model of each person and each job ad and could thus find the perfect job for each person. A concept that we at matched.io are more than familiar with, as we too have been focusing on the perfect match between developer and company since day 1.
What are the limits of digital algorithms?
If you always listen to the same three pop artists on Spotify, the algorithm will also suggest similar songs and artists. This is good, because you can be sure to hear new music that you like. On the other hand, however, it can be difficult to think outside the box under these conditions. So in a way, you're moving in a bubble with contents that constantly have a similarity to each other. At the same time, and not only on Spotify, famous artists are rewarded by having their music or posts suggested or displayed to a larger audience thanks to the algorithms. This makes it difficult for new, smaller artists with a smaller reach to enter the market and keeps users stuck in this bubble. It's a similar story with Netflix. If similar films are suggested to those you have already seen, there is often a lack of variety coming from films outside the bubble. Of course, users can still watch or listen to media from other genres, as just because the suggestions are among the most listened to or watched media, there are enough playlists and rankings that suggest other movies as well. It is then up to the users themselves to actively choose these alternatives and thus burst the bubble or at least make it bigger.
The broad topic of algorithms and artificial intelligence is a playing field with countless variables, formulas and networks. The more subtly algorithms work, the more efficiently a service can function. It should not be disregarded where their limits lie, but also their opportunities. They often make everyday life easier (who goes out to do the week's shopping without a shopping list?) and researching a wide variety of topics (thanks to search engines) is much faster and more efficient. Topics that really interest us and that really suit us are given priority over those with which we have had little or no contact. Nevertheless, it is always important that such algorithms find a balance between interest in old topics and enthusiasm for new things. Suggestions based on one's own interests and those of other users are becoming more and more accurate, but this process is certainly not a guarantee for good suggestions.