At the very beginning of this course, we briefly talked about the importance of artificial intelligence in today’s society and the society of the future, but we could not discuss it in detail at that time, since we had not yet presented enough technical concepts and methods to base the discussion on concrete concepts. Now that we have a better understanding of the basic concepts of artificial intelligence, we have a much better starting point for a reasonable discussion of the implications of current artificial intelligence.
The first consequence: algorithmic bias
Artificial intelligence, and machine learning in particular, is used in many industries to make important decisions. Therefore, the notion of algorithm bias should be mentioned. This means that discrimination on the basis of ethnicity, gender or other factors is often included in the algorithms when deciding on job applications, bank loans, etc. Algorithmic bias is not a hypothetical threat invented by academic researchers. It is a real phenomenon that already affects people today.
Online advertising
It has been observed that online advertisers such as Google tend to show ads for lower paying jobs to female users compared to male users. Also, searching for an African-American-sounding name may show you an ad for a criminal records access tool that you probably wouldn’t have seen otherwise.
Social networks
Because social networks essentially recommend content based on other users’ clicks, they can quickly magnify existing biases, even if they were very small to begin with. For example, LinkedIn has been observed to ask a user searching for professionals with female names if they actually meant a similar male name: so when searching for the name Andrea, the system would ask: “Did you perhaps mean: Andrew?” If people occasionally click Andrew’s profile, maybe just out of curiosity, the system will suggest Andrew even more often in subsequent searches. We could mention many other examples, but you have probably heard about them in the media. The main problem with using artificial intelligence and machine learning instead of rule-based systems is their lack of transparency. This is partly due to algorithms and the fact that the data is a trade secret that companies do not usually make available for public scrutiny. Even if they were made available, it is often difficult to determine which part of the algorithm or data elements are the cause of the discriminatory decisions.
In other words, the last point means that companies like Facebook and Google must at least explain their algorithmic decision-making processes when providing services to European users. However, it is still not clear what exactly counts as an explanation. For example, is a decision made using a nearest neighbor classifier (Chapter 4) considered an explainable decision, or would the coefficients of a logistic regression-based classifier be a better choice? But what about deep neural networks that can incorporate millions of parameters and learn from terabytes of data? There is currently intense debate about the technical implementation and interpretability of decisions based on machine learning. In any case, the General Data Protection Regulation could improve the transparency of artificial intelligence technologies.
Another consequence: Seeing is believing – or isn’t it?
We are used to believing what we see. When we see a leader on television saying that his country will launch a trade war against another country, or when a spokesperson for a well-known company announces an important business decision, we tend to trust them more than a news story someone else wrote about their statement. Similarly, when we see photographic evidence from a crime scene or a demonstration of a new technological gadget, we attach more importance to it than a written report explaining how things look. Of course, we are aware of the possibility of falsifying evidence. With programs like Photoshop, we can create photos of people in places they’ve never been and with people they’ve never met. The look of things can also be changed by simply adjusting the lighting or having a person pull in their stomach in the cheap ‘before and after’ pictures that advertise the latest diet pills.