The impact of Artificial Intelligence on business is growing rapidly. Artificial Intelligence is becoming increasingly intelligent; machines interact with each other more easily and are able to accurately process more and more unstructured data. With new technologies, inevitably different views and opinions arise. While some experts see Artificial Intelligence mainly as a threat, other experts acknowlegde the huge potential for business to support employees in their day-to-day jobs.
In our blog What is Artificial Intelligence? we posed a clear definition of what Artificial Intelligence is, and showed its relation to Machine Learning. Now, we will review both the advantages and threats in Artificial Intelligence and see to cope with them.
You might ask yourself: “Why should I apply Artificial Intelligence in my organization in the first place?” A justified question in our opinion. First of all, the interaction between humans and Artificial Intelligence can make a very effective team in many business processes. However, applying Artificial Intelligence should never be a goal on itself, rather a means to achieve business goals. Below, we have listed the three main advantages of this partnership between humans and computers in Data Science.
ADVANTAGE #1: EFFICIENCY
Let’s face it, computers are faster. Therefore, from a business point of view, it is interesting to assign tedious and well-specified tasks such as certain types of administrative work and straightforward calculations to a specialized Artificial Intelligence algorithm. Not only does it reduce the costs of these processes significantly, but it also frees up employees’ time to focus on more complex, creative and emotional problem solving. For example, we see that many customer service centers are operating much more effectively towards a positive customer experience when they have Artificial Intelligence applications to support them to focus on the most important matters. Besides, their working days are more interesting because they can now focus on typical human related tasks.
For human beings it is hard, or even impossible, to be completely objective about the world around us. It is well known in psychology that humans are masters in finding (not necessarily existing) patterns that confirm our beliefs. We call this confirmation bias. The human brain is focused on our perception of truth. On the other hand, Artificial Intelligence is by definition objective. It doesn’t form an opinion before having analyzed a sufficient amount of data. In contrary to a human being, an algorithm does not make any presumptions on what the expected outcome of an analysis should be. Therefore, Artificial Intelligence is much better suited for tasks requiring objectivity. For example optimizing conversion rates for potential customers by determining the best personal discount at the right time.
ADVANTAGE #3: SCALABILITY
Humans are able to oversee things better if they omit the (seemingly) unimportant details. However, sometimes details are essential to really comprehend a certain situation. One of the advantages of an Artificial Intelligence algorithm is that it can take the smallest details (e.g. all individual customer journeys) into account and translate this to the bigger picture. We especially see these advantages in organizations where commercial and logistic actions have a strong interdependency. A good example is the challenge of managing the on-shelf availability (logistics) when products are being highly promoted (commercial) in the market. Here, such a holistic Artificial Intelligence approach typically leads to a improved business strategy and better results for both customer (better on-shelf availability) and the organization (more sales).
The three advantages described above combined give rise to better insights and decision making. Artificial Intelligence supersedes human intelligence by processing a vast amount of data and find potentially undiscovered patterns or relationships on a corporate level. At the same time it enables human employees to work on a creative and emotional level which is better suited for the human skill set.
Although Artificial Intelligence has proven its worth, it is far from the point where it can substitute a human brain (if it ever gets there at all). As I explained earlier, humans also bring something to the table in the partnership between human and computer. Below we discuss some threats that could arise when Artificial Intelligence is not applied correctly in your business.
THREAT #1: BLACK BOX
When it is not clear in detail how an algorithm exactly came to an end result, we speak of such an algorithm as a black box. Some magic happens inside but we do not know exactly how the magic works. My experience is that most end users of Artificial Intelligence applications are more interested in the results rather than how the magic inside works (and that is fine actually, not everyone has to have the passion of a data scientist to understand such complexity). But how do you know if it works? How do you know that the output is correct? Unfortunately, the answer is that you often don’t. Some Artificial Intelligence algorithms are so complex in their model structure that their own creators cannot entirely comprehend how it came to a certain output. This could become a problem when Artificial Intelligence supports you in taking the right business actions or your doctor in prescribing the best treatment trajectory for your illness. It is therefore of uttermost importance that Artificial Intelligence is well tested out-of-sample and thoroughly finetuned to the desired business application.
THREAT #2: GOAL OPTIMIZATION
The second threat of Artificial Intelligence lies in the fact that it mainly tries to do one thing: optimize its goal (for which it is explicitly programmed). This goal could be anything you want: maximize market share, revenue or profit. Humans are very good in intuitively taking extra information into account. Computers are not. An example can be found in an algorithm that focuses on optimizing market share. It is likely to set all prices to zero because that action will increase market share the most. However, no sensible human being would ever do that because it will generate zero revenue and negative margin. Although this is an extreme example, such undesired strategies could follow from a poorly implemented Artificial Intelligence algorithm. We therefore have to explicitly or implicitly program the Artificial Intelligence algorithm such that these undesired side effects will not occur in the solutions in your organization.
THREAT #3: ALGORITHMIC BIAS
A third threat arises when the future training dataset of the algorithm depends on today’s actions of the algorithm. To better illustrate this, we will use a real-life example. Suppose an organization uses Artificial Intelligence to recommend interesting products in real-time to their (potential) online customers. The product that is recommended to the customer has a higher conversion probability because of the exposure the recommendation creates for this product. Coming night, the algorithm will be retrained with today’s fresh data. It will learn that this product generates high conversion rates, resulting in even more exposure tomorrow. Eventually this will create a bias of the algorithm and a convergence to a smaller set of products. This is the opposite of what you want your recommendation algorithm to do. Experienced data scientists know how to tackle this by either adding a healthy dose of randomness to the algorithm or by constructing hybrid techniques that do not solely need transaction data to come up with an interesting product recommendation (e.g. by using images, descriptions or reviews of products).
Artificial Intelligence applications are typically trained to do one well-specified task very well. I strongly believe that we are not even near the point where Artificial Intelligence can solve complex business problems autonomously (without humans overviewing the situation). To really utilize the advantages of Artificial Intelligence, while coping with the possible threats, we must be thoughtful about where to implement Artificial Intelligence and how to use it. This means we should keep autonomy and control over important decisions. Therefore, it’s crucial to have a team of qualified data scientists and business experts who master these skills and exactly know what they are doing when handling data and implementing Artificial Intelligence algorithms.
As long as control remains in capable human hands, Artificial Intelligence can complement us to be more efficient and more effective, creating more value and ultimately resulting in a better life for consumers and employees. Let both humans and Artificial Intelligence applications perform those tasks in which they are best. And above all, let us together exploit each other’s strengths.