Everybody talks about Artificial Intelligence, including experts on the 15th Austrian Consultants Days. Fascinating presentations showed the possiblities brought by the new technologies, but critics also had their say and warned about possible dangers.
To understand the impending dangers and safely make use of the new possibilities, it is necessary to understand some basics. Current success stories like the success of the AlphaGo system are centered around the area of Deep Learning. This technology goes back to developments that started in 1986, but only in recent time, the broad public has been noticing them.
So why did time seemingly stand still for Deep Learning? There are several reasons going back to necessary ongoing developments, eventually making the method applicable in practice. A central aspect was the rapid increase in processor speed over the course of those 30 years, as Deep Learning requires tremendous computing powers.
Successfull application currently includes, beside beating humans in brain games, the detection of objects on images and in realtime-videos, as well as language processing, and bio-medical applications. Deep Learning has also been recommended for customer relationship management and personalized advertisement for some time now. Currently, however, reasonable amount of incorrect decisions can still occur. Therefore, I would not let such a system cultivate customer contacts without human supervision. A pilot project by Microsoft, presenting an automated system on Twitter as a promotion of their endeavors in Artificial Intelligence, ended in desaster when users “taught“ concepts like antisemitism to the system.
A central factor to prevent such problems is to tightly control which data will be used for training by such a system, and which data should not modify its decisions. Of course, it is not possible to have such control for the input that is to be classified. For example, scientists managed to provoke wrong decisions with input images that do not make sense at all for humans, but have been classified by a system with an assumed reliability of 99.6%.
It is only possible to solve such problems by gaining a detailed understanding on the reasons an Artificial Intelligence has for making certain decisions. Publications on the IEEE VIS 2017 show that a great effort has been put into motion in the field of Visual Analytics to contribute to that task. However, at the moment, state-of-the-art research uses all available capacities to support the training process and develop tools aimed at experts in Artificial Intelligence. Users from industry and services will have to rely on ready-made systems, that can do unbelievable things, but still require a certain amount of human surveillance.
Summary: Artificial Intelligence will revolutionize decision making in companies on the long term. At the moment, however, it would be careless to leave the final decisions to computers. Systems which make decisions of Artificial Intelligence more transparent are desparately needed to create a seamless transition from current practice to future concepts.
Gathering customer data is traditionally considered courteous for SMEs. A host, greeting his regular customers with a handshake is equally favored to a financial service provider who knows enough about the customers to make smalltalk about the family. As a matter of fact, humans are naturally equipped with a top-notch face recognition algorithm and a highly intuitive database for personal data: the brain.
Advancements in IT forces entrepreneurs into a balancing act. On the one hand, they are expected to keep this kind of courtesy up and transfer it to digital communication. On the other hand, many customers have become highly sensitive regarding their personal data. Moreover, many information system providers enforce certain kinds of data to be gathered – mostly for technical reasons, but in practice those are often not set in stone but derivate from a lack of communication between system users and system providers. They also stem from a low priority decision makers put into the flexibilization of those specifictations. Superficially, it seems to be easier to just request data from the customers, but is this really true?
Gathering customer data with an information system is more systematic than in a typical conversation between humans. Moreover, data are typically stored on a long-term basis. However, what really makes data gathering uncanny for humans is the lack of transparence about their application. As a matter of fact, vast data gatherers like Facebook or Google are theoretically capable of learning a lot about their customers, including things they do not want to disclose. A typical SME, however, rarely possesses a sufficient amount of data for comparison. By the way, gathering customer data is not limited to data manually entered into forms. Recognizing customers with technologies like Cookies is principally the same as recognizing the face of a customer, as done manually by the entrepreneur himself.
The Cookie technology is also a gut example for the ambition of politics to regulate the topic. Unfortunately, the result is satisfactory neither for enterprises, nor for customers. Enterprises are confronted with additional technological effort to provide the so-called “Cookie-Banner”. Customers typically are still not able to dissent from the use of this technology.
It is a more sensible approach for SMEs to provide their customers with a reason for all information that is gathered, optimally by showing the advantags for both sides. For example, a small trade business or logistics company can ask for their customers' phone numbers, but at the same time provide status messages on their phones. In many cases, it is useful to make data gathering optional. If a customer wants to leave a field empty, but then is prevent from submitting a form, he is likely to enter something false. The resulting effect is called “Dirty Data”. Expertens for data analysis consider empty fields to be much easier to handle than fields with intentionally false data. If customers are aware of the fact that by leaving empty the field for the phone number, they relinquish their status messages, they are often much more inclined to enter their correct number, compared to customers the enterprise forces to enter something.
As stated earlier, empty fields seem to create technical problems. It is true that older database systems have strict regulations which data are necessary to clearly separate datasets. In practice, however, it is often possible to design those factors of identification in a flexible way, using the information customers provide willingly. In situations when two customers provided identical information, the stage of the business process typicall is one where unique customer identification does not have true benefit. For example, a service provider comparing various kinds of wares typicall wants to know which kinds of products are popular in certain groups of wares, but not individual customers.
Conclusion: SMEs have to balance out the interests of their customers in individual solutions with protecting their privacy. Modern information systems provide methods to create customer datasets with partial information, letting certain fields empty. Providing additional benefit often encourages customers to provide more information of their own free will.
Addendum, November 21st, 2017: The upcoming legal validity of the EU General Data Protection Regulation as of May 25th, 2018 causes some changes: some things that have been “best practice” till now will become legally binding. More details will follow here in the near future.