
Artificial intelligence (AI) is no science fiction anymore but has long been an integral part of our everyday life. It is contained in countless applications and tools that make our lives easier in both our private and professional environments. The majority of the assistive systems that we use every day belong to the so-called weak AI. Such systems usually work rule-based and are limited to a clearly defined area of application. They are not able to think outside the box. The latter is only possible if AI is used, which is based on a modern understanding of Machine Learning.
So-called Deep Learning algorithms are based on artificial neural networks whose structure and function is modelled on the neurons of the human brain. Due to a large information base and the structure of the neural networks, deep learning systems can independently link to new content. They are also able to create analogies without being confronted with certain key words in the context of a question. You know comparable situations from your everyday communication with other people. For example, in the case of the statement “This is the sports car among telephones”, it should be clear to everyone that the word “sports car” is to be understood in this context as a synonym for “top of the range”. Accordingly, if you ask a Deep Learning-based chatbot or voicebot the question “What is the London of France?”, it will correctly answer with “Paris”. In this example, the bot presents the appropriate solution without the keyword “capital” being mentioned.
The more meaningful data such systems process, the more reliable they become. Thus the “learning effect” ideally leads to a continuous Self-optimization.
But how does Deep Learning work in practice?
Suppose an insurance company wants to automate its customer dialogue and set up a modern voice portal to adequately answer customer queries. In the ideal case, there is already a sufficiently large number of documented customer queries. This data basis is used to train the voicebot on its industry-specific field of application and to train it with the logic and range of formulations used by insurance customers familiar. Additionally, data from sector-specific or general knowledge databases, for example online encyclopaedias such as Wikipedia, can be added. Basically it is possible to adapt the text collection exactly to the requirements of a company.
In order to enable the AI to link and analyze the existing data and to draw precise conclusions, the immensely large database must be made manageable. Words that are not meaningful in the application context are deleted, whereas relevant terms are reduced to the word stem. In this way, instead of various word forms (e.g. “asked”, “ask”, “asks” etc.), only the basic form of each word (in this example: “ask”) is retained, so that terms in the existing text collection can be weighted more precisely. The text or input information is then converted into vectors. Thus, distances can be calculated in vector space and words can be weighted according to their context-dependent significance.
As soon as the available database, meaning the entire text collection, has been prepared, the AI has an easy job. In our example, the voice portal would now have the task of processing incoming requests to search for patterns based on the training data. In the casually formulated statement “my car is scrap”, for example, the Deep Learning system recognizes the relevant terms “car” and “scrap” and logically assigns the query to the department responsible for motor vehicle insurance. In addition, the AI can also take over the complete identification of the concern by asking the customer specific questions to specify the damage.
With the aid of these deep learning systems, it is even possible to process terms that do not appear in the underlying data set. For example, if a customer says, “It’s about my tooth crown”, the request can be processed even if the word “tooth crown” would not be part of the text collection. Conventional bots are not capable of such feats. Modern AI systems, however, are able to assign unknown terms by means of the mentioned distance measures, as long as they have a similarity to words contained in their data set. In our example, for the recognition of the term “crown” it would be sufficient if the text collection contains a word like “dental treatment”.
One bot, numerous advantages
Unlike conventional chatbots or voicebots, it is not necessary to think through all potential dialogue paths in advance and to consider the entire formulation range of the customers. A system based on modern AI methods only requires a sufficiently representative, appropriately prepared database to be able to work completely independently. For example, by making terms unknown to it accessible without outside help.
The advantages are obvious: On the one hand, the quality and flexibility of the dialogues are improved, on the other hand, the scalability of the solution is considerably simplified. If the insurance company from our example wants to extend its voice portal to other company areas or cover a larger number of use cases, no new dialogue paths have to be worked out, the preparation of a suitable database is sufficient. The nice side effect? Not only time is saved, but also a lot of money. Furthermore, the maintenance effort for modern AI systems is comparatively low. In order to adjust the chatbot or voicebot to changed conditions or new requirements, only the data set has to be exchanged or adapted. Everything else is done by the system itself and even more: the longer it is used in a certain area of application, the more reliable it works.
IP Dynamics has the necessary know-how and the database to make deep learning algorithms even more efficient through industry-specific adaptations, for example of word dictionaries. In addition, the portfolio includes ready-made modules, for example, for concern recognition (especially in the insurance sector) as well as solutions for identification and authentication (e.g., for information on the processing status of transactions or the recording of missing data) or callback management. A further, decisive advantage: In the Contact Center of IP Dynamics all relevant customer service channels converge – telephone, e-mail, chat, messenger, social media. The combination of a modern AI system with our omnichannel platform offers considerable potential: In the complete solution, all incoming data are effortlessly captured, linked and evaluated. And the AI gets smarter and smarter and smarter and…