- Trends & Technology
Artificial intelligence – what does it really mean and how do we use it in practice?
The buzzword artificial intelligence (AI) was everywhere this year – in specialist articles as well as on news sites and social media. It’s hard to find a technical development as hotly discussed at the moment as this one. While some – such as Space X and Tesla CEO Elon Musk – see artificial intelligence as a threat and warn against it: “Mark my words – A.I. is far more dangerous than nukes”,1 others point out how AI can be used for good. One of the most well-known proponents of AI is Facebook CEO Mark Zuckerberg. He firmly believes in it and is putting his faith and money in the fact that “the technology will save us from ourselves”2.
From simple automation to artificial general intelligence
Although the term artificial intelligence has existed since 19553, many people still struggle to grasp its meaning and only have a vague understanding of the technologies behind it. In order to better understand the technology, it is necessary to look at different levels of automation. These can be divided into four levels: script-based automation, robotic process automation, rule-based software bots and cognitive computing/artificial intelligence.
The simplest type of automation in the software industry are what we call macros. These enable the automation of recurring work steps, especially within one program. To do so, specific command chains within a single program or across a small number of programs are specified in the script and processed by the software. A typical example of this is the creation of reports in Excel. A data set can thus be processed by recording manual commands for data aggregation, cleansing, sorting, analysis and output, which are then performed automatically by a macro.
Robotic process automation
The second level of automation is formed by so-called mini bots or robotic process automation (RPA). This software can carry out repetitive tasks across one or more systems and simulate the role of a human in the operation of a computer. For this, the individual work steps of an employee are recorded and processed by the software. Among other things, RPA is used for standard processes in HR that typically require a lot of manual processing across different systems, such as for the accounting of business trips. The RPA solution can log into multiple applications, compile data from different sources and enter it into one system such as SAP, perform checks and calculations and automatically send off the output. This only works for rule-based processes and activities where all required data is available in a readable electronic format in a consistent structure.
Rule-based (software) bots
A rule-based (software) bot can react to different (user) inputs by using predefined rules to perform set actions. A well-known example of this are rule-based chatbots used for customer service on websites. The customer asks a question in a message box or via a messenger and the software can pick up keywords and react with predefined answers. These terms and the corresponding answers always have to be determined beforehand and preprogrammed. If the chatbot cannot respond to the input, because none of its predefined rules are applicable, then the message is forwarded to a human employee. The employee can then process the customer’s query and subsequently define new rules to expand the chatbot’s abilities.
This technology can be further developed with self-learning rule-based bots. In this case, the software learns from the additional input of the human employee by using it to autonomously create its own new rules, thus improving itself without additional programming. This type of self-learning software is the first level of automation that can be classified as artificial intelligence.
Cognitive computing/artificial intelligence
Cognitive computing technologies form the highest level of automation. They simulate human thought processes in software-based models. This is done by using self-learning algorithms to teach a system a certain task. Machine learning is where algorithms analyse large quantities of data and use it to identify patterns and make predictions or decisions. Thanks to cognitive computing, software can learn to carry out a wide variety of tasks. Simple examples include language recognition and processing (speech recognition & natural language processing) such as Apple’s Siri or Facebook’s facial recognition and identification function.
Software that is capable of reaching human-level intelligence in certain areas and carrying out specific tasks at least as well as a human is referred to as weak or narrow AI. Huge progress has been made in these areas in recent years. Well-known examples: The world’s best chess, Jeopardy, poker and Go players have all since been beaten by software that was trained with machine learning.
Technology that demonstrates similar intelligence to humans across many different disciplines is referred to as “artificial general intelligence”. Although companies such as IBM have developed solutions that can be used across numerous industries and for a variety of tasks, we are still a long way away from developing comprehensive, human-like artificial intelligence.
AI in practice: automation at innogy
At innogy Consulting, we too are occupied with the topic of artificial intelligence and support our customers in identifying, strategically evaluating, planning and implementing various technologies. That includes chatbots as well as more recently in the HR department. My colleague Daniel Sochaczewski works as a principal consultant at iCon and specialises in digital aspects in his projects. Among other things, this role saw him work with the start-up “Precire”, a company which has developed a technology for inferring a person’s personality traits from the way they talk and communicate. He revealed to me in an interview exactly how that works and how innogy can use this solution for the identification of suitable new employees.
David Gölz: In your last project, you worked with our client on an exciting use case in the field of artificial intelligence. How did you come across this use case?
Daniel Sochaczewski: At iCon, we always want to give our customers the most up-to-date insights into new trends. For this, it is necessary to keep our finger on the pulse at all times. To ensure we do, we always look out for exciting concepts, solutions and companies across completely different channels – including, for instance, at conferences, seminars or in the personal spheres of our consultants. In line with our self-perception as management consultants, we are always bringing the outside-in perspective into our projects and thereby offering our customers considerable added value. This also includes the comprehensive expertise found in specialised start-ups. In this way, we met the exciting start-up Precire at a networking event on artificial intelligence and instantly recognised the potential of their solution for our customers.
David Gölz: So you instantly recognised that Precire’s technology can be used to solve the concrete problems faced by our customers. What is the value proposition for our customers?
Daniel Sochaczewski: Precire’s solution supports HR in the selection of candidates. When hiring, the personality of a candidate is of considerable importance alongside his or her qualifications in the field. With the AI solution from Precire, we can objectively identify personality traits and thereby objectify the “gut feeling” of recruiters concerning the cultural fit of a candidate – something that was not previously possible.
It sounds very vague here but can be put into concrete terms in the form of hard numbers. For instance, time to hire is reduced by up to 60%, early phase fluctuation is reduced by up to 45% and process costs go down by up to 35%.
These are big numbers, especially in the current age of the “war for talents”. If you consider that with the use of artificial intelligence only around half as many recruited candidates proceed to quit their positions as was previously the case, the solution quickly pays off.
And the AI solution has yet another advantage: the (social) competence profiles generated by Precire can be used to create training material to systematically strengthen and sustainably anchor these employee competencies.
David Gölz: Using technology to measure the cultural fit of a candidate sounds very complicated. Can you give us a simplified account of how the technology works?
Daniel Sochaczewski: The Precire technology recognises learned characteristics in language from which it derives linguistic, psychological and communication-related traits. Alongside natural language processing techniques, it also records specific language patterns (word order, regular expressions etc.). Using these language patterns, objective predictive models can be generated based on reference data sets.
From speech samples of approximately 15 minutes in length, Precire is capable of not only drawing conclusions from word level, including word order and word category, it can also analyse the language right down to the level of personality. The result is an objective personality and competence profile of the person whose language was analysed. Using this method, the algorithms do of course become more precise the more speech samples are analysed.
David Gölz: How did we introduce this to the innogy Group?
Daniel Sochaczewski: The Precire technology offers its greatest advantages in the recruiting process. That’s why our team at iCon actively sought out our contacts in innogy’s HR department and gave them an initial rough sketch of the idea over the phone.
This is where the benefits of a long-term, trust-based relationship come into play: we immediately got the go-ahead to introduce the solution to the relevant people together with Precire at a half-day workshop. I then prepared the workshop in great detail together with the Head of Psychology at Precire. Everything had to be perfect, and not just the storyline, the content and presentation. We wanted to make the solution’s advantages tangible to a certain extent. For this reason, we used two innogy job advertisements that were current at the time to draw up desired profiles for the candidates being sought in the job ads. In the workshop itself, we then analysed speech recordings live and compared them with the desired profiles. The results were there for the customers to see and experience in real time. They could see how the dashboard and results were presented. It made the workshop an all-round success.
David Gölz: Where do we go from here?
Daniel Sochaczewski: To implement the solution, “genuine” desired profiles for job advertisements have to be created. The scientific basis, which Precire already has the framework for, must be fine-tuned according to the specifics of the industry and company. We then take speech samples from employees that are already working at innogy in the positions being sought. The pilot project can then get underway and we are excited to see the results.