A shiny, minimalistic designed office building, surrounded by nature, including green roof, invites to enter. It could be today, or ten years in the future. As the whole company is supported by strong Artificial intelligence (AI), so is the building, as it opens the door after a short face scan confirms our appointment.
Speaking about AI in corporations, it must be understood that we are talking about an intelligent system connecting all parts of the organization. There are no single AIs managing buildings, robots, or supporting employees for individual tasks, but it is one holistic algorithm located in the Cloud. Parts of it can function temporarily separated in the Fog, as continuous interconnection may not be feasible or practical. Up to that, small parts of the algorithm can work temporarily autonomously at the Edge, for example, implemented directly in safety cameras or drones. This is to ensure that such tools would work in emergency situations, disconnected from the main algorithm.
Today’s individuals actively use online information for their decision-making, including sources like Wikipedia or, more and more, also Generative AI chatbots like ChatGPT or Bard. In opposition to early Digital Transformation of companies, Generative AI works bottom-up, as employees bring their apps to work (could be on a computer or smartphone) and companies must react to this. A Generative AI guideline helps to define a safe space, for example, defining what kind of information can be shared with public Gen AI tools, and which only can be used with a limited number of vetted Gen AI company apps.
A technology that can be used efficiently by companies and other organizations. Not only that algorithms and robots can replace employees in the factory and office, but management levels are also in scope. Due to their high salaries, the efficient automation of CEOs would have an immense cost-benefit. However as algorithms cannot be held legally responsible, this is not an option. Nevertheless, middle managers are controlled by higher-level management, so part of their functions can be automated, especially if their responsibility is limited, as for example, the management of a warehouse.
An efficient AI is not implemented in a single point, but holistically in the whole organization. It can be part of physical robots, available as a chatbot to explain rules and regulations, and be part of the board to consult about company investments and strategy. The AI includes expert knowledge from external databases, the internet, and also from observed decision-making by the company’s employees. Due to the infused expert knowledge, the AI always represents one pair of eyes inside the four-eyes principle. The algorithm is the first required to decide, before a human must confirm the correctness of the decision. If the human and AI do not align, upper management must approve. For lower value routine decisions, the AI can even act autonomously, like ordering office supplies or approving smaller payments.
As Generation Z uses chatbots to directly receive the required information instead of searching for it inside documents, this is a service they expect also from their employer. Already the first companies offer a chatbot replying to such questions based on internal guidelines. A potential problem is that regulations may not be sufficiently precisely worded, giving space for different interpretations. Unprecise training data (here the guideline) can cause the chatbot to hallucinate, meaning to create credible but wrong information.
Keeping guidelines simple should reduce the grey areas as much as possible, giving clear guidance on what is expected from the employee. This is beneficial for the reader, independent of whether it is a human or a chatbot. Besides the pure definition of the topic, guidelines may include a motivational part, explaining the importance of the topic and potential impacts on organization and society. Such additional information may confuse the algorithm and lead to a higher risk of hallucination. Companies offering a chatbot to explain guidelines and processes have to adapt the way they create such documents. Regulations may include hashtags and labeling inside the different chapters or even include a simplified description of the regulation targeted to the algorithm.
The risk of hallucination can be reduced, but not eliminated. Responsible for the information given by the Gen AI bot is the human. Accordingly, the explaining and interpretation of guidelines may be automated, but organizations still want to keep the human in the loop, as for example, the responsible human officer may have to release the chatbot’s answer.
The AI is not limited to communicating regulations, but it goes one step further. It analyzes the regulations to identify potential weak points and loopholes. It can align the numerous internal documents to find missing steps or regulations which contradict each other or external laws. With this, the AI collaborates with the various guideline owners.
The chatbot is the interface to the AI and can act tailor-made for the asking employee. If the bot is connected to the HR database, it understands who is asking and can consider this information for its answer. For example, the allowance of gifts and hospitality may be different for an IT employee than for a colleague working in the procurement department. For a sales employee, the AI is aware of the attended customers and their risks, for example, if they are governmental owned or if there are indications of potential corruption. The chatbot can take different forms, for example, it can act based on the company founders’ biography, including outer appearance and voice.
To support the organization and AI governance system, the HR department may have a Digital Twin for each employee, including their basic information, curriculum, character, qualifications, and certifications. Depending on the job profile, information like performance (results, but also observations via sensors like cameras) or mood (via sentiment analysis using emails, social media, or voice calls) can be added. This data is the basis for the algorithm to understand how to explain internal regulations to the individual employee. Furthermore, the data gets used to select employees to create the ideal project teams, suggest employees for training and career moves. The AI analyzes the employee’s opportunities, but also risks, as for example, vulnerabilities to corruption or other compliance topics. Here the organization can proactively support with additional training. The digital twin includes a 3D avatar, used for company meetings in a virtual reality platform, up to the Industrial Metaverse, to interact with other digital twins. Using the avatar instead of the laptop’s camera supports to take off stress, as employees do not have to focus on how they get perceived by the other participants of the meeting.
This prediction is based on already existing technology. Sophisticated sensors and algorithms can shift the work environment from motivation towards continuous control. It is up to society to decide if and how such possibilities should be allowed, controlled, and limited.
Comments on this publication