**2. Concept of organization digital twin**

At this article the organization digital twin (ODT) refers to the mathematical environment that simulate organization human capital productivity. To be able to simulate the reality the digital twin must meet following requirements:


#### **Figure 1.**

*Concept of organization digital twin.*

*Deep Learning Applications*

result, not the data of the past.

Bayesian-stochastic-nonsymmetric-signaling game.

one solution in solving leadership challenge.

mance rather than offering simplified solutions.

best value and reward. When RL is utilized at model-driven AI the model simulates the behavior's effect in the value. The agent tries to learn the best behavior by following the model's reward signals. Thus, the behavior of the agent determines the

Game Theory is a branch of mathematics that are used to model the strategic interaction between different players on a context of predefined environment. At management game theory there is predefined organization environment where the players are leaders and team members as workers or employees. Each player has incentives that drives their behavior in the game. Management game is nonsymmetric because leader has specific and non-changeable characteristic compared to workers. Workers are motivated in maintaining and improving their work performance and personal self-esteem. Team leaders are motivated in maintaining and improving team performance, which is related to team leader personal profit incentives. Team leader knows that team performance is essential for achieving team profit targets. Workers know that their personal incentives will improve if their work performance is good. Thus, if there are problems at work the rational policy would be to tell the problems to supervisor so that problems can be solved. In addition, solving problems may improve workers self-esteem, having hidden psychological incentive. This organization environment form state space for strategic-

Nash equilibrium is a concept of game theory where optimal outcome is the balance where all players incentives are considered and fulfilled in optimal way. If team leader gives positive feedback for raising the possible problems, it will have positive effect on workers' self-esteem, fostering workers policy to inform the problems by signals. Solving the problems will improve group performance which foster leader's policy to encourage workers signaling game. This way workers and supervisor may find equilibrium of policies (strategies) which lead to general-sum game where optimal and sustainable team profit performance is achieved. However, this article explains why this optimal equilibrium is difficult to achieve in reality. Bersin [1] study reveal that 89% of managers think that leadership is important issue, but current leadership programs bring only minor value in improving leadership quality. This article argues that modern reinforcement learning artificial intelligence gives

In addition to administrative role, the HR management has important function on adding competitive business value to an organization management (for example see references [2–6]). Managers need predictive measurements that indicate how business is developing and how to improve it. Human assets are essential for creating competitive advantages, thus interest in performance management has increased. Fleetwood and Hesketh [7] argue that researchers should better understand the complexity of the organization environment and seek to open a "black box" of causal relationships between human resources and organizational perfor-

Several studies indicate that employee psychological well-being has tendency to predict business value of an organization (for example see references [2, 8]). However, management can be confused of how to improve well-being and how much effort should be invested in well-being development at different situations to gain sufficient payback. Research reveals that organizations expect artificial Intelligence to help reducing managerial biases related to human issues and to improve productivity and employee experience [9]. Beside the hopes, researchers are also concerned that artificial intelligence may cause serious harm if the organization context is oversimplified by using data driven machine learning algorithms [10]. This article argues that AI can help solving difficult management problems

**38**

#### *Deep Learning Applications*

Markov property means that the future is not determined by the past data, thus supervised learning regression analytics cannot be solely applied in creating ODT. Markov rule is one backbone for creating ODT digital twin and for utilizing Reinforcement Learning where the behavior of the agents determines the future.

The state transition from state to state follows Markov chain where all necessary information is transferred from past to the present. Therefore, the probability of transition from the current state to the next state depends only on the current data and the activity of the players. In the digital twin, this current data must be able to determine the reality presented by the twin. The data in the twin can be measured and verified from reality, thus creating a feedback loop from the real world. This model verification against reality is also necessary for learning purposes so that ODT can learn to refine the transition functions to match the real world (**Figure 1**).
