Blog Archives

Johtamisen tekoälyllä kilpailuetua

Seuraavassa avaan lyhyesti tekoälyn toimintaperiaatteita ja selitän, miksi johtamisen tekoäly luo merkittävän kilpailuedun.

Tekoälyn soveltaminen johtamisessa mahdollistaa viisaampien päätöksien tekemisen ja ennakoivan johtamisen. Kyse ei ole pelkästään ongelmien välttämisestä, vaan ennen kaikkea ihmisten voimavarojen paremmasta hyödyntämisestä työntekijöiden ja organisaation parhaaksi.

Tekoäly (artificial intelligence) voidaan jakaa kahteen yläluokkaan ja kolmeen osa-alueeseen. Tekoälyn kaksi yläluokkaa ovat data- ja malliohjattu tekoäly. Dataohjatussa tekoälyssä käytetään algoritmeja datan syy-seurausvaikutusten tulkintaan sekä ennustamiseen. Malliohjatussa tekoälyssä tavoitteena on kontekstin syvällisen ymmärryksen (mallin) avulla muuttaa käyttäytymistä, jotta voidaan saavuttaa parempi tulevaisuus. Malliohjattu tekoäly perustuu vahvistetun oppimisen algoritmeihin (reinforcement learning). Dataohjatussa tekoälyssä käytetään ei-ohjattua (unsupervised) ja ohjattua oppimista (supervised learning).

Tekoalyn_periaate

Ei-ohjattu oppiminen (unsupervised learning) hyödyntää luokittelematonta dataa. Algoritmien avulla voidaan tunnistaa datassa olevia asioita ja indikaatioita, jotka liittyvät johonkin ilmiöön. Datasta voidaan selvittää, mitkä tapahtumat liittyvät todennäköisesti esimerkiksi sairauspoissaolojen kasvuun, työkyvyttömyyteen tai ei-toivottuun irtisanoutumiseen. Sen avulla saadaan suosituksia; esimerkiksi urakehityksessä tiettyyn koulutukseen hakeutuneille suositellaan jatkokursseja, jotka on koettu hyödyksi. On selvää, että organisaatiokontekstin ymmärrys on tärkeää, sillä ei-ohjattu oppiminen ei itsessään ymmärrä asiayhteyksiä ja syy-seuraussuhteiden kausaalisuutta.

Ohjattu oppiminen (supervised learning) on datan hyödyntämistä tapauksissa, joissa tieto on luokiteltua ja lopputulos tiedetään. Analysoinnissa käytetään regressiotarkasteluja, jotka ovat monelle tuttuja mm. Excelin kautta. Ohjattu oppiminen hakee datapisteitä noudattavan funktion, jonka avulla voidaan ennustaa tulevaa kehitystä. Esimerkiksi työelämän laadun (QWL) ja sairauspoissaolojen välillä on havaittu yhteys: työelämän laadun huonontuessa sairauspoissaolot lisääntyvät. Voidaan siis hakea funktio, joka mallintaa työelämän laadun yhteyttä sairauspoissaoloihin ja siten ennustaa sairauspoissaolojen kehitystä mittaamalla työelämän laatua.

Vahvistettu oppiminen (Reinforcement learning) opastaa käyttäytymään siten, että lopputulos on parempi. Se siis vahvistaa käyttäytymistä, jonka avulla saadaan optimaalinen lopputulos pitkällä aikavälillä. Vahvistettu oppiminen vaatii luotettavan mallin, joka simuloi organisaation toimintaa ja johtamisen vaikutusta siihen. Mallia voidaan ”pyörittää” eteenpäin, jolloin nähdään, miten johtamiskäyttäytyminen vaatii tiettyjä uhrauksia (aikaa ja kuluja), mutta järkevästi toteutettu johtaminen tuo tulosta myöhemmin ja tämä tulos maksaa moninkertaisesti takaisin (vrt. ROI eli return on investment). Vahvistetun oppisen tekoäly siis mallintaa kontekstin ROI vaihtoehtoja, vahvistaen sellaista käyttäytymistä (strategiaa), joka johtaa maksimaaliseen takaisinmaksuun eli tuottoon pitemmällä aikavälillä.

Kaikki tekoälyn osa-alueet voidaan valjastaa johtamisen avuksi esimerkiksi seuraavasti: ei-ohjattu tekoäly seuraa datavirtaa ja indikoi, että jossain ryhmässä on henkilöstön suorituskyky heikentynyt. Pulssityyppinen henkilöstökysely käynnistyy tällöin automaattisesti ja mittaa työntekijöiden kokeman työelämän laadun. Tiedot menevät tekoälyavusteiseen simulaatioon, jossa esimies voi tekoälyltä kysyä, mitä johtamisaktiviteetteja kannattaa toteuttaa, jotta saadaan paras takaisinmaksu ROI tuottona. Ennakoiva johtaminen parantaa työyhteisön työelämän laatua, jolloin tuottavuus paranee ja uhkaavat sairauspoissaolot vältetään.

Mitä tämä tarkoittaa yrityksen kilpailukyvyssä? Yksi tuottavuuden suurimpia haasteita on esimiestoiminnan huono laatu ja suuri hajonta. Vain erittäin harva organisaatio on kyennyt ratkaisemaan tämän ongelman. Yritys, joka käyttää tekoälyä johtamisen apuna, kykenee nostamaan esimiestoiminnan laatua merkittävästi. Työelämän laadun parantuminen tuotantotekijänä nostaa tehollista työaikaa kuormittamatta työntekijöitä. Yrityksen tehollisen työtunnin kustannus voi näin olla yli 20% alhaisempi kuin kilpailijalla, vaikka yritys maksaa henkilöstölleen parempaa palkkaa (vrt. Time-Driven-Activity-Based-Cost, BSC, Kaplan). Lisäksi tyytyväisemmät asiakkaat tuovat kilpailuetua, sillä henkilöstön kokema työelämän laatu parantaa asiakaskokemusta.

Rohkaisen johtajia käynnistämään tekoälyn “evoluution” omassa organisaatiossa. Dataa pitää kerätä ja ymmärtää aiempaa paremmin. Ihan ensimmäiseksi tekoälyn hyödyntäminen vaatii tiedolla-johtamisen tason nostoa ja seuraavaksi rohkeutta lähteä kokeilemaan ja oppimaan.  

Lisätietoa marko.kesti(at)ulapland.fi

Game theory approach to human capital management

Game theory provides mental models to integrate behavioral capital with finance and HR data. In this blog post, I will illustrate what management game theory is and why it is so powerful in connection with artificial intelligence (AI).

Behavioral capital is becoming increasingly important in creating business value. These “soft skills” are difficult to acquire because human social context is complex due to different personalities and biases. However, there is an emerging new science that will solve this problem and foster organizational performance. Game theory sheds light on management behavior and helps illuminate the relationship between the actions of management and the performance of subordinates. It helps illustrate why some organizations fail at change management or face high staff turnover. Game theory is science that applies mathematics to better understand human decision-making and social behavior. Game theory is key in creating new generation model-driven artificial intelligence to reinforce managers’ behavior and create sustainable competitive advantages.

Every leader, manager or supervisor is playing a game that includes the following game theory principles:

  1. Strategic game
  2. Bayesian game
  3. Stochastic game
  4. Non-symmetric game
  5. Signaling game
  6. Non-co-operative or co-operative game
  7. Zero-sum or general sum game

Strategic game: A leader’s behavior today affects an organization’s profits after twelve months. This phenomenon of long-term effects makes leadership strategic. Every leader has a certain management mindset or policy that he or she follows, either consciously or subconsciously. There are also human biases that dictate leadership behavior. In addition, there are personal assumptions about the rewards of leadership behavior. Some leaders are able to predict future rewards while others think only about fast rewards or avoiding possible punishment, which may be strategically unwise.

Bayesian game: The management game is Bayesian, meaning we have to make decisions with imperfect information. Managers’ have prior assumptions about their leadership behavior’s effects. With experience the prior assumptions may change as learning from doing gives better understanding about the context and behavior’s causalities. This is called reinforcement learning that rational persons naturally have, and it is also included at Bayesian game theory. Leaders operate at organization environment that is complex and may sometimes be hard to comprehend. However, leaders know certain probability distributions upon which they can base their decisions. Rational leaders utilize the brain’s natural phenomenon of reinforcement learning despite the imperfect information from complex environment.

Stochastic game: The management world is stochastic, which usually leads to negative surprises. One cannot expect that each day’s activities will be fulfilled as planned. Often there are stochastic interventions which require our attention. Also, humans are heavily affected by current moment bias in which short-term reward (or avoiding immediate punishment) is valued more than long-term reward (which would require different actions and more strategic thinking). The stochastic world is evolving and manager behavior will have a great effect on the outcome.

Non-symmetric game: Leaders, managers and supervisors are all in non-symmetric positions compared to their subordinates. The leadership power of managers is controlled by management systems. Thus, managers usually have different strategies than their subordinates. While worker focus on doing their tasks, managers have to think about whole team collaboration and performance. In addition, there are myriad personal and social features that form the way leaders use their non-symmetric power to influence subordinates.

Signaling game: The behavior of the leader can modify a team’s culture and culture dictates the signaling game. When there is common trust that problems are solved in a positive way, there will be more signals about possible problems and development needs. In addition to staff comments and feelings, there are also signals from management systems. For example, a digital leaderboard can signal increased sickness risk and recommend activating early intervention for preventing absences. In this case, the digital system analyzes data (i.e. staff inquiries and other data) and sends out alarms and offers advice for action.

Non-cooperative or cooperative game: In the famous prisoner’s dilemma there are two prisoners who can’t communicate but are forced to choose to either cooperate with each other or act in their own best interest. In the context of an organization, communication is not restricted, but the same type of social decision dilemma is present in every organization and team: Do the employees choose to cooperate, or do they act on their personal best interest? A non-cooperative mindset reduces productivity and may cause severe performance problems. Using game theory, it is possible to foster a more cooperative mindset in which employees innovate and solve problems in a positive atmosphere.

Zero-sum or general sum game: In zero-sum game, a gain for one player results in a loss for the other player. A zero-sum game mindset is harmful for organizations because it prevents cooperation. In the general sum game, the players help each other to achieve rewards and higher performance levels. General sum players are also willing to take risks together. They are focused on winning and are ready to make sacrifices to secure long-term rewards. Both, the zero-sum and general sum game leave marks at organization data, thus it is possible to analyze which type of game the organization plays. Data-driven AI helps identify exiting culture and model-driven AI helps with teaching which behaviors lead to the general sum game.

Why is game theory combined with machine learning an incredible breakthrough? Because we can model complex human behavior in an organizational context using mathematical modeling. First, we have to make digital presentations for organizations that model the effect of management decisions on fiscal and human performance, then we must implement Markov sequences at this digital twin and start running reinforcement Q-learning with the Bellman function. This may sound complicated (and it actually is), but when the digital twin world is built, it explains why human capital management within organizations is so difficult. AI can provide advice for managers in their decision-making, forming sort of a crystal ball that reveals future alternatives and reinforce behavior for winning.

There are significant benefits in utilizing game theory, data-analytics and machine learning in an organizational setting. I believe there is going to be an emerging new management science in this field. Game theory provides mental models to integrate behavioral capital with finance and HR data. This is not an easy task, but this revolutionary research has already begun. At this stage, it is essential to have close research collaboration with companies where data is created every day and performance problems are difficult to solve. It seems that Kurt Lewin’s saying, “Nothing is more practical than good theory,” applies here as well.

Marko Kesti
M.Sc., Dr., Adjunct Professor (human capital productivity)
Research Director, University of Lapland
markokesti.wordpress.com
https://www.linkedin.com/in/markokesti/

HR-AI helps solving wicked management problems

Companies are reinventing the performance management in their organizations (Bersin 2018). The HR-AI helps achieving this aim.

 

Traditional business management makes decisions with simplified iteration and using mental shortcuts called cognitive biases. Cognitive biases are assumptions of how the world works. Humans substitute complex issues with biases. Human performance management is too difficult to make sense of as there are just too many ifs involved: if a key person leaves, if strategy implementation fails, if customer satisfaction drops, if employee performance declines, if absence increases, etc. Therefore, cognitive biases drive leadership behaviors. However, what happens if these cognitive biases are wrong and/or harmful?

New management game theory and artificial intelligence (AI) algorithms make it possible to predict leadership behavior’s effect on business. The architecture consists of a human capital production function, motivation theories, and several evidence-based rules. For AI, management decision-making is a prediction problem, and solving it is possible through the use of an augmented reality simulation game. The simulation game predicts the future outcome according to management behaviors. Managers will learn to make better decisions from the simulation. Artificial intelligence (AI) will help to optimize human resource management decisions.

Artificial intelligence plays several rounds of simulations in milliseconds, and remembers the most valuable management practices for long-term success. AI also suggests actions to manage the decision-making process. A manager uses human judgment, because some of the the AI-suggested actions may not be reasonable in a real-life situation. Humans are good at estimating which actions are best for a specific situation, but humans are poor at predictions. Humans have several cognitive biases, which are based on wrong assumptions, and that harm long-term success. While AI can see into the future and can predict the long-term result, it does not take into consideration all situational realities. Thus, the best results are achieved when AI and human beings work in collaboration.

Human resources management AI is an intelligent prediction machine. Its prediction accuracy can be increased for each specific organization. AI has the ability to learn, and this learning is not limited by harmful biases. Prediction accuracy improves with more up-to-date data, listening to employee feedback continuously, and comparing the simulation prediction to the real-life realization.

Management_AI_architecture_dark

Figure 1. HR-AI architecture

One problem is the cognitive illusion that management competence is in order, and performance problems are due to other reasons (plenty of ifs). Supervisors’ leadership practice skills may be very poor and, therefore, there may be a tendency to neglect necessary leadership activities. The team-leader may justify omitting performing HR-practices, because it seems to be more important to use precious work time to maximize profits than to invest time into soft, human leadership practices. However, this is a wrong assumption. Management problems are serious, because behavioral cognitive biases are difficult to overcome and require practice-based learning to substitute these biases with better behavior. AI-based simulation learning may solve this problem.

 

REFERENCES

Bersin, J. (2018). HR technology disruptions for 2018: Productivity, design, and intelligence reign. New York, NY: Deloitte. Retrieved from http://marketing.bersin.com /rs/976-LMP-699/images/ HRTechDisruptions2018-Report-100517.pdf

Agrawal, A. (2018). The economics of artificial intelligence. Commentary McKinsey, April 2018.

Kahneman, D. (2012). Thinking, fast and slow. Location: Penguin Books.

Kesti, M. (2018).  Architecture of Management Game for Reinforced Deep Learning, Intelligent Systems Conference 2018 6-7 September 2018 | London, UK. (conference paper, not yet published)