Featured Post

Game Analytics - Big Data And Business Intelligence(BI)

Games generate more data then an average application because of the game state machine . Terabytes  of data can be accumulated in a short pe...

Tuesday, August 30, 2016

Leveraging Artificial Intelligence(AI)/Machine Learning(MI) In Gambling Games

Artificial Intelligence(AI)/ and Machine Learning(ML) have been in existence for decades with AI applications being used in a number of different domains. Video games, in particular, have leveraged AI to assist and direct gaming characters/avatars to act in a human way and to assist the characters in their interaction with the game or other  players. Gambling style games have not leveraged AI and Machine learning techniques(to my knowledge). However, this may be changing as "traditional" gambling games have to sort out a way to reinvigorate the space and grow the market and attract a younger audience.  Online gambling potentially provides the ideal  environment for AI/ML because game play   generates a significant  amount of data that can be used to profile players, spot trends, increase revenue per player and increase engagement.

The real question is what form of AI is more appropriate for the  gambling space versus the video game space where some AI is already in use? It is common knowledge that video  games use a "form" of AI. However, the AI  approach used in video games  may or may not be appropriate  for traditional gambling games that are simpler, less "role" playing oriented and lack the  massively multi-player characteristics of some video games. Online poker may be the exception to this characterization.

What are the options in terms of AI approaches?

Machine Learning(ML) - Until recently computer programs have been  hard coded anticipating what a problem will be and  reacting  to  a fixed set of scenarios  providing recommendations or executing code that drives a process to conclusion. Traditionally, computer programs have  not been able to react to events, learn from these events and automatically change a program's behavior  based on that interaction. Machine learning changes all of this as it includes  ways to  react to  unexpected events by analyzing data patterns and making recommendations based on that analysis. This approach could be used in gambling games to   solve the problem of a player continually winning or losing a game, skewed payouts, consistent losses with not enough wins, etc.  In an ML driven environment the games math model, bonus structure and payout structure could be modified during play to improve the overall experience and to increase revenue per player.

Case Based Reasoning(CBR) -  CBR is an AI approach that matches the goal of the outcome with specific "cases" that are appropriate for a scenario. This concept could be extended into the gambling environment  for specific game types,  game "states" and player profiles.  Case based reasoning has been used for diagnostic purposes to determine a specific method to address a symptom or issue. In gaming the case could be the "characteristics" of  game play and or the history of game play of a particular player. It could also be applied to the amount of the "wager" and loses to date or in a game. In a multiplayer or head to head game the game play of the players and opponents could be adjusted over time as the game progresses. Players may also chose a "case/category" to be placed in before the start of a game.

Decision Tree   - Decision trees were some of the first AI systems constructed. A decision tree starts at a point where a problem or goal is stated and then guides an individual or program  through a process of questions and answers to help that person or machine to achieve a goal. These systems are relatively static. Consequently if a scenario has not been experienced or anticipated there is no other option. In a gaming environment  it may be difficult to anticipate all possible scenarios . However, once the decision tree becomes robust it can be an effect coaching tool to help a player execute a reasonable gaming strategy or drive a slot experience based on player profiles.

Neural Networks - Neural networks are essentially 'learning" systems that evaluate behavior  and  or activity and develop "advisory" approaches to solve problems. To be effective neural networks need a constant flow of data. Neural networks would appear to be a good approach for massively multiplayer gambling games  such as poker or perhaps blackjack where a a number of players  continually engage and react in a gaming environment create large and ever changing data sets. Slot games are fertile ground for Neural Networks because they accumulate large data sets that can be associated with a single player and groups of players.

Post Play Analysis Learning  -   Although  reacting to game play activity during a game using AI/ML is the ideal scenario, post analysis of game play data is also important. Mining data from a game data depository that contains game play data that has been accumulated over time is an excellent way to determine trends,  player habits, causes of drop off and engagement. MI in particular is a great way to to do this as MI is geared to evaluating large data sets. The analysis of this data could be used to determine better ways to monetize games, improve the "stickiness" of games.    

It may seem odd to think of   gambling games  as a domain that can take advantage of advanced computing approaches.  However, given the amount of data that is generated from gambling games it actually is a perfect forum for AI/ML.  Slot machines are actually  a form of computing device that accumulate large amounts of data making for an excellent source of data for MI/AI.  From an economic perspective  gambling, relative to other gaming domains, has a high revenue per game coefficient.  This coefficient could be improved through the use of ML/AI. The real challenge is to determine what AI approach or approach is or  appropriate for a given "game".



Kevin Flood is the CEO of Gameinlane and FitCentrix, Inc. Kevin spent time at the Machine/Learning Lab at MIT as a technology transfer fellow engaged in model based reasoning, case based reasoning and rule based systems to enhance diagnostics systems.  He created a company called AI Squared as a result of that research.   Kevin has worked for and with US land based casino operators helping them evaluate social casino and iGaming platforms for the purpose of joint ventures and acquisitions in addition to launching online gambling operations in Europe. Gameinlane is also startup "friendly" understanding the unique value new gaming companies bring to the marketplace.  Kevin frequently speaks at gaming conferences around the world providing him with a unique perspective on this very interesting business sector. Kevin can be reached at kflood@gameinlane.com  and or twitter  at @kflow1776.


No comments: