AI Gamification for Improved Business Outcomes

Malcolm Dickson
Nov 3 · 7 min read
Michał Parzuchowski on Unsplash

“The relationship between human intelligence and artificial intelligence (HI + AI) will necessarily be one of symbiosis. The challenge and potential of exploring this co-evolutionary future is the biggest story of the next century and one in which a closeness in development velocity is a necessity.” — Bryan Johnson

Entities like DeepMind illustrate the advancements of artificial intelligence, or AI, through gameplay analysis, and then the subsequent ability for the AI to recognize winning patterns unnoticed by its human competitors. Herein lies the HI + AI development velocity illustrated.

Imagine applying AI-based gamification to desired business questions and hypothetical outcomes. For example, given a sufficiently broad set of data, a business might ask an AI system, what is the best strategy to maximize revenue? Revenue has many variables, which is why it makes a compelling initial question. The number of variables underlying a revenue stream maps well to the infinite number of possibilities in hands of Texas Hold’em.

Recently an effort was made to apply AI to Texas Hold’em by Carnegie Mellon and Facebook. The designers of the Texas Hold’em AI model simplified the game model, reducing the infinite variables in a hand down to a very few, what they thought to be critical few variables. The AI model designers observed that a lost bet impacted that player's next bet, which the AI designers labeled as a regret factor. The AI design leveraged that regret value within their AI model to win subsequent hands, if only slightly. A series of small wins resulted in an overall game win, every time.

These and countless other examples are very targeted, for good reason. AI’s weakness is generalization. Today’s AI models are targeted, built for specific situations, specific patterns.

What makes AI and neural nets good at pattern recognition?

What goes under the heading of AI today takes the underlying form of neural nets. These systems are also referred to as machine learning and when the layers that define the neural net become sufficiently deep, referred to as deep learning. These systems range from relatively simple python code to highly complex and expensive to run systems. A neural net is a codebase which when run constructs a series of nodes, each node containing some logic taking an input and generating an output value.

Neural nets vary in the number of nodes in a column of nodes and depth of rows of nodes. A node in one column gets its input from all the nodes in the preceding column and sends its output to all the nodes in the subsequent column. This network of layers of interconnected nodes resembles, by design, the network of neurons in a brain, albeit in a very small way. Note: there are somewhere around 86 billion neurons in a human brain, as a comparison of complexity.

The code within a neural net neuron is specifically designed to refine its output relative to the idea of success. The concept of success is a per model definition. Let’s say the idea of success was to distinguish images that contained a cat face (keeping this relatively simple). The neural net comprised of the array of interconnected nodes needs to be “taught” what the pattern of cat face looks like.

The process of teaching a neural net is the process of refining how each of the interconnected neurons passes their values to their connected neurons to achieve a “success” outcome. With enough training on various different images of cat faces that yield an improvement curve of success means that the neural net is forming the correct pattern of values between their connected neurons to statistically determine a cat face when next presented with a unique image.

Recognizing patterns in various degrees of complexities is the strength and core value of today’s AI and its constituent neural nets.

Applying AI/neural net pattern recognition machines to business have the ability to enable companies to see meaningful patterns in their data much faster than humans. Companies that know where and how to apply AI to identify and then adjust to changing business environments much faster than their non-AI competitors, enable those companies to further extend their market breadth.

“Companies plan to spend big money on AI — as much as $77.6 billion by the year 2022… Optimized decision-making is what AI is for.”

Develop Your Leaders for an AI-Driven Future

Today’s AI adoption constraints typically fall into two major categories. First, there is a human resource constraint. Demand for either data scientists or AI professionals who build the AI models is high with a small available resource pool. The second typical constraint is appropriately tagged data. Enterprise is rich in data, the data needs to be prepared for subsequent AI models.

Startups like DotData will help address some of these constraints. In fact, there is a growing demand for AI startups as the preferred consultant for enterprise AI efforts.

“Artificial intelligence and machine learning top everyone’s lists this year, and CIOs are twice as likely to go to startups for these solutions than to established vendors. More than eight in 10 executives, 81%, say they are investing in AI and ML, with the majority, 67% relying on startups” For Artificial Intelligence Capabilities, CIOs Prefer To Work With Startups

In order to address today’s enterprise AI adoption constraints, accessing existing enterprise data will need to be more friction-free, frictionless in an ideal world. And, the ability to leverage the wealth contained within that data needs to be frictionless, removing the human expertise barrier currently needed to build the AI neural net models and train the resulting model.

Let’s consider for a moment an abstracted model of what we think of as a “game.” In other words, let’s say we build an AI model of the abstraction of game, within which we could fit just about any implementation game. The idea of win is the desired output when the AI model is run. The result is a sufficiently generalized AI-based game model. When applied to any game rules, the AI model would analyze patterns within the rule constraints to find the pattern of moves that result in the “win” output.

Now, let’s move this AI model into the business environment. We can equate the business question into the definition of “win” and the business data against which the question is asked to the game rules. The AI engine, given a question would traverse the data with the goal to arrive at the most efficient solution to the question, the win. Inevitably, like Go and Texas Hold’em, there will result in patterns outside the human paradigm, business value patterns.

If you’ve made it this far then you’re probably realizing that AI models are specific and somewhat expensive. So, how might this scenario unfold over the next few years given the resource constraints mentioned so far? Amazon’s Alexa has a command library SDK, a skills kit, enabling others to write domain-specific commands. Alexa goes so far as to have a business-centric implementation, albeit early and limited in commands. Yet it’s quite easy to see the trend with the business that will eventually yield a vocabulary extended to a generalized business data pattern recognition command library.

It’s not hard to imagine a time in the not too distant future where a CFO don a VR headset to “play” with models of their company’s financial data and is able to ask the system AI to show data patterns that would result in an improved margin. Or, imagine a CEO wearing the VR headset looking at a dataset representing the company’s universe of opportunity, and asking for the system AI to extract the largest set of common attribute data across that universe opportunity that would yield the greatest market differentiator.

There are no technological hurdles to accomplish this scenario, only the will to succeed or win if you will.

The business questions define the data environment and the win parameter. With those two elements defined, let the games begin.


References:

  • AI beats professionals at six-player Texas Hold ’Em, https://www.google.com/amp/s/www.newscientist.com/article/2209631-ai-beats-professionals-at-six-player-texas-hold-em-poker/amp/
  • DeepMind’s AI beats world’s best Go player in latest face-off, https://www.newscientist.com/article/2132086-deepminds-ai-beats-worlds-best-go-player-in-latest-face-off/#ixzz63vrXvTfY
  • DeepMind AI now keeps up with ‘StarCraft II’ Grandmasters, https://www.engadget.com/2019/10/30/deepmind-ai-starcraft-ii-grandmaster-level/
  • But what is a Neural Network? | Deep learning, chapter 1, https://youtu.be/aircAruvnKk
  • 5 ways AI will evolve from algorithm to co-worker, https://www.techrepublic.com/article/5-ways-ai-will-evolve-from-algorithm-to-co-worker/
  • For Artificial Intelligence Capabilities, CIOs Prefer To Work With Startups, https://www-forbes-com.cdn.ampproject.org/c/s/www.forbes.com/sites/joemckendrick/2019/10/31/for-artificial-intelligence-capabilities-cios-prefer-to-work-with-startups/amp/
  • Alexa Skills Kit, https://developer.amazon.com/docs/custom-skills/built-in-intent-library.html
  • Alexa for Business, https://www.google.com/amp/s/www.techrepublic.com/google-amp/article/cheat-sheet-alexa-for-business/
  • Develop Your Leaders for an AI-Driven Future, https://www.gallup.com/workplace/267410/develop-leaders-driven-future.aspx
  • Demis Hassabis tells Jim Al-Khalili why he wants to create artificial intelligence, https://www.bbc.co.uk/sounds/play/m0009zbj

©️ 2019 Malcolm P. Dickson All rights reserved

Data Driven Investor

from confusion to clarity, not insanity

Malcolm Dickson

Written by

Augmenting AI with AGI as a Service.

Data Driven Investor

from confusion to clarity, not insanity

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade