Game On: How War Games and AI Hallucinations Can Level Up Your Strategy
Thanks to my wife Steph's recommendation, I recently had the chance to listen to Malcolm Gladwell's "Serious Games" episode of his "Revisionist History" podcast, which was especially inspiring for me given my interest in game theory.
In this episode, Gladwell explores how war games can be used to help military planners prepare for unexpected events. He notes that:
We're good at making lists of things we expect to happen, but we're not very good at making lists of things we don't expect to happen.
This can leave us vulnerable to unexpected events which can catch us off guard and leave us struggling to adapt.?Who expected COVID-19’s disruption?
Gladwell argues that war games can help military planners prepare for unexpected events by simulating different scenarios and thinking through possible outcomes. War games can reveal unexpected insights and outcomes that may not have been apparent otherwise. By exploring these scenarios in a controlled environment, military planners can develop more effective strategies for dealing with unexpected events.
It helped me realize that the concept of utilizing war games beyond the military. In business, unexpected events such as market disruptions, supply chain interruptions, or new technologies can catch companies off guard and leave them struggling to adapt. Generative AI can help businesses prepare for these unexpected events by simulating different scenarios and generating new ideas that might not have been thought of otherwise.
Shall We Play a Game?
Let's say that a consulting company has received an RFP from a potential client in the healthcare industry. The RFP outlines a project to develop a new patient management system that integrates with the client's existing electronic health record (EHR) system. The consulting company wants to develop a winning proposal that addresses the client's needs, reduces risks on all sides and stands out from the competition.?
By utilizing a large language model such as GPT-3, the consulting company can delve into various scenarios, both anticipated and unforeseen. The technology is essentially playing a game of "what-if" where the consulting team can explore different outcomes and uncover potential innovative solutions to problems they may not have even considered. Here are some examples:
Expected Scenario:
The consulting company expects that the client is primarily concerned with the technical specifications of the patient management system. They anticipate that the client wants a solution that is highly scalable, secure, and integrates seamlessly with their existing EHR system. The consulting company can use GPT-3 to generate technical documentation, such as system architecture diagrams, performance metrics, and security protocols, to include in their proposal. Some potential prompts for GPT-3 in this scenario could include:
"Generate a system architecture diagram that integrates the patient management system with the client's existing EHR system."
"Provide a list of performance metrics that demonstrate the scalability and efficiency of the proposed solution."
"Generate a set of security protocols that address potential vulnerabilities and ensure compliance with industry standards."
As Gladwell say, we’re good at these sorts of lists.
Unexpected Scenario:
Now things start to get interesting. With GPT-3 as their war game simulator, the consulting company can prompt the AI with general prompts to uncover potential solutions that may not have been obvious otherwise. The language model can simulate different scenarios and help the company explore how to adjust the solution, staffing, and provisioning to accommodate the changes. To begin anticipating the unexpected, the consulting company can initiate GPT-3 with a set of general prompts, such as:
"What are the most important factors for the client when it comes to the patient management system?"
"What aspects of the system would the client value most?"
"What are the key factors that will differentiate our proposal from the competition?"
"How can we make sure our proposal meets the client's needs beyond the technical specifications?"
领英推荐
"What are some potential blind spots or overlooked areas in the RFP that we should be considering?"
During gameplay with GPT-3, a fascinating aspect is the ability to keep asking it questions to generate additional prompts, uncovering both valuable and ridiculous insights. This approach ultimately leads to improved strategies. For instance, if the prompt "what aspects of the system would the customer value most" reveals insights into specific preferences for user experience, GPT-3 can provide more detailed prompts to explore potential solutions further, such as:
"Describe your ideal user experience for a patient management system."
"What are your top priorities when it comes to user interface design?"
"How important is ease of use to you when it comes to patient management systems?"
"What do you dislike about existing patient management systems on the market?"
“Generate three different user interface designs for the patient management system, each with a different look and feel.”
“Describe how user feedback can be incorporated into the design process to ensure that the final solution meets the client's expectations.”
“Generate a set of usability metrics that can be used to measure the effectiveness of the user interface.”
The game carries on as the consulting company generates mocked-up user interfaces, establishes a framework for usability metrics, and more. By simulating various scenarios using large language models, the consulting company can explore both the anticipated and unanticipated possibilities and craft a proposal that effectively addresses the client's requirements while minimizing the risk of failure and distinguishing themselves from competitors.
Leveling up your game with hallucinations
The saying "One man's bug is another man's feature" applies perfectly to the use of AI-based war games, as these simulations depend on exploiting a feature in AI systems that many consider a nuisance.?Let me explain.
Generative AI models like ChatGPT are based on deep neural networks that are trained on vast amounts of data to learn patterns and relationships between words and phrases. Once trained, the model can generate new text by predicting the most likely word or phrase to follow a given sequence of input text.
However, because the model is designed to generate text that is novel and creative, it can sometimes "hallucinate" content that is not based on any specific input. This can result in the generation of new and unexpected ideas, concepts, or language that may be surprising or even surreal.
In fact, someone recently shared a hallucination that mistakenly believed that I was “the founder and CEO of Broadridge Financial Solutions.” This could be because of my work with IBM in the blockchain field and Broadridge's previous involvement in blockchain initiatives. It's possible that an AI model recognized this connection and generated the false information.
Hallucinations can lead to game-changing solutions. Say the consulting company was using a language model to generate user interface designs for a client's app and received a response that included an unexpected feature - a chatbot for customer support. Upon further consideration, the team realized that this could significantly enhance the app's user experience and differentiate their proposal from the competition.
Although it's crucial to verify the accuracy of AI-generated content before making decisions based on it, such content can provide valuable insights and generate new ideas. In the context of war games, the unexpected scenarios and hallucinations generated by AI models can uncover unique insights and creative solutions that can give businesses a competitive edge. By embracing the unexpected and using AI to explore new scenarios, businesses can develop innovative strategies and stay ahead of the competition.
Next steps…
Now that you've learned how war games and AI hallucinations can level up your strategy, it's time to give it a try. Whether you're a business looking to stay ahead of the competition, or a curious individual interested in exploring the power of generative AI, there's no better time to play than now. I'd love to hear from you about your experiences and insights gained from conducting simulations. And if you're looking for additional resources to deepen your understanding, here are some references that might be helpful:
Serious Games. Revisionist History. (2018, July 26). Gladwell, M.
Shall We Play a Game? (2007, June 2). The Economist, 72.