Ctrl-Z Strategy: What if Every Business Decision Had an Undo Button?
Strategy Moves from Prediction Machines to Imagination Machines
Games and Simulations: The Rise of Imagination Machines
In the world of artificial intelligence (AI), it’s easy to think we’ve already seen it all. Self-driving cars, world-champion Go programs, hyper-realistic voice assistants—these breakthroughs once seemed futuristic, but have quickly become the new normal. Yet beneath the surface, a new frontier is emerging, one that promises to push us beyond mere “prediction machines” into a world of “imagination machines.” Leveraging advances in generative AI, agent-based simulations, and game-inspired training, we are on the verge of simulating entire futures. And the implications for business strategy, scientific discovery, and human creativity are enormous.
From Prediction Machines to Imagination Machines
For the past decade, the AI landscape has been dominated by systems that excel at analyzing massive data sets and predicting outcomes. In his book, Making Sense of Chaos: A Better Economics for a Better World, complexity scientist J. Doyne Farmer critiques how traditional models too often reduce the future to linear extrapolations from the past. Farmer advocates for new, more nuanced approaches—particularly agent-based simulations—that can capture the unpredictability and feedback loops inherent in complex systems like economies.
At the same time, large language models (LLMs) and generative AI have burst onto the scene, making creativity and ideation suddenly more accessible. These “imagination machines” do more than predict what comes next in a sequence of words or numbers. By combining context, memory, and a capacity for novel output, they can help us envision scenarios not strictly dictated by historical data. If “prediction machines” revolve around data-driven forecasting, “imagination machines” push us to explore multiple hypothetical futures—and then help us decide which ones we want to make real.
Why Games?
A classic catalyst for AI breakthroughs has been the use of games. Demis Hassabis, cofounder of DeepMind, has repeatedly called games a “gym for the mind.” They have clear rules, built-in metrics for success, and enormous complexity. Research teams can train AI systems in a controlled environment, and in doing so, gain rapid iterative feedback that’s often unavailable in the real world.
Clear Objectives: Games like Chess and Go come with clearly defined goals (checkmate, surrounding territory). Systems can quickly identify whether they’ve won or lost, which is crucial for reinforcement learning.
Synthetic Data Generation: AI agents can play these games millions of times, producing massive training data. No need to wait for real-world events or risk real-world consequences.
Complexity: Analyzing games and their branching factors (b) and depth (d) introduces levels of complexity:
\(Complexity=b^d\)\(Chess: 35^{80} \)\(Go: 300^{200}\)In comparison, a real-time game like StarCraft has a branching factor so high that researchers sometimes call it “effectively infinite.”
\(Starcraft: 10^{200^{36000}}\)
When DeepMind’s AlphaGo defeated the world champion Lee Sedol in 2016, it showcased a kind of machine creativity, generating moves that no human master had considered. From there, DeepMind applied the same reinforcement-learning methods to complex protein-folding challenges. Such leaps illustrate how game-trained AI can transcend its original environment to tackle high-value scientific and business problems.
Beyond Training: Simulations as a Glimpse of the Future
Games are not only about beating grandmasters or leveling up your character. They also serve as prototypes for simulating real-world environments. This is evident in the 2023 Gran Turismo movie, which dramatizes the true story of Jann Mardenborough, a gamer who translated his Gran Turismo racing-sim skills into a bona fide career on professional racetracks. As one quote from the film puts it, “These players have clocked more track time than you have in your entire career.” Hours in a simulation might not replicate every nuance of physics, but they come surprisingly close.
The same logic applies in areas beyond motorsport. Think of agent-based economic simulations, in which virtual “agents”—each representing a consumer, a firm, or a financial institution—interact according to specified rules, revealing emergent behaviors. Farmer and others argue that such simulations can model policy interventions or test scenarios without waiting for real-world crises to unfold. In short, a well-tuned simulator is both a laboratory and a crystal ball.
The Age of Strategic Imagination Machines
With the arrival of generative AI, we’re moving from reactive analyses to proactive, imaginative scenario-building. Consider:
VOYAGER – An experimental system that uses GPT-4 to autonomously explore and master tasks in the game Minecraft, building a “skill library” and refining its capabilities through self-generated goals.
Generative Agents – Systems that integrate large language models with rich memory structures and reflection loops, enabling entire virtual communities to form relationships, pass on information, and coordinate group events like Valentine’s Day parties—all in a simulated environment.
These and similar projects signal a move away from single-task optimization toward multi-agent ecosystems capable of creativity, complex collaboration, and continuous learning. Call them “Strategic Imagination Machines.” They can generate business scenarios, prototype new product strategies, or test organizational changes—in a safe, synthetic environment.
Augmented Intelligence and the Krypton Factor
One of the biggest shifts in this new era is the move from working alongside machines to truly thinking with machines. Traditional AI systems have been exceptional at parsing large datasets and making predictions. Now, generative AI and agentic simulations provide us with divergent thinking on tap, offering a flood of possible futures or solutions.
But here’s the catch: when everyone has access to the same generative tools, the baseline for “superhuman” insights rises dramatically. That’s the essence of the “Krypton Factor,” where if all of us become “Supermen,” then being super is no longer a competitive advantage. Instead, the real differentiator is human expertise—the domain knowledge and creative instincts that enable us to:
Define the Right Problems: By knowing what truly matters, you can direct the AI toward meaningful simulations rather than trivial ones.
Combine AI’s Divergence with Human Convergence: Machines can suggest countless scenarios, but you need to interpret, select, and refine.
Inject Unique Data and Context: If the AI’s knowledge is purely generic, its simulations may be shallow. Proprietary data or specialized insight can give your simulations unique depth.
In short, augmented intelligence is not about handing everything over to AI. It’s about leveraging AI’s ability to explore and test possibilities, while you—guided by your personal “mutant” expertise—decide which outcomes are worth pursuing. That synergy is what propels us beyond mere predictions and into the realm of strategic imagination.
Lessons from Gran Turismo, Chess, and Go
Across everything from Mardenborough’s success in racing sims to AlphaGo’s victory over Lee Sedol, we see common threads:
Safe Training Ground: Simulations allow you to fail fast and often. In racing sims, you can spin out a thousand times without risking real injury. In strategy games, you can try bizarre gambits without losing your entire marketing budget or brand reputation.
Accelerated Learning: As the Gran Turismo movie highlights, you can log more virtual hours than you ever could in real life. A human with a chess engine can train, exploring 100’s of games in the time it takes a human to play a single tournament.
Cross-Domain Transfer: Once mastery is achieved in a simulated domain, that knowledge can often be adapted. Insights from StarCraft AI might improve supply chain management. Protein-folding breakthroughs from AlphaGo-inspired systems might open new frontiers in drug discovery.
From Games to Business Simulations
The corporate world is quickly catching on to the power of simulation. Historically, large organizations have relied on analytics teams to forecast demand or set pricing. Now, with generative AI and multi-agent models, we can imagine entire digital “twins” of companies, markets, or even entire industries. Instead of static spreadsheets, leadership teams can “play” these scenarios, adjusting variables to see how a change in policy or a sudden supply shock ripples through a simulated environment.
We’re also witnessing early experiments in “agentic workflows,” where AI “agents” are given tasks—like researching a competitor or refining a product offering—and left to orchestrate complex steps autonomously. Much like Centaur Chess where humans and AI collaborate to play chess, bringing AI simulations together with humans will allow strategists to collaborate to explore multiple possibilities before picking a unique competitive advantage.
Why This Matters Now
Easier Access to AI: Tools that were once the domain of specialized PhDs are increasingly accessible through user-friendly platforms.
The Creativity Factor: Generative AI allows for “blue-sky” thinking, factoring in not only known risks and probabilities but also novel ideas and scenarios. Augmenting humans and GenAI will open up novel solution spaces.
Faster Iteration: We can rapidly prototype entire new strategies. Instead of running a month-long pilot in the real world, we test multiple futures in a meeting.
The potential for innovation and disruption is immense—so is the ethical responsibility. Once your imagination machines are robust enough to shape real-world decisions, the stakes become very high. Issues of bias, resource allocation, and governance come to the fore. That said, if we approach these systems thoughtfully, we might unlock new paths toward solving global challenges, from climate change to social inequality.
Closing Thoughts
“Where do you get to practice your decision-making under pressure?” Demis Hassabis
His answer is games—those safe playgrounds for the mind that, ironically, prepare us for the real world. Today, we stand at the crossroads of game-inspired simulation, agentic workflows, and generative AI. It’s a moment that shifts us from simply predicting the future to actively imagining—and shaping—it.
As these imagination machines move beyond the research lab and into our everyday tools, we have the opportunity to transform how we train, learn, and innovate. Whether you’re racing in a virtual car or testing a new corporate strategy, simulations can compress years of trial and error into hours of game time. By harnessing this power carefully—and combining it with your uniquely human “mutant” knowledge—you can ensure that your augmented intelligence becomes a true competitive advantage, not just another commodity.
Thanks for reading!
If you enjoyed this exploration of games, simulations, and the next era of AI-driven imagination machines, consider subscribing. Feel free to share your thoughts, questions, or experiences in the comments below—after all, the best simulations are the ones we build together.