Tragically for creatives, the value of novelty is often eclipsed by the safety of consensus. This is especially true when you’re working with AI, where overvalued consensus is a barrier to innovation. As users of what is popularly called “generative AI” and what technologists prefer to call Large Language Models or “LLMs,” we face an important challenge: breaking free from the echo chamber of conventional thinking.

Large Language Models, despite remarkable capabilities, heavily favor answers that align with the status quo, resulting in safe but boring responses. This reliance on consensus not only stifles creativity and maims attempts at art, but also risks perpetuating a cycle of relentless mediocrity. If we aim to innovate with AI, we must disrupt this, and force these models to help us explore ideas that challenge convention and ignite change, because if we can’t, perhaps in some contexts, we should consider abandoning the use of these tools entirely.  Brainstorming with early input from consensus thought is… a profoundly counterproductive exercise.

Recognize what you’re dealing with. LLMs are something like calculators for words—they are designed to reflect the average of all available data and mostly repeat and amplify what they’ve overheard. If you want more than a Family Feud “survey says” style response, you will have to push the software forcefully to color outside the lines. This is especially important when the consensus is wrong. A college professor friend of mine recently pointed out that she’s trying to help students contend with the effects of generations of systemic racism, and that when her students lazily give in to letting the LLM think for them, this is not only cheating them out of the opportunity to acquire insight, but also teaching them to embrace the exact consensus ideas that she wants them to learn to question and disrupt.

Also, these models hallucinate sometimes.  Especially when we tell them to be creative.  That’s a significant risk, and we should treat it like one.

Nevertheless, you’re probably still reading because you want to explore strategic prompts that attempt to mitigate some of these limitations. By deploying these methods, we can attempt to stimulate LLMs to venture beyond the safe and familiar, outside the boundaries of conventional thinking, in a process designed to open new neural pathways – for us. These approaches are designed to harness some of the untapped potential of LLMs and inspire transformative thinking in you.  Here are five ideas to get you started:

1. Prompting for Disruptive Trends

Strategy: Ask the LLM to identify emerging trends or disruptions within a specific industry or market that are not widely recognized.

Example 1: Instead of asking the LLM to list current tech trends, prompt it to focus on unpopular market segments that could be disrupted by new advances in technology.  This might lead to unique insights about upcoming innovations that aren’t yet mainstream.

Example 2: Request the LLM to identify sectors or markets currently overlooked by major players but that show signs of significant potential for disruption. This approach could uncover niche opportunities that have not yet been exploited.

Why it Works: This strategy challenges the AI to move beyond established trends and identify novel, potentially game-changing insights, directly addressing the need for creative thinking and breaking free from consensus, but you have to explicitly ask it to do this.

2. Reverse Engineering Successful Strategies

Strategy: Analyze unconventional strategies or innovations that have succeeded and propose how these principles might be applied to a different field or problem.

Example 1: Review a successful guerrilla marketing campaign and ask the LLM how similar tactics could be used in the healthcare industry. This could lead to fresh approaches in a field that typically relies on traditional methods.

Example 2: Examine an unexpected business model, like subscription-based services in an industry where they are rare, and prompt the AI to suggest how this model could transform a different sector.

Why it Works: This strategy leverages existing successful outliers to spark creative ideas in new contexts, pushing the AI to move beyond standard approaches and explore innovative applications.

3. Alternative Analysis Frameworks

Strategy: Use less common or unconventional analysis frameworks or metrics to evaluate opportunities and generate insights.

Example 1: Instead of traditional financial metrics, prompt the AI to use a framework based on social impact or environmental sustainability to assess potential investments. This might reveal opportunities that align with broader values beyond financial returns.

Example 2: Request an analysis based on cultural or psychological trends rather than purely economic factors. For instance, evaluate a new product idea through the lens of cultural shifts and societal needs rather than market size alone.

Why it Works: This approach challenges the AI to apply non-traditional metrics, encouraging it to provide insights that are not limited by conventional business analysis, thus fostering more creative and impactful ideas.

4. Voice of a Respected or Unexpected Figure

Strategy: Prompt the AI to generate content in the style or voice of someone who is admired or unexpected in the context of the topic, especially if that person doesn’t usually discuss it.

Example 1: Ask the AI to write about the future of technology from the perspective of a historical figure like James Garfield. This can lead to novel insights or an interesting way of describing or thinking about a topic.

Example 2: Request a summary or a strategy plan written as if it were authored by a comedian like Robin Williams. Your goal is to trick the model into doing something unusual.

Why it Works: This strategy encourages the AI to adopt unexpected perspectives, fostering unique and unconventional responses that break from standard analytical approaches and provide fresh viewpoints.

5. Requesting Outlying Viewpoints

Strategy: Identify the prevailing consensus on a topic and specifically ask the AI to generate and defend an outlying or contrary viewpoint.

Example 1: If the consensus is that remote work is the future, prompt the AI to explore and justify reasons why in-office work could make a significant comeback, challenging the prevailing narrative.

Example 2: If the general advice is to focus on digital marketing, ask the AI to argue for the resurgence of traditional marketing methods and explore scenarios where they might outperform digital strategies.

Why it Works: This strategy directly addresses the need to challenge prevailing norms and generate non-consensus views, encouraging the AI to produce insights that defy the status quo and inspire new thinking.

Conclusion

Relying on consensus-derived responses from AI tools can limit our potential for innovation and change. The tendency to stick with conventional wisdom and average answers often leads to safe outcomes, but safe is boring. “Good” is frequently the enemy of “safe.” To innovate, even a little, we therefore need to push boundaries and habitually prompt LLMs to move beyond the average of all available ideas.

Forcing the LLM to abandon its cozy cocoon of consensus will require some badgering and intentional effort on your part. By using the strategic prompts we’ve discussed, you can experiment with new ways to challenge the model to generate responses that break away from the norm and in the process, spark your own human creativity. Cultivate a mindset that values innovation over predictability and embrace the messy journey to new ideas.

Ultimately, pushing AI to deliver unconventional responses isn’t just about leveraging technology more effectively; it’s the difference between the potential to create value by amplifying creativity and the possibility of attenuating human creativity by elevating consensus to a pedestal it doesn’t deserve.

Instead of thinking for us, LLMs can take our ideas, and apply new combinations, patterns, and perspectives before returning them to us for further analysis. This conversational model for LLM use can be very helpful.