As use of generative AI proliferates across organizations, there are traps that can undermine any benefits it can bring. What’s needed to avoid any traps is the right mindset.
The leading traps of AI were explored by Elisa Fari and Gabriele Rosani, both with Capgemini Invent’s Management Lab, in their latest book, HBR Guide to Generative AI for Managers.
It’s understandable why people want to embrace genAI in a fast and furious way. “As you begin to use genAI in your daily work, you’ll experience a mix of emotions, including excitement, curiosity, and perhaps even a touch of apprehension,” Fari and Gabriele Rosani write. “Adopting the right mindset around generative AI will help you explore and experiment confidently and responsibly.”
They outline the leading traps that arise with human-AI collaboration:
Too much trust in AI: This comes from excessively trusting gen AI output, “without exercising their critical judgment, driven by laziness of a superficial impression that AI’s responses ‘sound good enough,’” Fari and Rosani warn. They recommend actively probing AI’s reasoning “by asking for clarifications and better articulation, requesting counterarguments, and identifying weak points.”
The risk of fabrication: It’s risky to “accept gen AI output as factual without verifying it,” the co-authors state. “Many are not even aware of the risk of AI’s fabrication. The autorotative tone of language models further fuels this risk.” It’s important, then, to ”validate statements against established facts from reputable sources and consult experts, particularly on unfamiliar topics.”
The tendency toward conformity: Be as specific and within context as possible, to avoid AI’s tendency to deliver bland, generic output lacking diversity and originality. This involves prompting AI “with contextual information for example, about a company’s values, unique value proposition, brand, and so on,” Fari and Rosani advise. “Ask AI to consider it as a guideline throughout the creative process.”
The speed trap: There’s a natural tendency to type, click or advance too hastily when working with technology, they state. Perhaps people working with AI “should slow down, and actively participate in the conversation. They should articulate their own perspectives and counterarguments.”
Solo trap: Some people may simply choose to work with AI and stop interacting with other humans altogether. This is not healthy. “This can reduce interpersonal communication and knowledge sharing within the team, resulting in more siloed work and a lack of diverse perspectives,” Fari and Rosani warn. They urge regular breaks from such ”solo AI interactions to engage face-to-face with colleagues. Seek feedback, integrate diverse viewpoints, and encourage peer learning.
To avoid these traps, Fari and Rosani urge developing a “genAI mindset” – embracing all that AI can offer, but keeping humans engaged – and skeptical.
This includes interacting with systems in a conversational way, trying and testing different AI models, and using it responsibly. A genAI mindset is all about continuous learning. “Hands-on testing reveals capabilities, limitations, effective usage techniques, risks, and potential mitigations.” the co-authors point out. “Adopting a learning mindset – asking ‘what it?’ – can help you reap the full benefits.”
This also represents important steps to take in skills development, they add. “Experimentation will help you understand which skills your team needs to develop most. Start where your testers encounter specific challenges or barriers in using gen AI. For instance, your team may find it challenging to effectively prompt AI to perform certain tasks or may struggle when crafting a suitable outline for a human-machine dialogue. Testers may encounter common pitfalls, such as placing too much trust in the machine, only to realize later that additional verification was indeed necessary.”
Knowledge of prompting techniques, “both basic – simple queries – and advanced – structured prompts – is an important skill, and is likely where you should start your upskilling efforts,” they state. “Many companies have also created ‘prompt academies’ to train their employees, while offering a platform to share and collect learnings, known as ‘prompt libraries.’”
