Everyone is a Prompt Engineer in the Age of GenAI
Everyone is a Prompt Engineer in the Age of GenAI
Every worker in the year of 2024 will be a prompt engineer. Since the inception of ChatGPT in 2022, there has been a tremendous interest, financial and human resources pouring into generative AI research, app development. My prediction is that all companies will try to make generative AI work for them in the year of 2024. Innovation and entrepreneurship scholar, Ethan Mollick, argues that with generative AI, organizations need to reimagine how organizations organize, and create different work processes. Business leaders should take responsibility to actively participate in this redesigning future of work with AI as an essential part of the organization processes. The main argument hinges on the idea that with generative AI, a team work process would be something akin to the idea of “collective intelligence” between humans and AIs where a team’s productivity is a synergy between human workers and their AI colleagues. Each AI team member can do a job, or perform subtasks, sub-processes. The figure below demonstrates an example of how a normal process can incorporate AI workers into the work process to speed up the delivery of a particular assignment. In essence, we’re entering a work culture where each person would bring with them a suite of AI agents to augment, automate, and enhance their work.
The ideal process that Ethan Mollick painted is very intriguing. An open question for business leaders and each worker is how could they actually implement this ideal work process in practice. I would argue that in 2024 every worker should learn the art of prompt engineering, and should perform in one way or another the job of a prompt engineer. While the title of prompt engineering as a new job category after the introduction of chatGPT, it should instead be conceptualized as a set of skills that every worker should master in this GenAI age. Any organization that creates a structure, a work culture of learning, upskilling that promotes prompt engineering as an essential skill for their worker will get ahead in this race to harness generative AI power. Prompt engineering should no longer be a specialized skills that are relegated to a group of specialists called prompt engineers. Similar to the introduction of personal computers, type writing is no longer a specialized job. Everyone now learns how to type in grade school.
Chief Scientist Jaime Teevan of Microsoft recently pointed out that in order to unlock tremendous power of large language models, we (yes the collective we) should not focus on fine-tuning a large language model or build a new large language model, but rather the main values to organization are in mastering prompt engineering, and coming up with creative prompts that unlock the productivity promises of large language model. She further observes that prompt innovation has mainly come from the research community, which should not be the case. Prompt engineering innovation should come from billion large language model users on a daily basis. In other words, as organizations incorporate GenAI in their workflows, any employee can innovate, and add to the collective power of prompt engineering.
A recent research between Harvard Business School and the Boston Consulting Group, examining professional consultants’ productivity gain when using generative AI found that “professionals who skillfully navigate this frontier gain large productivity benefits when working with the AI, while AI can actually decrease performance when used for work outside of the frontier” (emphasis is mine). The tasks that the authors examined were non-routine tasks which make automation less ideal, and that the introduction of chatGPT ushered in “an entirely new category of automation, one whose abilities overlapped with the most creative, most educated, and most highly paid workers.” For workers who could resolve tasks that are outside of the frontier, they exhibit “centaur behavior.” These users “switch between AI and human tasks, allocating responsibilities based on the strengths and capabilities of each entity. They discern which tasks are best suited for human intervention and which can be efficiently managed by AI.” Don’t we all want to have this “centaur behavior” now?
Besides, the authors found that workers who have access to GPT-4 with an overview of how to effectively use prompts performs slightly better than workers who are only given access to the tool alone. This finding is very important. Even when the tool is made available, workers with training will perform better than workers without prompt engineering training on average. This insight has profound implications for workforce training, and educational curriculum design for wider AI adoption.
In the year of 2023, we have observed that businesses and educational institutions originally took a risk-averse stance with regard to chatGPT. Now they have put guardrails and created internal systems that let employees take advantage of chatGPT. ChatGPT is now conceptualized as another essential tool for a business akin to Google. The question is no longer about whether it should be allowed. The question has become how to unlock chatGPT and other large language model tools in the form of chat at the full scale based on what we have learned about how this tool is being used at work.
To take advantage of these tools, I suggest that prompt engineering should become an essential skill for every worker. This has a few implications for the educational system, and workforce training programs:
Educational institutions
Prompt engineering should be introduced early in one’s educational journey. Students will use chatGPT. The question is how they can use chatGPT to help them get the most out of it.
I recently helped a college students in critiquing their scholarship and internship essays. I read their essays with an editorial eye, and suggested places to cut, and word choices to use. Yet, the students instead of taking my suggestions to micro-edit every sentence, they input the entire essay in chatGPT with my suggestions for the machine to rewrite the entire paragraph, sometimes an entire essay for them. What I would have done instead is that instead of asking chatGPT to write an entire paragraph for them, the student should have asked chatGPT to critique every sentence, and suggest micro phrases, and changes, and make the differences known with proper rationale. This is a process that chatGPT can do relatively well. In this case, the student was too focused on churning out a well-formatted essay in a short period of time, instead of paying attention to word choices, their personal voice, sentence structures, transitions, namely, the many things that make each essay unique and personal to an individual. This student would benefit so much from getting personalized guidance on what it takes to write a great motivation essay, what good writing principles are, and how to use chatGPT to achieve those goals. In a sense, chatGPT is a personal assistant, but the more detail instructions one uses to give it, the more one would get out of it. This is akin to when an artist creates a piece of art, using a predetermined template is different from the process of exploring an idea, and drive it to the fullest conclusion of it using a tool such as an easel or a camera. It all starts out with an idea, while the process could be learned, and could be figured out while this artist exercises their agency to take advantage and master the tool.
Prompt engineering should not be conceptualized as a skill that is for engineering students only. Every student should learn it. Each specific prompt engineering style might depend on the subject matter at hand for example writing classes would utilize chatGPT very differently than say in a math class.
Having an agreed upon standard of what kind of prompt engineering is for what levels, and how to design a good prompt is an important critical thinking skill that prepares them to succeed in the future jobs.
Organizations
Many organizations are now having their internal chatGPT-equivalent tool. Prompt engineering should not only be a course that engineers and technical people take. Instead, everyone should have skills to write a good, functional prompt. Each organization should also come up with training programs akin to what I mentioned above for educational institutions.
Each organization should proactively upskill their workforce. There are many training courses on different platforms for their workforce to use nowadays. This must be something that businesses should encourage.
What I am most afraid of is that every business is trying to introduce tools such as ChatGPT in their organization, but if it is not incorporated into the workflow of each worker on a daily basis, it would become irrelevant. This is where HCI scholars, UX researchers will have a role to play. The difference between introducing a tool, and realizing the productivity gain is tremendous. The question is not about introducing the tool, but to empower their workforce to effectively use such tool to unlock the most productivity gains. This is the question of “How.” Any organization able to solve this question of upskilling their workforce and encouraging them to use chatGPT and the equivalent tools would unlock the most potentials faster, and for sure will be better off in the age of GenAI.