Why can't you read my mind?!
The irresistible pull of new technology and its implications for entrepreneurship in global health.
My assistant and I had a long-running joke. I would ask her to do a task and explain what I needed. (Okay.... so my instructions were generally rushed and cryptic.) She would expertly complete the job, except it was different from what I intended. I would look at her and ask, "Why can't you read my mind?"
I have been remembering those interactions since I started working with the Large Language Model (LLM) ChatGPT. I discovered that the same traits follow me from my interactions with assistants, both real and virtual, resulting in less-than-optimal responses.
I have had the pleasure of working with some very talented assistants. As I learned more about engineering prompts for ChatGPT, I realized there is little difference in communicating with intelligent people versus smart algorithms. This is not a defense of my human interactions but a recognition that I need to get smarter with how I communicate- whether with humans or machines.
Many reasons humans misunderstand one another are similar to why interactions with LLMs can be frustrating.
Task requirements may be communicated poorly, and assumptions and biases can lead to misinterpretation or misunderstanding. A lack of clarity about the desired outcome, insufficient context, and lack of relevant perspective can lead to misalignment between the desired goal and the delivered response.
Of course, LLMs don't have bad days or poor listening skills, and they can't see nonverbal cues. The point is that the clarity and coherence of communications must match the complexity of the task.
My longing for a mind-reading assistant notwithstanding, the more productive approach is to learn how to communicate and interact with people and LLMs effectively.
Prompt Engineering
If you Google 'prompt engineering,' you will learn that there are courses in creating effective prompts and using LLMs to achieve your desired output. One such course is offered through Coursera by Jules White, an Assistant Professor of Computer Science in the Dept. of Electrical EngiĀneering and Computer Science at Vanderbilt University.
https://www.coursera.org/learn/prompt-engineering
I strongly recommend this course. Through a series of short, ~10-minute videos, Dr. White does an excellent job walking through the art and science of developing better prompts.
As emphasized by the course, we have yet to develop the best practices for writing these prompts, and we need to be open to exploring different approaches to writing prompts and providing feedback to the LLM to refine the responses. The creativity of the prompts, the types of questions that get asked, and the instructions on delivering a response are crucial for effectively using these tools.
Interactions with LLMs are more like conversations than queries, and you can instruct the LLM to act as a particular persona. For example, you can require it to respond from the perspective of a person, or even some inanimate object, an animal, or some system. The LLM can act as a research assistant, a colleague from a specific discipline, or an architect to help design a process or computer program. And just like that, you can create a "super team" of experts from different disciplines.
But I will let ChatGPT speak for itself in responding to my prompt:
I am writing an essay for scientists in global health describing the advantages of LLMs in developing better public health policies. Write a 300-word description of how LLM works and its advantages to policymakers in low-income countries
ChatGPT
Language Models (LLMs) have emerged as powerful tools that offer significant advantages to policymakers in low-income countries when developing better public health policies. LLMs, such as OpenAI's GPT-3.5, are advanced artificial intelligence systems trained on vast amounts of text data and designed to generate human-like responses to text prompts.
One key advantage of LLMs for policymakers in low-income countries is their ability to bridge information gaps. These countries often face resource constraints and limited access to up-to-date research and data. LLMs can help fill this void by providing instant access to a wide range of information and knowledge. Policymakers can pose questions or input prompts related to public health challenges, and LLMs can generate comprehensive and contextually relevant responses based on their training data. This enables policymakers to make informed decisions backed by the latest evidence, even in the absence of local expertise or resources.
Furthermore, LLMs promote inclusivity and democratization of policy development. Low-income countries may lack the expertise and infrastructure to conduct large-scale studies or consult with a diverse range of stakeholders. LLMs can assist policymakers by offering a broad perspective on public health issues. They can process diverse inputs, consider various perspectives, and generate policy recommendations that account for the specific needs and challenges of the population. This helps ensure that policymaking processes are more inclusive, representative, and responsive to the realities on the ground.
Another advantage of LLMs is their ability to aid in forecasting and decision-making. Policymakers can leverage LLMs to predict and model potential outcomes of different policy interventions. By simulating various scenarios, policymakers can assess the potential impact of different policies on public health indicators, resource allocation, and overall outcomes. This information empowers policymakers to make evidence-based decisions and prioritize interventions that are most likely to yield positive results.
Moreover, LLMs provide a cost-effective solution for low-income countries. They offer an efficient means of accessing and processing information without the need for expensive infrastructure or extensive human resources. LLMs can be accessed remotely and at any time, making them highly convenient for policymakers working in resource-constrained environments.
In conclusion, LLMs offer numerous advantages to policymakers in low-income countries. They bridge information gaps, promote inclusivity, aid in forecasting, and provide a cost-effective solution. By harnessing the power of LLMs, policymakers can develop more informed, evidence-based, and effective public health policies, ultimately improving the well-being of their populations.
LLMs remember previous parts of a conversation, so positive and negative feedback helps refine the responses, and the conversation can be shared via a link so that others can contribute and help refine the conversation.
Click here and view the rest of the conversation I started above.
Implications
My previous essays on the Velocity Made Good Substack have focused on addressing challenges in global health where the complexity and interconnected risk factors render any easy fix inadequate. These problems require insight from many perspectives, and we must seek out collaborators from relevant disciplines to get their views.
LLMs act as well-informed experts who can constitute a "super team" by taking on the persona of different experts in rendering a response. It can even suggest a persona best suited to help think through a problem to raise the level of discourse!
LLMs can act as research assistants, colleagues, or architects when needed. Learning to manage this "team" will become the key to success and require the creativity and patience to master the communication skills needed to get the most value from virtual "super teams."
For example, we can ask the "team" to brainstorm different approaches to solve a problem, test them, and get feedback. We can ask the LLM for any given situation to suggest alternative approaches. Some of them might be great, and others might be terrible, but just the fact that we are looking at different options will contribute to our sense of holiteracy and help us better think through the problem and figure out the solution that makes the best sense.
Admittedly, the richer the insights, the more complex the synthesis task. While LLMs can provide information, human creativity, and persistence are still required. We can try out any number of wild-eyed ideas with our team. However, we still need enthusiastic individuals with an expansive network of contacts to muster the will to implement the best of these ideas.
CODA
Despite the doomsayer's concerns about the existential risk of artificial intelligence, after listening to White's course, I believe that our innate desire to solve problems, and the power of LLMs, will be the source of the irresistible pull of this technology.
The skills and courage required to be an entrepreneur will become even more critical, particularly those who figure out how to manage best their virtual team. Importantly, these LLM-assisted entrepreneurs will become better at challenging top-down solutions imposed by experts with little or no knowledge of the local context.
Challenging the status quo and gaining a heightened awareness of the importance of context can lead to truly innovative solutions to global health challenges. Moreover, the skills acquired by LLM-assisted entrepreneurs from low- and middle-income countries can set the stage for them to profoundly impact the worldwide economy.