Do People Need to Know How to Write Anymore?
Tell Me a Story
Do We Need to Know How to Write Anymore?
I like teaching, and last semester was particularly enjoyable. I taught Tech Innovation on the west coast in Stern’s Tech MBA program, and two sections on Systematic Investing. I brought machine learning to Wall Street in the mid-90s, and have been trading systematically since then, so I’ve experienced a wide range experiences with AI in investing that provide a rich background for course. I start each three-hour session with roughly 45 minutes of free form discussion as long as it’s related to Finance or Artificial Intelligence, but it sometimes strays into life, the universe, and everything.
A major requirement for the Systematic Investing class is a term project, where students are expected to apply the thinking and concepts from class to an investment area of their choice. The project requires asking an original question and answering it using the scientific method. It’s an opportunity for them to be creative. For example, an interesting question someone asked last semester was whether and why options for some stocks become mispriced just prior to earnings announcements. Options are financial contracts that give the buyer the right to buy or sell an asset at a stated price within a specific period. The project investigated whether an algorithm could profit from the mispricing.
Having taught my Systematic Investing course for almost twenty years, I know what to expect from the projects. But this time, I was in for a bit of a surprise: in one of my sections, students turned in very polished papers. But they were written mainly by ChatGPT.
The larger question this raises is whether people need to learn how to write anymore.
We’re already seeing the application of AI to all kinds of creative tasks, so what’s stopping GPT from doing all our thinking and writing for us as well? Could sophisticated prompt engineering do a better job?
The Art of Storytelling
The ChatGPT written projects were different from the good human ones in two ways.
GPT written papers had a distinctive bias. They read like articles from scientific conferences or journals, and were padded with highfalutin’ words and math. The results were dressed up in fancy language. The words meticulous and robust figured prominently, as in, the data were collected or cleaned meticulously, the analysis was meticulous, and the models and results were robust. These are typical claims in academic articles.
In contrast, the good human-written projects asked interesting and well-defined questions, and most importantly, the answers were narrated like a story. What was the source of the idea and the thinking behind it? What was the question? Why was it interesting? What were the data collection and analysis challenges? How did the idea get refined during the investigation? Were there any surprises? The story brought out the process of discovery and learning.
Stories are incredibly powerful, but creating a good story is an art that is shaped from within, forcing us to connect the dots into a coherent arc. The creator of TED, Chris Anderson, describes it well in his book on public speaking, but it applies equally to writing. I wrote my TED talk multiple times before testing it on a few audiences, and memorized it because every word was deliberately chosen, although I allowed myself some wiggle room during the talk based on what I was sensing from the audience.
One of the hardest tasks for scientists and business people is coming up with the right story. A trading strategy that tells a story is a lot more compelling to an investor than one that just presents the numbers. In investing, a good story not only explains the numbers, but has predictive power – it predicts the future numbers as well. In other words, the story establishes expectations about the future.
Stories integrate the science into a larger picture. In my podcast conversation with Philip Tetlock, author of the book Superforecasting, he pointed to the success factors of humans who are good forecasters. They are open-minded, incessantly curious, understand probability, tolerate dissonance, revise beliefs continually as data become available, search for the right analogies, and anchor themselves with the right “base rates,” meaning, how often things occur on average. Teams of superforecasters are even better because they communicate in terms of well-defined questions and information.
The valuation guru Aswath Damodaran is an example of a superforecaster in Finance. In addition to his quantitative framework for valuing companies, Damodaran exhibits the traits of Tetlock’s superforecasters and makes heavy use of analogies in structuring a problem. For example, if you want to understand the food delivery potential in a new country, anchor yourself with data from a country where it already exists, and then make adjustments. I discussed the integration of narrative and numbers on a previous Brave New World podcast with Damodaran, whose heavily followed blogs, Musings on Markets, are all about the integration of stories and numbers.
Creating a clear and concise story is hard. It requires continuous refinement. The process of writing leads to clarity that does not emerge otherwise.
Still, there’s a lot of room for AI to improve our writing. You can ask GPT, for example, what a good question is. It will tell you that a good question has relevance, clarity, testability, falsifiability, novelty, importance, and is feasible to investigate. That’s a good checklist, but it’s still your job to come up with the question and its context. Once you have a well-defined question, you can ask it to summarize the state of the art related to your question and think about how to weave it into the narrative. In the old days that research would have taken weeks. Now it takes seconds.
The larger takeaway is that the better defined your question and the more you know, the more AI will enhance your thinking and the creative process. On the other hand, the less clear you are, the more generic and undifferentiated it will make you. While it may sound coherent, it will lack depth.
Writing is a process of continuous improvement without end. I read that the comedian Louis C.K puts his best joke from his last show at the beginning of the next one, and since he always needs to finish stronger, his last joke gets better. Writing is no different.
My latest Podcast
My latest podcast guest on Brave New World is Vlad Barbalat. Vlad is the President of Liberty Mutual Investments and the Chief Investment Officer of Liberty Mutual Group, one of the largest property and casualty insurers in the world. Vlad is a former student of mine from 20+ years ago, and I have fond memories of him and his two buddies sitting in the front row. They’ve all had very successful careers.
Vlad and I had a free-flowing conversation on a range of topics – the immigrant experience in America, universities, the world of insurance and investing, and what it means to be a leader in a large business enterprise. So, check out my episode with Vlad here:
Until next time.
V/