A Moveable (AI) Feast

Author’s note: I am currently on sabbatical in Paris, and this column is a hackneyed but heartfelt homage to the late, great Ernest Hemingway.

Inviting AI to the Table

The first book that I read after arriving in Paris was Ethan Mollick’s “Co-Intelligence: Living and Working with AI.” Mollick, a Wharton faculty member in entrepreneurship and innovation, offers a number of interesting perspectives on various roles (co-worker, tutor, coach, etc.) that artificial intelligence (AI) might come to play in our lives. My biggest takeaway from the book, however, was Mollick’s advice to “always invite AI to the table,” a call to action to simply start trying things with generative AI (GenAI) and develop a feel for where it does (and does not) have a positive impact.

It’s still pretty early in the process, but in the past few weeks, I have used GenAI tools to look for thrift stores and sort them geographically (fast and effective), create turn-by-turn walking tours around different neighborhoods (less successful) and write drafts of recommendation letters on my behalf (no comment). It’s a start.

Cognitive Amplifier

One morning, I headed out to a café to get together with Julien Boubel, one of the first people that I met after arriving in Paris. A Silicon Valley veteran now based in France, Boubel is the co-author (along with Pierre-François Ferley and Laurent Garnier) of the new book, “A Job for Christmas: Unwrap the Secrets to an AI-Fueled Job Hunt.” Boubel and his co-authors assert that without leveraging AI in a structured way, 21st century job seekers are bound to work harder and be far less successful in finding the right role, and that both employees and employers will lose out as a result. The book then leads its readers through a structured job search and hiring process, providing a series of concrete suggestions on why and how to utilize AI tools every step of the way.

Note: After my recent conversation with Boubel, I decided to send a copy of this book to all of my MBA students from last year, not only to help improve their professional prospects but to assuage my own guilt for being absent for their upcoming job search and graduation this spring.

Much like the AI-charged methodology presented in “A Job for Christmas,” there are thousands of books and software vendors exploring ways to use AI as a cognitive amplifier. A few have been in direct contact with me recently:

• Motivated by his experience as an Italian language instructor, my former MBA student Valerio Marachi has started a company called Duets.ai, which uses GenAI to analyze the dialogue between foreign language students and their tutors and then automatically create customized exercises to accelerate the learning process.
• Leveraging his extensive experience with call center operations and technologies, my college classmate Jeff Swanson is involved with an early-stage venture called XSELL Technologies that seeks to provide AI-based real-time coaching to improve the productivity of call center agents.
• Update.ai, led by my friend Josh Schachter, seeks to analyze the content of Customer Success Manager (CSM) calls with their customers to summarize the conversation, identify action items and deliver insights to help CSMs improve customer outcomes.

Displacement Effects

All of these are examples of AI being developed to improve effectiveness and efficiency, capturing what labor market researchers call “productivity effects.” But today, it is also not hard to imagine a world in which AI soon eliminates the need for human involvement in many tasks (“displacement effects”). Historically, there has been a lot of economic research on the impact of technology, but most of what we imagine might happen as a result of GenAI is still largely speculative.

In a recent conference paper, Dandan Qiao (National University of Singapore), Huaxia Rui (University of Rochester) and Qian Xiong (Tsinghua University) examine data from online labor markets in an attempt to understand the impact of GenAI on freelancers’ workloads and wages by examining patterns before and after the launch of ChatGPT in late 2022. Although productivity and displacement effects are both evident, the net results are quite varied across different specializations, with productivity effects having already been eclipsed by displacement effects for some role categories, most notably translation and localization. Based on their extensive data analysis, the authors also offer up an unnerving conjecture: once AI advances beyond a certain threshold of competency for a particular skill set, the market demand for human labor associated with that specialization inexorably declines. Yikes

Risks to Humanity

I enjoyed a recent café conversation with Stuart Russell, a longtime AI researcher, co-author of the standard university AI textbook and a leading voice on AI safety and ethical AI development. What Russell calls the “standard model” for artificial intelligence models – one that is particularly familiar to economists and operations researchers – involves defining intelligence in terms of the ability to optimize some fixed and measurable objective and then designing the system to achieve that objective. Russell acknowledges that this approach has been “increasingly successful” but anticipates that it will eventually bump into a variety of limitations. 

Russell has been thinking about the ethical and safety issues associated with AI for many years. In his 2019 book, “Human Compatible: Artificial Intelligence and the Problem of Control,” Russell argues that the current approach to AI development poses significant risks to humanity and proposes a new model for creating beneficial AI systems. Specifically, he starts with the foundational principal that AI systems must be designed to always defer to the needs and desires of human beings, even though – and especially because – those “requirements” are neither fully articulated nor completely known. As such, he asserts that AI systems must be designed to operate with humility because of the inherent uncertainty about both objectives and outcomes. He has also called for improved governance of AI research and development.

More recently, Russell has poured a lot of his energy into launching the International Association for Safe and Ethical Artificial Intelligence (IASEAI), an independent organization committed to ensuring advanced AI systems are guaranteed to operate safely and ethically, benefiting all of humanity. IASEAI’s inaugural conference was held in early 2025, and the organization seeks to pursue its goals through influencing public policy, supporting research and education efforts, and building a community that includes business, governmental and scientific stakeholders. I am most grateful to Russell, his colleagues at IASEAI and everyone who is supporting IASEAI’s mission. I will be looking for ways to contribute, and if you are reading this, please consider joining me.

In the age of increasingly powerful AI with vast capabilities that we do not truly or fully understand, it is hard to say how far away we are from either artificial general intelligence or human extinction, but each day it feels like we are getting closer to both. “There is never any end to Paris,” Hemingway wrote in his Parisian memoir “A Moveable Feast.” I want to believe that he was right, but I also worry that he may have spoken too soon.