Keeping It Human in a Digital World (2)

Language is the primary way we communicate, learn, and transmit knowledge. It’s fundamental to our human experience, and so this shift has huge implications:

We can automate things we never could before. Now, software can support us with open-ended goals and use unstructured data. This breakthrough unlocks the opportunity to apply machine power to activities long considered uniquely human. It’s why it is now possible to use machines for tasks that require reasoning and what we think of as creativity.

We can turn to AI for help in turning vague train-of-thought notes into an actionable project plan, developing a marketing strategy for a still-forming product idea, or creating a strategy to smooth our relocation to a new city. It can synthesize key points from any document, describe the meaning of a cartoon, or create an image from a sentence of text. We can snap and upload a picture of a problem around the house—a hole in the wall, or a broken appliance—and receive recommendations on how to repair it along with a shopping list for the hardware store. These new capabilities open the door to practical, software-driven support across a vast new spectrum of daily challenges and needs and could fundamentally change how we approach everyday problem-solving.

We can command machines in our everyday language. We will also see the impact of our newfound ability to get machines to do our bidding simply by using the language we write or speak every day. Historically, we had to learn how to code or use complex software to get the machines to do a lot of specialized tasks for us. Only those with this technical expertise could really tap into these capabilities. Now, for a growing range of specialized tasks, we can simply instruct machines using everyday language to produce results that once demanded years of skill development.

We can instantly create complex illustrations or videos by entering a few sentences (or even a phrase) into a text box. We can describe the functionality we want for our website—for example, a scroll bar or a spinning prize wheel—and get functional code within seconds. We can compose songs in any style with just a few descriptive words. This shift could democratize creation, allowing people from all backgrounds to bring their ideas to life without technical barriers.

Conversation with machines can enhance our thinking. Just as talking through ideas with a person helps us clarify thoughts and generate new insights, conversing with AI can enhance our thinking process. The back-and-forth dialogue that we use when working with AI externalizes our thoughts, which can trigger new associations and perspectives we might not have considered on our own. AI’s ability to respond in natural language means it can serve as a tireless thought partner or intelligent sounding board for our thoughts, offering diverse viewpoints and helping us explore ideas more deeply.

The idea that talking out loud can help us think has roots in philosophical discussions dating back to Ancient Greece and Rome. Much later, German writer Heinrich von Kleist articulated this concept in his 1805 essay “On the Gradual Formation of Thoughts During Speech” where he explored how we discover new ideas through the process of speech itself. Essentially, actively speaking can transform abstract, obscure ideas into more concrete ones, making speech a creative process that amplifies our thinking power.1 This phenomenon plays a big role in a learning theory called collaborative problem-solving, which has been well-supported by modern research to enhance our problem-solving abilities.2

Conversation can also support associative thinking, where one idea or concept leads to another. Our brains create networks of related concepts, so when one idea is activated, it can trigger a cascade of related thoughts. Research links associative thinking to creativity, problem-solving, and richer communication.3 AI easily and rapidly generates a broad range of ideas, providing us with more diverse opportunities to engage in associative thinking, potentially expanding our creative and problem-solving capabilities.

The Accidental Breakthrough

Large Language Models (LLMs), which are behind the recent acceleration of AI, were originally developed to be word prediction machines. Input a phrase, and they were designed to predict the most likely next word. We don’t fully understand why this original design led to the advanced capabilities that we can access today—capabilities that surprised even those working at the front edge of AI. “The crazy thing,” writes Ethan Mollick in his book Co-Intelligence: Living and Working with AI, “is that no one is entirely sure why a token prediction system[i] resulted in an AI with such seemingly extraordinary abilities.”4 Adam Cheyer, the cofounder of the startup that created Siri, explains, “I’ve been working for almost forty years in AI and more than thirty years in conversational AI and I am one of those who say: I never thought I would see what’s happened in this last year in my lifetime.”5

Discovering the Alchemy of Minds + Machines

People have never before had broad access to technology that can create, that can collaborate, that can help us think—all the thought, all the creativity, it had to come from us. But now AI can augment our decision-making and creation. From the C-suite to college students, we are all simultaneously discovering how to navigate this opportunity and the challenges it is throwing our way—it’s that new.

We are at the dawn of a new relationship with machines. This may not be a relationship you asked for, or are even sure you want. But those who can develop an effective working relationship with AI will discover that when we responsibly combine human intelligence and machine intelligence, we can achieve what neither could do alone.

This contributes to the sense of magic and mystery surrounding AI in this moment. Through math and patterns in our data, these systems uncover an underlying order in our world that we may not even be conscious of, and use it to create in ways we once thought only humans could—doing it ever faster, more affordably, and sometimes even better. As this capability is being pulled into our physical world through devices, apps, and robotics, it is not only becoming more entwined with our daily lives, but also pushing us to reevaluate what truly is human and where we still need and want human interaction and creation.

iIn the context of natural language processing used by Large Language Models (LLMs), a token refers to a single unit of text. This can be a whole word, part of a word, or even a punctuation mark. Modern LLMs are designed as token prediction systems, meaning they predict the next word (or “token”) in a sequence based on the context of the preceding tokens.

Notes