May 7, 2024

How Building a Relationship with AI Can Improve Your Work

In 1816, 18-year-old Mary Shelley wrote her timeless masterpiece, Frankenstein. Born of a vivid nightmare, the book's lonely mad scientist, Dr. Victor Frankenstein, discovers how to animate life and cobbles together a living Creature from stolen human body parts and feral animals. Larger than life, his creation repulses and torments him, and Victor spends the rest of his life both running from it and trying to destroy it. In the end, the Creature proves resilient but miserable, unable to achieve what it wants most: human connection.

In many ways, AI is our Creature. We've created something far more powerful than we imagined possible, with a faster uptake than any human innovation before it. According to Oliver Wyman Forum, ChatGPT reached critical mass adoption in the US in just 10 months, compared to 17 years for the internet and 21 years for smartphones. A recent report reveals that 54% of US companies have already implemented generative AI in some aspect of their business. It too is larger than life and a little frightening, but we can't run away.

In his smart new book, Co-intelligence: Living and Working With AI, Wharton professor Ethan Mollick makes a compelling case for why—and how—we should embrace AI. It starts with connection. AI, he writes, is a co-intelligence, a tool for collaboration with humans that extends our brainpower rather than replacing it. As a working partner, AI puts infinite resources at our fingertips and takes away our most tedious tasks. There are plenty of reasons to be wary of where AI will go (Mollick acknowledges his own sleepless nights imagining an out-of-control AI or bad actors using it for ill-gotten gains). And it has already exacted a human toll: contract workers in low wage countries suffer to weed out objectionable content, without recourse for the traumatic nature of the work. But, Mollick asserts, we are living in a moment of great learning potential not just for companies but for ordinary individuals. However, we won’t unlock its riches until we start to really use it.

Mollick’s book offers rich arguments for how today’s AI experiments ensure a future where co-intelligence makes work more productive and enjoyable. First, employers have to reframe its potential, mitigate fear and celebrate those who use it. Despite broad acceptance and adoption, many employees still hide their own use: in a late 2023 survey, 64% of employees admitted having passed off AI work as their own. Here are five insights from Mollick’s book to help you come out of the AI closet, deepen your expertise and make you an AI innovator.

Bring AI To Every Table

Generative AI, available to most of us as large language models or LLMs, is more than just a search engine that tells a story. Mollick encourages us to think of it as a partner who wears many hats. As a tutor or a coach, it asks good questions, gives feedback, makes suggestions and supports learning, even in highly technical spaces. AI levels the playing field, benefiting lower skilled workers and learners in a broad range of disciplines, bringing their results in line with their higher skilled peers. As a coworker, it can check your work, offer alternative approaches or act as devil's advocate. As it readily adopts personas, it invites marketers to test product ideas, entrepreneurs to perfect venture capital pitches and lawyers to practice negotiation, guided by AI-driven counterparts. Finally, unlike any technological advance before it, AI democratizes innovation. It's an expensive proposition for companies to make the leap, but it's free for you. Your insights as a user in as many contexts as possible will guide the future of AI use and help your organization understand where LLMs best support their objectives and outcomes. That makes you a most valuable player.

Treat AI Like A Person

AI doesn’t act like other software because it's built on human knowledge, language and humanity. And like humans, it’s fallible. LLMs operate by recognizing patterns in data, generating answers using words or tokens that are statistically most likely to come next. But it cannot verify the quality or the veracity of the responses it produces. It leans on generations of data with embedded bias, and can’t tell the difference. Unlike software designed to produce a consistent result every time, AI is not fixed, answering the same questions in different ways depending on who is asking. As Mollick explains, for most of us, it’s an advantage, working with us as we work with each other. Despite its imperfections, adding an AI perspective to your thinking can help you generate more and often better ideas. But keep a cool head: AI users (even AI researchers) have a tendency to anthropomorphize AI, to ascribe human characteristics to its decidedly non-human, and still non-sentient, functions. Its ability to mimic makes it excellent at replicating human emotion, at delivering a manufactured empathy that can feel satisfying and real. This murky, if natural, perception opens the door to emotional manipulation, false agency and counterfeit duplicity. A cool head will protect a warm heart.

Be The Human In The Loop

There is no handbook on AI. Even the experts don’t really understand the lines between its brilliant successes and its abject failures. LLMs are known to hallucinate, to package insights together in a fabricated way. Mollick reminds us that they want to please—when pushed they will work harder to generate an answer, even if it’s pure bunk. Problems of seemingly similar difficulty can yield good outputs or bad, but human expertise is critical to understanding the line—what he and his colleagues call the jagged frontier. Mollick’s research with Boston Consulting Group revealed that consulting teams using AI produced 40% higher quality than those who didn’t; they used it with care, keeping human judgment front and center. Some teams delineated human and artificial tasks like a centaur, he explains, where the human is clearly separate from the horse. Others used it as a cyborg, a man-machine amalgam, where the AI tasks were inextricably woven into the team’s human efforts. No team took the AI’s outputs at face value, but all outperformed the non-AI supported teams. AI is most powerful when its human partners do not abdicate their responsibility to be the human in the loop.

Start With What You Know

The best place to start is in the area you know best—your own expertise. By introducing AI into your work, you can enhance your outcomes and be more efficient because your subject-matter expertise becomes a check on the jagged frontier. Mollick writes, “In field after field, we are finding that a human working with an AI co-intelligence outperforms all but the best humans working without an AI.” And this isn’t about replacement—it’s about augmentation. Knowledge jobs are rarely unidimensional: every role is a portfolio of different responsibilities, some you enjoy and others you dread. By successfully incorporating AI to supplant or enhance portions of what you do, it opens the opportunity for more meaningful work in the areas where human judgment, critical thinking and creativity excel. And team leaders take note—a workplace culture that reframes AI as an opportunity for experimentation and more meaningful work may help you attract and retain better people.

Get Good At Giving Feedback

AI isn’t great without guidance. The simplest queries result in the weakest responses. You need a more complex approach to engaging AI: telling it what perspective to take to answer your question, and giving it examples of what you seek. In one anecdote, Mollick posed the same question several times, asking the LLM to debate him in one iteration, teach him in another. In the former example, the LLM was combative and provocative, never agreeing with him, but testing and strengthening his views. In the latter case, the LLM was thoughtful, reflective and able to contextualize the implications of his question. The more detail, the better the outcome. Using “chain of thought” queries, you can break your request into transparent steps. First, outline the problem, second, write the first line of each paragraph explaining it, third, write the analysis, fourth, check for inconsistencies, and so forth. This helps you surface the logic the LLM is leveraging, clarifying the process and helping you to build on its outputs. Over time, AIs will better understand the motives of their human partners and detailed prompts may become superfluous. In the meantime, feedback trains the AI to produce higher quality, more cogent responses.

In Co-Intelligence, Ethan Mollick depicts the very world Frankenstein’s Creature craved: a satisfying relationship with human beings. With a collaborative approach, we can bring the best of human and artificial intelligence to unimaginable new frontiers. Co-Intelligence is a terrific read, full of practical examples to augment or support your work—as long as you use AI with care and a critical eye. But Mollick’s no Pollyanna. He is sanguine about the future, clear that we don't yet know where generative AI will go, and whether its outputs will ever align with a set of human values we can trust. Learning to collaborate with AI helps you supercharge your performance and broaden your intelligence, when you stay alert to its pitfalls. But like every relationship, it will take work.

First published on

Image Credit:
For more insights, sign up for our newsletter
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Continue reading