Co‑Intelligence: Living and Working With AI
This 15-minute audio course introduces Co-Intelligence: the collaborative partnership between humans and AI. The first lecture defines the concept, arguing that the future is not humans versus machines but humans plus machines becoming something greater together. It frames AI as an augmenting partner rather than a replacement, using everyday examples of AI assistance and prompting listeners to reflect on their own relationship with technology. The second lecture grounds the idea in a concrete object: a smart coffee mug. Through a dialogue between a narrator and the mug's AI voice, the lecture walks through the sensing, feedback, and suggestion cycle that turns an ordinary tool into a collaborative partner. It slows down at the moment the mug predicts a needed break before the user feels fatigue, showing how sensors, real-time data collection, and predictive assistance support user wellbeing in daily life. The final lecture tackles the paradoxes surrounding AI's impact on human creativity. It opens with two contradictory claims—that AI will replace human jobs and that it will create new opportunities—and holds both in tension while investigating the central question: will AI ultimately expand human creativity? Following the evidence, it examines the fear of loss as a false lead, then uncovers AI-augmented creative tools in design and music as the true lead. The conclusion offers a framework for evaluating AI as a co-creative partner and shows that, when we ask the right questions, Co-Intelligence amplifies rather than replaces human creative capacity.
Welcome to Co‑Intelligence: Re‑imagining Human‑AI Partnership
Welcome to Co‑Intelligence: Re‑imagining Human‑AI Partnership
A single consultant sits down at a laptop, opens a chat window, and types a question she could have answered herself in an hour. She's one of 758 consultants at Boston Consulting Group who were assigned to work with GPT-4 as part of a field experiment, and some of them had no AI access at all. What happened to her output, Billy, and to thousands of workers like her, is the story behind an idea called co-intelligence. Because when researchers compared the groups, the people working alongside AI didn't just get a little better. They finished twelve percent more tasks, over a quarter faster, and their work was rated more than forty percent higher in quality. Those numbers surprised me the first time I saw them, and they should change how you think about your own workday.
So what does co-intelligence actually mean, moment to moment? It means treating AI less like a search engine you occasionally poke and more like a colleague you think with. The consultants in that experiment didn't hand everything over. They chose how to divide the work, and the researchers watched two distinct patterns emerge. They called the first pattern Centaurs, after the half-human, half-horse creature of myth, because these people split tasks cleanly, keeping the judgment for themselves and delegating the rest. The second pattern they called Cyborgs, people who wove the AI into every step of their workflow, passing work back and forth almost continuously. Neither pattern is the one right way. They're options, and the choice between them is yours.
Now, the twist in this study is what convinced me the story isn't simply "AI makes everything better." The researchers described AI's abilities as a jagged technological frontier, meaning performance can differ sharply across tasks that look similarly difficult. Think of it like a cliff line that jumps around unpredictably. Two assignments might seem equally hard to you, and the AI will breeze through one and fail badly at the other. On eighteen tasks inside that frontier, the AI-assisted consultants crushed it. But on one task outside the frontier, they were nineteen percentage points less likely to reach a correct answer than colleagues working with no AI at all. Even a prompt-engineering overview didn't remove that risk. The practical lesson: judge AI by task, not by reputation. A success on one assignment tells you nothing about the similar-looking one next door.
And there's a second finding here that I, no, wait, this is the part that gets me. The gains weren't spread evenly, and they didn't go where most people would predict. In the consulting experiment, below-average performers improved by forty-three percent relative to their baseline, while the top performers improved by seventeen. A separate workplace study of more than five thousand customer-support agents found something similar: access to a generative AI assistant raised issues resolved per hour by fourteen percent on average, but novices gained around thirty-four percent while experienced agents saw minimal impact. The AI was essentially spreading the practices of the best workers to the less experienced, like an apprentice quietly absorbing a master's habits. That study also reported improved customer sentiment and better employee retention, which is, yeah, which is remarkable when you consider most workplace tools promise none of that.
So what's the small shift for you this week, Billy? It's not learning fifty prompt tricks. It's deciding, task by task, whether AI belongs inside this work at all. Because the frontier is jagged, the skill that pays is checking the output against the task, and choosing your pattern, Centaur or Cyborg, deliberately rather than by accident. The consultants who thrived weren't the ones who trusted the AI most or least. They were the ones who stayed in the loop, kept their judgment, and let the machine do what it demonstrably does well. That's co-intelligence in one breath: a partner you collaborate with, verify, and never blindly obey. Start with one small experiment this week, maybe ask AI to critique something you wrote. Watch where it helps and where it quietly fails. That map, drawn from your own work, is worth more than any statistic I could quote you.
The Smart Mug: Tracing Co‑Intelligence Through a Simple Object
Speaker 1
Okay, so you want to explain co-intelligence to me with a coffee mug? I picture a scientist's desk covered in papers, and you hand me a mug.
Speaker 2
Hand me the mug, then. Ember makes these, and what's inside is a built-in heater plus precision sensors. Somebody picks a temperature, and the mug works to hold the drink there.
Speaker 1
So it's a thermostat with a handle. Where's the intelligence in that?
Speaker 2
Now, that's the right suspicion, and it's half right. A basic feedback loop runs the whole thing: measure the drink's temperature, compare it to the chosen setting, and use the heater to close the gap.
Speaker 1
Measure, compare, adjust. That's my home thermostat, though.
Speaker 2
The difference is who supplies what. The person supplies the preference, the desired drinking temperature. The device carries out ongoing maintenance. That division of roles is the whole co-intelligence idea, shrunk down small.
Speaker 1
Okay, but a thermostat also holds a setpoint I chose. Why is the mug a partnership and the thermostat just a gadget?
Speaker 2
Because the mug follows you. Lift it off the charging coaster and the heater and sensors keep working on battery, over three hours of warmth with the sliding lid on. It maintains the goal away from its dock, in the middle of your morning, not just where it's plugged in.
Speaker 1
So the coaster is the—
Speaker 2
The power source, yeah. The battery is what lets the partnership travel.
Speaker 1
A mug that, that follows me around keeping promises. I'll admit that's a nicer picture.
Speaker 2
And it goes further. Ember says the sensors can detect liquid level too, not just temperature, and convection currents help spread the heat evenly. It's sensing more than it advertises.
Speaker 1
It knows how full it is? That's more data than I expected from a cup.
Speaker 2
Which opens the door to the app. It stores temperature presets, notifies you when a drink reaches the chosen temperature, pairs multiple mugs, even sets the LED color. But here's the part I like: Mug 3 can be set without the app. The app is optional.
Speaker 1
Why does optional matter so much?
Speaker 2
Because the human keeps control of the basic interaction. Think of it like a colleague who can work alone but checks in when you want them to. Augmentation, not autonomous intelligence. The user still chooses what to drink and what temperature is desirable.
Speaker 1
So the mug helps with the bounded operation, and I keep judgment over the goal. But does that actually work? That's the claim I want to test.
Speaker 2
Fair question, and the evidence is mixed. In 2024, Michelle Vaccaro, Abdullah Almaatouq, and Thomas Malone ran a meta-analysis, a hundred and six experiments, and found human‑AI combinations performed, on average, worse than the better of humans or AI working alone.
Speaker 1
Worse?
Speaker 2
On average, yes. Adding a partner doesn't guarantee better results. Combinations lost on decision tasks; the authors saw more promise in content creation.
Speaker 1
So I should be skeptical of my own mug argument, then.
Speaker 2
I agree that the mug is a straightforward case. The task is bounded, the failure mode is lukewarm coffee, not a bad decision. Research on human‑AI collaboration emphasizes outcomes depend on the task and how the work is divided. The mug divides the work well.
Speaker 1
Okay, but the mug quietly collects data about my habits. Who's watching that?
Speaker 2
That's the shadow side. The Federal Trade Commission actually has guidance here: build security into connected devices from the outset, not as an afterthought, and practice data minimization, limiting what personal information a product collects and how long it keeps it.
Speaker 1
Data minimization meaning, what, collect less on purpose?
Speaker 2
Exactly. And communicate clearly with consumers about data use, especially when collection goes beyond their reasonable expectations. A mug pairing with a phone is an Internet of Things device, a sensor exchanging information over the internet. Most people don't expect their cup to have a data footprint.
Speaker 1
So my morning coffee has a privacy question attached. Is there a bigger idea behind putting computers in mugs, or is this just gadgets?
Speaker 2
There's a lineage. Back in 1991, Mark Weiser at Xerox PARC described computing woven into everyday surroundings, technology receding into the background. He argued computers could become so integrated into daily life that people would scarcely notice them.
Speaker 1
So the mug is, what's the phrase, ubiquitous computing made drinkable.
Speaker 2
Made drinkable. And maybe that's why it teaches co‑intelligence better than a chatbot. There's no conversation to perform, just a quiet partner holding up its half of a shared goal.
Speaker 1
The person supplies the preference; the machine maintains it. That's the mug in one line, isn't it.
Speaker 2
it doesn't pick your temperature for you. The next time a tool feels collaborative, that's the test I'd apply, who's holding the goal, and who's doing the holding.
Speaker 1
A scientist's desk covered in papers, and the clearest example was in his hand the whole time.
Speaker 2
Usually is.
Paradox or Promise? Investigating AI’s Impact on Human Creativity
Two things sit in front of us, and they cannot both be true. In 2024, Anil Doshi at University College London and Oliver Hauser at the University of Exeter ran an experiment on generative AI and creative writing, and people with AI help wrote stories judged more novel, more useful, better written, more enjoyable. At the same time, in the same experiment, the AI‑assisted stories were more similar to one another. Individual writers got better while the collective output got more alike. Which is it, then? Does AI expand creativity, or quietly shrink it? This has been considered for a while, and it will not be resolved early. I'm going to work it like an investigation, clue by clue, because the answer genuinely isn't settled.
So let's look at the crime scene, so to speak. The researchers recruited three hundred participants who were not professional writers, and asked each to write an eight‑sentence micro‑story for a young‑adult audience. Some wrote alone. Some got a single AI‑generated story idea. Some could pick from as many as five. Independent evaluators then scored the results without knowing who'd had help. The AI‑assisted stories scored higher on novelty and usefulness, and the gains were larger for participants whose baseline creativity scores were lower. Among less‑creative writers, five AI ideas came with roughly a ten percent bump in novelty and usefulness. Think of it like a rising tide that lifts the shallow boats highest.
Now, a natural lead here is the standardized creativity tests. In 2024, Kent Hubert, Kim Awa, and Darya Zabelina compared GPT‑4 with 151 people on divergent‑thinking tasks, things like suggesting alternative uses for objects, and the AI scored higher on originality and elaboration. Earlier, Mika Koivisto and Simone Grassini compared three chatbots with 256 people on an alternative‑uses task, and the chatbots won on average. But this is the lead that turns out wrong, or at least incomplete. Divergent‑thinking tests measure one narrow kind of idea generation; they don't establish human‑like creativity across everything creative people do. And in the Koivisto and Grassini data, the very best human ideas matched or beat the chatbots. The machines win on average; the humans still hold the ceiling. That distinction matters more than the headlines suggested.
So why does the same tool amplify one person and flatten another? A clue from poetry research. When participants were assigned an editor role while working with generative AI, their poems were rated less creative than poems written with no AI at all. But the deficit disappeared when the same people were positioned as co‑creators, and the effect tracked their creative self‑efficacy, their belief in their own ability to produce creative work. The tool was identical. The framing changed the outcome. Which means collaboration isn't automatic. You have to actually co‑develop, not just accept.
If you asked me to wager, here's where I'd put money. On professional writing tasks, Shakked Noy and Whitney Zhang found that ChatGPT cut completion time by forty percent and raised assessed quality by eighteen percent, with the biggest gains for weaker writers. I'd bet confidently on that pattern repeating: AI lifts the floor. But would I bet on AI raising the ceiling of what's creatively possible? No. Not even money. A study found no general ChatGPT advantage on a divergent‑thinking task, and one low‑creativity prompt believed to come from ChatGPT was actually associated with lower scores. Researchers called the whole tension a social dilemma: individuals benefit from AI ideas even as widespread use risks narrowing the range of content produced. Both things can be true at once. That's the paradox, and I, I can't argue it away.
What does this mean for your week, Billy? When you sit down with AI tomorrow, ask it for options instead of answers. Have it generate five directions for the thing you're making, then discard three and focus on the strange one, the one you might not have said out loud. Keep the judgment about which idea to pursue with you, because that division, options from the machine, choice from the human, is exactly what the research points to. And remember, the similarity effect in the story experiment showed up even when people saw just one AI idea, so variety has to be something you deliberately protect. So, does AI ultimately expand human creativity? Honestly, the jury is still out, and reasonable people land on both sides. The tool amplifies whoever is holding it. What it amplifies depends on how you show up: as an editor outsourcing judgment, or a co‑creator keeping it. That choice is yours each session, and nobody has settled it for you.