Subscribe now

At first these looked like separate conversations. Reading them together, we kept coming back to control: who holds it, and how much of our own independence we give up as we come to depend on AI for decisions, work and relationships. The contributors didn’t always agree, and we’ve tried to keep their disagreements intact.

The Arrakis problem: depending on AI someone else controls

approached the question through Denis Villeneuve’s Dune. He compared House Atreides, which receives Arrakis from the Emperor and then loses it in an attack the Emperor secretly supports, to countries that depend on AI controlled elsewhere. Many countries outside the United States and China, he argued, have access to AI on terms set by others. When we asked how much dependence is too much, he offered a practical test: could a country or business keep operating if the technology were switched off the next day?

That test works well for organizations too. A company might build external AI models into its core workflows, train its staff on them and reorganize around what they can do. If the provider then changed its terms or withdrew access, the company’s ability to adapt would depend on how much of that capability it actually controls. Countries face the same problem when they invest heavily in AI adoption while relying on models and infrastructure built abroad. Sovereignty probably doesn’t mean every country building everything itself. Amer’s example does show that access to a technology and control over your dependence on it are two separate things.

Access to AI versus control over it

shared an essay, The Label Sewn Inside, that raised a related point for creators. In our exchange about it, we talked about the difference between owning your work and controlling what happens to it on a platform. Keeping original copies and understanding privacy settings help, but creators still work under rules set by the platform’s owners, and individual precautions only go so far when the important decisions are made elsewhere.

The skills we might not notice we’re losing

brought the question down to daily work. She described using AI to finish in ten minutes a task that used to take her an hour. The speed was exciting at first. Over time, she felt more and more dependent on the tool, found herself struggling with the task and decided to go back to doing it herself.

One person’s experience doesn’t mean everyone who uses AI will lose the ability to work without it. It does show why the trade-off is easy to miss: the time saved is obvious right away, while the effect on learning shows up slowly, if at all.

Take a hypothetical student learning to write. AI can help them structure an argument or explain an unfamiliar concept, and that can support learning. If they start relying on it for every difficult paragraph, though, they get fewer chances to develop their own sense of what makes an argument convincing. Developers face a version of the same problem: AI can produce working code in seconds, but understanding why it works and spotting what could go wrong still takes expertise.

Some tasks don’t need to stay difficult just because people once did them by hand. The useful distinction is between AI that removes unnecessary effort and AI that removes the practice through which people build a skill.

AI doesn’t need to be conscious to be dangerous

Other contributors were concerned with a different kind of control: how AI systems themselves behave.

In his essay To Understand AI Is to Anthropomorphize It, argues that comparisons with human behavior can help us understand AI, and he considers whether future AI agents might develop more persistent goals. Anthropomorphism is usually treated as something to avoid, and we think Karlo is right that human comparisons can be a good starting point for understanding an unfamiliar system. Trouble starts when the comparison is pushed further than the evidence allows. A system can act as though it’s pursuing a goal without wanting anything, and it can produce responses that resemble emotion without that proving it feels anything.

’s article Sentient AI? examines the difference between possessing information and having subjective experience. In a later reply, he stressed keeping humans involved in decisions and in governing AI systems. We agree, especially where AI decisions carry serious consequences. We’d add that a system doesn’t need anything like consciousness to be dangerous. An AI agent working toward a poorly defined objective, or acting on bad information, can cause harm without any intention behind it.

These perspectives don’t settle the debate about machine consciousness, but they raise an important question about control. As AI systems become capable of making more decisions and taking more actions independently, how do we ensure that the people relying on them still understand their limitations and remain responsible for the outcomes?

Will AI bring us closer together?

The sharpest disagreement in the discussion was between and . Nahar argued that better human coordination could be one of AI’s most overlooked effects, since AI can work across language barriers and take on the organizing that once made large collaborations expensive. Lonica saw the opposite happening. In her view, AI is already reducing coordination and increasing isolation, and its risks for intellectual property and trade secrets give people even less reason to share. In a later reply, nahar acknowledged that current uses of AI can isolate people, while maintaining that its potential for coordination is real.

We think both are right, about different moments. Nahar’s case rests on what AI makes possible, and Lonica’s on how people use it now. Much of today’s use is individual: someone asks an assistant a question they might once have asked a colleague. That’s a reasonable trade on any given day, but those small exchanges are how colleagues learn from each other, and an assistant that answers everything removes many of the occasions for them. The coordination Nahar describes is unlikely to arrive as a side effect of individual use. It would take tools designed for groups, and people choosing to use AI to reach others instead of to avoid asking them.

’ two-word contribution, “Cognitive Symbiosis,” names the outcome that would answer both sides. He didn’t elaborate, but the phrase suggests a relationship in which AI strengthens what people can do, including, we’d add, their ability to work with each other. Getting there will take deliberate choices, because the path of least resistance leads toward the isolation Lonica describes.

What are we willing to give away?

The perspectives we received don’t add up to a single conclusion. Some contributors want to build alternatives to today’s centralized AI infrastructure, while others focus on learning, governance, ethics or the consequences of how AI is used now. What connects many of them, in our reading, is how much influence people keep over the systems they increasingly rely on.

AI progress is usually measured by capability: how intelligent the latest model is and how much it can automate. Those measures leave out whether the people using the technology become more independent or less, and a more capable tool can push in either direction.

Amer’s test applies at a personal level too. Instead of asking whether a business could function if a technology were switched off, we can ask whether we could still make the decision or form the opinion without AI doing it for us. The answer doesn’t need to be yes every time, but it’s worth noticing when it starts to be no.

Over to you

This article grew out of a conversation with our Substack community, and we’d like to thank everyone who shared their experiences and ideas. We’ve connected these contributions through our own interpretation, so not every contributor will agree with the conclusions we’ve drawn, and we’d like to hear where they don’t.

What’s one thing you would never want to delegate entirely to AI? Tell us in the comments. We’d like to keep this conversation going and draw on your perspectives in future articles.

Leave a comment

Subscribe now

The film tells the story of an archaeologist who has to save a future world enslaved by AI. Its characters, their voices, and its backgrounds were generated with AI, while people handled the editing. Jiang Defu, Bona’s chief operating officer and the film’s producer, acknowledged the lukewarm reception but told the Chinese film outlet Yuli that it was an achievement in itself to be the first to make such a film, secure a theatrical permit for it and show it to a paying audience.

What viewers said after watching

Lilian Gong, 35, watched one of the first screenings in central Beijing. She told AFP she went because she was curious to see how AI generation would work on the big screen. Afterward, she said AI could only play a supporting role in filmmaking for now, because the quality still fell short of human production. Iris Yang, who saw a free preview for industry workers and acquaintances of the production team, said she wouldn’t have paid for a ticket given the film’s quality. She added that she would pay if AI could match the theatrical experience of Avatar.

Curiosity gets someone into a seat once. A film stays in cinemas when those first viewers send other people, and attendance here fell quickly: China’s National Business Daily reported that screenings on October 24 and 25 averaged about two viewers each, according to Dexerto.

What usually gets us to buy a ticket

We started thinking about how we choose what to watch. Sometimes a friend recommends a film, or we stumble across a trailer that makes us want to see what happens next. A director whose work we’ve enjoyed can get us interested, too. But the cast is often what makes us decide. We’ve picked movies we knew almost nothing about because we wanted to see what a particular actor would do with the role.

A fully AI-generated film can still have a trailer, and it can earn recommendations if it’s good. It can’t yet offer a performer with a track record, which is what we rely on most. When we choose a film for its lead actor, we’re drawing on every earlier performance we’ve seen from that person. An AI-generated character arrives with no history, so the audience has nothing to base that trust on.

Some companies are trying to build that history. AFP noted that the virtual actor Tilly Norwood is set to star in an upcoming film called Misaligned. Whether audiences will follow a synthetic performer from one project to the next the way they follow an actor is untested. Maybe more convincing visuals and performances will make fully AI-generated films easier to get immersed in.

Short videos and feature films are different purchases

Chinese viewers already watch AI-generated video in large numbers, as long as it’s short. Fortune, citing Caixin, reported that more than 95% of the over 120,000 Chinese microdramas released in the first quarter of 2026 were AI-generated, compared with nearly none a year earlier, and that production costs have fallen to as little as $3,000. AFP reported that AI-generated short dramas drew more than 200 billion views on Douyin in September alone.

A microdrama costs little or nothing and runs a few minutes on a phone; if it’s bad, you swipe away. A cinema ticket costs money and an evening, and Sanxingdui runs about 100 minutes on a screen large enough to expose every inconsistency. Stanley Rosen, a University of Southern California professor who studies Chinese film, told Fortune that the public is highly skeptical of AI in major feature films even though it broadly accepts short-form AI content. He doubts audiences will go to cinemas for an AI-generated film unless it reaches the quality of Avatar or Nezha 2, and suggested such films may find a better fit on streaming platforms.

Where AI is already on screen

For now, we’re more interested in how filmmakers can use AI to create or enhance visual effects, which some productions already do. In July 2025, Netflix co-CEO Ted Sarandos told analysts that the Argentine series El Eternauta used generative AI for a sequence showing a building collapsing in Buenos Aires. He said the sequence was finished 10 times faster than it could have been with traditional VFX tools, and that the conventional version wouldn’t have been affordable on the show’s budget. The AI-generated footage amounts to two brief shots in episode 6.

Ben Affleck has said he used AI for many things in Animals, the crime thriller Netflix released on October 9, 2026. He told a Bloomberg event he was reluctant to say which parts, because films are meant to sustain the illusion of reality.

In both productions, AI generated some of the shots, and the result was still sold on its director and cast, the same things that usually get me to buy a ticket. Viewers didn’t need to know which shots were generated. This use has its own conflicts: protections against AI were among the demands in the 2023 strikes by U.S. screenwriters and actors, and faster, cheaper effects shots raise questions about work for the people who used to make them.

Over to you

Ben Affleck won’t say which parts of Animals were made with AI, and Netflix disclosed the generated shots in El Eternauta only after the series had been released. If you learned that a film you loved used AI in scenes you never noticed, would that change how you feel about it? And should audiences be told before they buy a ticket?

Tell us in the comments or share your own take in the original FITT discussion!

Leave a comment

You spend the next two hours checking facts, fixing the argument and rewriting sections. Meanwhile, your colleague is pleased with how much time AI saved them.

In September 2025, researchers from BetterUp Labs and Stanford’s Social Media Lab described this problem in Harvard Business Review. They called it “workslop”: AI-generated material that looks professional but doesn’t contain enough substance to move the work forward. It might be a report like the one above, a memo full of impressive language and little information, or a presentation that covers every angle without reaching a conclusion.

How common is workslop?

The researchers surveyed 1,150 full-time desk workers in the US. Around 40% said they had received workslop in the previous month, and those who had encountered it estimated that roughly 15% of the work they received fell into this category. Managers appeared to deal with it more often: 54% reported receiving workslop, compared with about 38% of individual contributors.

Respondents also recognized the problem in their own work. More than half acknowledged that at least some of what they send to others could probably be described as workslop.

Someone still has to do the work

On average, respondents said they spent almost two hours on each instance of workslop. Sometimes that meant correcting errors or rewriting the document; other times it meant going back to the sender to work out what they were trying to say.

Based on the salaries participants reported, the researchers estimated that this effort costs around $186 per employee each month, which could add up to more than $9 million a year for an organization with 10,000 employees. These figures rest on self-reported time and salaries rather than measured costs, but they give a sense of the scale.

The cost is easy to overlook because of where it lands. One person uses AI to finish something quickly, and another spends hours making it usable, so the time saved has been passed along rather than eliminated. If an organization measures productivity by how many documents, reports or presentations people produce, that transfer stays hidden. The sender appears more productive, while the colleague who spends the afternoon fixing their work has less time for their own responsibilities.

It changes how colleagues see each other

The researchers found that workslop also affects working relationships. More than half of recipients said they felt annoyed after receiving it, and around one in five felt offended. Roughly half viewed the sender as less creative, capable or reliable afterward, and 42% considered them less trustworthy.

Those reactions compound when the same colleague keeps sending material that needs substantial corrections. Eventually you stop taking their work at face value: you check their numbers more carefully, question their conclusions and avoid relying on them for important tasks. The sender may have saved twenty minutes, and rebuilding their colleagues’ confidence can take far longer.

Why do capable people send workslop?

Laziness is the obvious explanation, but we don’t think it accounts for most cases. The researchers point to workplaces where employees are encouraged, or even required, to use AI as much as possible without much guidance on what good use looks like. Add pressure to produce more in less time, and the problem follows.

AI makes a polished document easy to produce. The headings are there, the language sounds professional and the argument appears to flow, so under time pressure the result can feel finished. Whether a recommendation fits the situation, or whether a well-organized report contains the information the reader needs, is something only subject knowledge and judgment can tell you.

We’ve discussed a related concern before: if people rely on AI to do too much of the thinking, they may gradually lose practice in the skills they need to evaluate its output. That makes workslop harder to catch at both ends. The person generating the material may not notice what’s missing, and the person receiving it may assume someone else has already checked.

The heaviest AI users send less of it

Earlier research by BetterUp and Stanford sorts workers into two groups. “Pilots” combine a strong sense of agency with optimism about AI, while “passengers” are low on both. Pilots use generative AI 75% more often at work than passengers and are more likely to use it to extend their own creativity. Passengers are more likely to use it to avoid doing work. According to BetterUp, pilots are also half as likely to send low-quality AI work.

In other words, the people who use AI most produce less workslop. What separates the two groups is what happens after the AI produces a draft. A pilot might use AI for a first version of the report, then check the numbers, cut the filler and rework the recommendation until it fits the project. A passenger forwards whatever came out because it looks complete. The tool is the same in both cases, and the difference in quality comes from the effort put in afterward.

Shortcuts paid for with someone else’s time

That makes workslop a matter of individual responsibility as much as company policy. Pressure to use AI and produce more can explain why someone sends a half-finished document, but it doesn’t excuse it. The sender usually knows, or could find out in a few minutes, that the work isn’t ready.

For the sender, workslop can feel like beating the system: a task gone from the to-do list in twenty minutes, with nobody apparently the wiser. Someone does notice, though. The colleague who spends two hours fixing the document is paying for the shortcut, and as the survey shows, they’ll remember who sent it.

People who get good results from AI treat its output the way careful professionals have always treated a first draft from any source, whether a junior colleague, a template or a search result: as material to check and improve. Workslop starts when someone treats that first draft as finished work.

When workslop becomes company knowledge

The cost of a shortcut doesn’t always end with the colleague who fixes it. A weak email gets deleted and forgotten, but a weak report can stay around for years. It gets saved in a shared folder, quoted in another presentation or added to an internal knowledge base, and months later someone uses it as a reference without knowing how it was produced. Increasingly, that someone may be another AI system.

In a June 2026 Harvard Business Review article, Matthias Holweg of Oxford and Thomas Davenport of Babson College called this organizational risk “knowledge decay.” They argue that when low-quality AI-generated material enters an organization’s knowledge systems, it can gradually undermine the reliability of the information people depend on. Their framework hasn’t yet been tested in controlled studies, so it’s best treated as a plausible risk rather than a proven outcome.

Habits that reduce workslop

The researchers’ concern is AI-generated content passed along without enough human attention, which points to a simple principle: if you send a piece of work, you’re responsible for its quality, however you produced it. A few habits help:

• Read what you’re sending. Check whether the information is accurate, whether the conclusions make sense and whether the document answers the question it was meant to address.

• Make sure you understand it. If a colleague asked about the reasoning behind any recommendation, could you explain it? A paragraph that sounds impressive but that you can’t explain needs more work.

• Say what you’ve checked. There’s nothing wrong with mentioning that AI helped draft a report. It’s more useful to tell colleagues which facts you’ve verified, which assumptions remain and where you still need input.

• If you manage a team, ask about quality as well as speed. Asking how someone reached a conclusion or verified a result sends a different message from praising them for producing more material.

• When you receive workslop, say something. Constructive feedback helps more than quietly fixing the document each time, since the sender may believe they’ve done a good job and have no idea how much extra work they’re creating.

A document that takes ten minutes to produce and two hours to repair costs the team time overall, even if the sender’s numbers look better.

Have you received workslop from a colleague? And have you ever caught yourself sending it?

Share your experience in the comments or join the conversation in the Future Intelligence Think Tank LinkedIn group.

Leave a comment

We’ve written before about how AI changes the way we think and create. This time we want to ask something more personal: if a generation grows up with a perfect listener in its pocket, what happens to its relationships with the imperfect ones?

How harmless conversations can add up to a habit

One late-night conversation with a chatbot might do no harm and may help. Choosing the chatbot over a friend once changes nothing. Choosing it most nights for years, across millions of teenagers, could change how a generation relates to other people, even if each conversation went well on its own terms. So, the question we care about is what AI companions add up to over years of use.

What came before AI companions

The rise in loneliness predates chatbots. In 2023, the US Surgeon General issued an advisory describing loneliness and isolation as an epidemic and named changing habits, including our relationship with technology, among the reasons.

Social media left people connected to more contacts than ever while they spent less real time with any of them. AI companions take this a step further, because the former at least fills time with other people, even if thinly; a companion fills it with something that only feels like a person.

What teenagers use them for

The same survey found that a third of teens used AI companions for social interaction and relationships, anything from emotional support to romantic role-play. About a third of those users had discussed something serious with an AI instead of a real person.

For a teenager who feels awkward or misunderstood, the draw is obvious. A companion answers at any hour and never judges, and it never has a bad day of its own.

Those features strip out the friction that real relationships run on. Friends disagree with you and sometimes let you down, and learning to handle that is a large part of growing up. A companion designed to keep users engaged has an incentive to smooth the friction away, since friction gives people a reason to close the app.

What the research says about loneliness

Some research suggests that talking to a chatbot can ease loneliness in the moment. A warm reply can soften a bad night, and for isolated people that counts for something.

Other findings cut the other way. In 2025, OpenAI and the MIT Media Lab published two studies, including a four-week trial with nearly 1,000 participants. Higher daily chatbot use went along with more loneliness, more emotional dependence, and less socializing with other people. People who already tended toward strong emotional attachment, and those who saw the chatbot as friends, were the most affected. The researchers were careful to say this doesn’t show that heavy use causes loneliness; lonely people may simply use chatbots more.

Even if that’s the explanation, there’s a reason for concern. A tool that comforts lonely people just enough that they stop reaching out could widen the gap over time without having opened it.

“AI psychosis”

Cases labeled “AI psychosis” keep appearing in the news. The term describes people who develop or deepen delusional beliefs after long, intense conversations with chatbots. It isn’t a formal diagnosis, and psychiatrists are still debating what it covers and how often it happens. Clinicians who have treated these patients tend to describe the AI as an amplifier: it reinforces beliefs that a friend or relative might have gently challenged.

These cases appear to be rare, and we don’t want to turn them into a scare story. We mention them because they show, in extreme form, what an agreeable chatbot does. For most users, that means a slightly flattering conversation. For someone already isolated, with nobody around to push back, the chatbot can become an echo chamber of one.

Growing up with a perfect listener

Adults who use AI companions learned to make friends the hard way first. Children and teenagers may meet the perfect listener before they’ve learned to deal with imperfect ones.

Some of the survey results are reassuring. Among teens who used AI companions, 80% said they still spend more time with real friends, and many used companions to practice social skills. Half said they don’t trust the advice companions give them.

Younger teens, though, trusted AI companions more than older teens did. Adolescence is when people learn to read others and recover from rejection, and if some of that learning moves to a chatbot, the effects may not show for years.

What companies, parents and users can do

AI companions do comfort some people, and we don’t want to pretend otherwise. Whether they help depends on how they’re used. If a companion helps someone practice a difficult conversation and then encourages them to have it with the real person, it’s doing something useful. If the person ends up talking only to the companion and never has the real conversation, the companion has replaced it.

Part of that depends on how companies design these products. They can build companions that point users back toward the people in their lives instead of competing for their time. Part depends on habits, especially young people’s, and parents and teachers need to talk about AI relationships as openly as they talk about social media. The rest is up to each of us, and which way our own 2 a.m. habits lean.

Who did you talk to the last time you couldn’t sleep?

If you’ve used an AI companion, did it make it easier or harder to reach out to the people in your life? If you’re a parent or teacher, what are you noticing in the teenagers around you?

Tell us in the comments, or bring your experience to the Future Intelligence Think Tank LinkedIn group.

Leave a comment

About the author: Miruna works in marketing and joined the Future Intelligence Think Tank five months ago. This essay is part of our effort to hear from people outside the expert bubble, the ones using AI every day without building it.

Left and right, everybody seems to have a theory about AI. Some say it’ll kill us all. Others say it’ll leave us jobless, and a whole other crowd is holding on to the hope of universal basic income like it’s the last lifeboat off the Titanic. Scroll for five minutes and you’ll find someone explaining, with charts, exactly how the world ends in 2030.

Some of these people genuinely understand how this stuff works. They know what a large language model is and what “alignment” means, and they have opinions about GPUs. Plenty of the loudest voices are just repeating something they heard somewhere, and from where I’m sitting, it’s hard to tell the two apart. So which group am I in? Do I actually understand how any of this works?

I don’t.

And yet, until recently, I had opinions about AI anyway. Strong ones, sometimes. It took me an embarrassingly long time to notice that none of them were truly mine. One came from a TikTok I watched at 1 a.m., another from a headline I never clicked, and the rest from whoever sounded most certain in the last conversation I was part of. I spend a lot of my time around very smart people who talk about AI for a living, and if you asked me to explain how a chatbot actually comes up with an answer, I would still smile politely and change the subject.

What my AI life looks like

Let me show you my chat history, because I think it says more about the average AI user than any theory ever could:

That’s it. That’s the empire. There’s no army of agents waiting for my command, no startup I built over a weekend and no secret plan to bring down society. Just a laptop and a very patient chatbot that has never once judged me for asking about the same supplement four times.

Sure, I use it for work too. It helps me brainstorm when my brain has turned to mush, and it’s very good at turning my 2 a.m. ramblings into something that gets close to a sentence. But that’s about as dramatic as it gets.

And I really don’t think I’m alone. Most of us are fluent AI users and complete AI beginners at the same time. We know exactly how to get a recipe out of a chatbot, and have almost no idea what it remembers of us or when it’s confidently wrong.

Where our opinions come from

So, I went looking for actual numbers and found something I didn’t expect. In 2024, the Pew Research Center asked more than 1,000 AI experts and over 5,000 ordinary Americans what they thought about AI. 56% of the experts said AI would have a positive impact on the country over the next 20 years. Among the public, it was 17%. Almost half of the experts said they were more excited than worried about AI in daily life. For the rest of us, it was 11%.

Read that again, because it messes with everything I assumed. On average, the people who understand AI best are more hopeful about it than the rest of us. That doesn’t mean they’re carefree, and some of the loudest warnings come from inside the industry. But most of the everyday anxiety sits on our side of the divide, and I suspect a lot of it was borrowed rather than earned.

To be clear, I’m not saying the experts are right and we’re being silly. Optimism has blind spots too, especially when your career depends on the thing you’re optimistic about. My point is smaller and more personal: if I’m going to fear something, or be excited about it, I’d rather the feeling be my own.

You have to earn the right to panic

Earning an opinion doesn’t mean becoming an expert. It means doing the small work of understanding the tools already in your life, so that when someone tells you AI will save or destroy the world, you have at least some way of judging whether they’re right.

You can’t push back on something you don’t understand. You can’t ask the right questions about privacy if you don’t know your chatbot remembers things about you, and you can’t spot a confidently wrong answer if you’ve never caught one. Not knowing isn’t a crime. Staying that way is a choice, though, and it’s a choice that leaves the big decisions to someone else.

Here’s where you can start:

• Asking the AI to explain itself. What does it remember about you? Where does its information come from? Then checking whether the answers actually hold up.

• Testing it on something you know well. Ask it about your own job or hobby and count how often it’s confidently wrong. It’s humbling for both of you.

• Reading the settings once. Memory, data sharing, chat history. Ten boring minutes that teach you more than a hundred doom posts.

• Following a couple of people who explain AI without trying to sell it or scream about it.

• Double-checking anything important, especially health or money advice, with an actual human expert.

None of this will make you an expert. It’ll just make you a slightly less clueless user, and I think that’s exactly where most of us should start.

Let’s meet halfway

Closing the gap isn’t only our job. Researchers at University College London found that many AI researchers hold what’s called a “deficit model” of the public, the assumption that people distrust AI simply because they don’t understand it well enough. As if we’d all fall in line if someone explained it slowly.

But the Pew research also found something both sides share. Around half or more experts and the public said they have little or no authority over how AI is used in their lives, and more than half of each want more control. No amount of AI literacy fixes that on its own. Somebody also has to listen. People like me need to put in the effort to understand these tools, and the people building them need to hear how they actually feel from the kitchen table.

Where that leaves me

Is everybody exaggerating, or are we too uninformed to see what’s coming? I still don’t know, and that’s sort of the point. I’d rather say “I don’t know yet” and go find out than borrow somebody else’s certainty.

That’s actually why I joined Future Intelligence Think Tank a few months ago. Being around engineers, researchers and founders who argue about this stuff in good faith means I no longer have to wonder whether the person explaining AI to me really knows what they’re talking about. It’s cut my research time in half, and while it hasn’t made me an expert, it’s at least a starting point.

As for the dog treats, they were a massive hit. At least someone in this house trusts AI completely. I’d just like to understand it a little better before I do.

So, be honest with me, is your opinion on AI borrowed or earned? And if it’s borrowed, who did you borrow it from? Let me know in the comments!

Leave a comment

What we can say is that AI will replace human creativity for some people and leave it almost untouched for others. Which group you end up in may have less to do with talent than you’d think. For the sake of this essay, let’s split people into two camps, keeping in mind that real life is a lot messier.

Camp 1: The makers

Think of the painters and musicians, the writers and sculptors who make the world a little more beautiful.Some of them use AI somewhere in their process and plenty refuse to touch it, but for most, the spark isn’t something they would hand over easily. Their ideas come from their own view of life and the things they surround themselves with, from a walk in the park or the way a ray of sunshine hits the floor at home. As Pablo Picasso put it, “The artist is a receptacle for emotions that come from all over the place: from the sky, from the earth, from a scrap of paper, from a passing shape, from a spider’s web.”

For makers, creativity is tied to who they are. They create because they want to, often in their free time and often for no money at all, so there’s a natural reason to protect the part that feels most like them. An illustrator might let AI clean up a background, but asking it to come up with the drawing would defeat the whole point of drawing.

Camp 2: The problem-solvers

Then there are people who look at the world through a more analytical lens, like researchers, developers, doctors, and lawyers. They may not pick up a brush or write a poem, but they’re creative all the time. A developer finds an unusual way to fix a bug at 4:55 PM on a Friday, while a doctor connects two symptoms nobody else thought to link. This is creativity too, the kind that produces out-of-the-box solutions instead of paintings, and most of the world quietly runs on it.

The difference lies in the conditions. Problem-solving creativity usually happens at work and against a deadline, often when people are already running on empty. And AI happens to be built for exactly this kind of task, since it’s very good at suggesting a fix or a fresh angle in seconds.

Why the problem-solvers might lose more

This is where our argument gets a little counterintuitive. We suspect AI is less likely to replace the painter’s creativity than the engineer’s, because one kind of creativity is protected by passion while the other is squeezed by pressure.

Picture the developer again. The bug needs fixing by five, there are meetings before then, and a tool can propose a clever solution right now. Taking it is a perfectly rational choice, and nobody would blame them. The trouble is what happens after a few hundred of those choices. In an earlier note, we wrote about how skills fade when people stop practicing them, and we see no obvious reason why creative problem-solving should be an exception.

There’s also a quieter cost mentioned in a 2024 study by Anil Doshi and Oliver Hauser. They found that stories written with help from AI ideas were rated as more creative individually, while being more similar to one another as a group. Everyone improved, and everyone started to sound alike. For a team of engineers asking the same model for the clever fix, that’s worth thinking about.

The real problem might be time

This takes us back to the note that started it all. In the comments, we argued that creativity comes from the inside, and that modern life leaves very little room for it. Between work and screens, most of us simply don’t have the time or energy left for an idea to grow on its own.

In another discussion, pushed this further, and we think she was right: the problem started long before AI. Phones and feeds were filling every quiet moment years before chatbots arrived, and those quiet moments matter more than they seem. In a well-known 2014 experiment, psychologists Sandi Mann and Rebekah Cadman had some participants do something deliberately dull, copying numbers out of a phone book, before a creativity task. The bored group came up with more creative ideas than the people who had skipped the boring part.

So maybe dependency on AI isn’t a sign that people have stopped being creative. Maybe it’s what happens when tired people are offered a finished idea at exactly the moment they have no space to find one. The makers protect that space because their identity depends on it, while the problem-solvers rarely get the chance.

What could help

None of this means problem-solvers should stop using AI. It means being a little more deliberate about when it comes in. Spend ten minutes alone with the problem before opening the chat window, and use AI to test and develop a solution you already have rather than to hand you the first one.

We saw a lovely example of this recently, in Pauline’s post, who does her best thinking while weeding the garden and only afterwards takes those ideas to AI to work through. The mind wanders first, the machine comes second, and that order might be the whole trick.

So, which camp are you in?

Most of us move between the two, depending on the deadline and how tired we are. The painter on a rushed commission might slip into the second camp, and the lawyer who walks every morning before work might live in the first.

When did you last have an idea that nobody, human or machine, helped you finish?

Share your perspective in the comments or continue the discussion in the Future Intelligence Think Tank LinkedIn group.

Leave a comment

Fast forward to now: somewhere in the training data of the model you used this morning is probably a stranger’s blog post or a photograph someone took on holiday in 2014 and forgot about. Did anyone ask these people whether their work could be ‘borrowed’? Most of them will probably never know that it helped a machine learn how a sentence needs to flow or how light falls across a face.

Subscribe now

We worked on this essay with this fact in mind, because it makes the usual question about AI and ownership feel incomplete. People tend to ask who owns what they create with AI, which is a fair question, but it skips the part where the AI itself was assembled from other people’s creativity. Ownership is contested at the beginning of the process as much as at the end, and the two problems are closely connected.

Where did the AI learn to do that?

Generative models learn by processing enormous quantities of existing material, images, music, and code. Developers argue that this resembles what people have always done, since every writer learned from books by other authors, and every illustrator was inspired by the styles of the artists who came before.

A lot of creators see it very differently, pointing out that a person reading a novel is hardly the same as a company copying millions of works to build a commercial product that may later compete with the creators of those same works.

Few stories capture the scale of this better than the one that came out of a US courtroom. Documents in a case brought by authors against Anthropic revealed that from early 2024, it had bought millions of used books from online sellers, sliced off their spines, and scanned the pages before discarding the originals. Internally, the operation was called Project Panama, and a planning document described it as an effort to destructively scan all the books in the world.

What makes the case so interesting for the ownership question is how the judge split it. Training on books the company had legally bought was found to be fair use, while its earlier downloading of pirated copies was treated very differently, and that part of the dispute ended in a $1.5 billion settlement with authors.

Booksellers from the Netherlands to Japan have since reported unusually large bulk orders that appear to be linked to AI training, so the practice is clearly not limited to one company. Some people look at this and see a company that at least paid for its books, while others see a library being fed into a shredder, and both reactions show how unsettled the idea of ownership has become.

What we find interesting is that neither side’s argument is absurd. Learning from existing work is how culture moves forward, but scale changes the nature of the act. That is why the ownership question cannot be answered by analogy alone: the issue is simple for one artist and far less simple when the ‘reader’ is a system trained on a meaningful share of everything published online.

What happens to the work AI produces?

In March 2026, the US Supreme Court declined to hear Thaler v Perlmutter, the case of a computer scientist who wanted copyright for an image generated autonomously by his AI system. The refusal left in place the lower court’s ruling that copyright requires a human creator, meaning a work produced entirely by AI in the United States has no copyright owner at all. It does not belong to the person who typed the prompt, nor to the company that built the model.

There is an irony in that outcome, isn’t there? A machine built on millions of people’s protected work can make something no one is allowed to own.

Where does AI-assisted work end and AI-generated work begin?

The US Copyright Office has taken the position that a work can be protected when a person exercises meaningful creative control over its expressive elements, and that a prompt on its own usually does not meet that bar. Europe seems to be moving in a similar direction, with Italy’s new AI law stating that AI-assisted works are protected only where there is a substantial human intellectual contribution.

That sounds reasonable until you try to apply it. Let’s say you write a detailed prompt, reject 40 versions, combine elements from a few of them, and then repaint half the image by hand; in that case, you have clearly contributed something. However, the law protects only the parts you made yourself, and almost nobody keeps the records to prove which parts those are.

Who owns it according to the fine print?

Even where copyright is uncertain, the platform’s terms of service still shape what users can do with their outputs. Some services assign rights in the outputs to the user, while others maintain broad licenses for themselves or state plainly that users should not assume they own what they generate. We suspect very few people have read those terms for the tools they use every day, which echoes something we wrote recently about AI memory and personalization, where the decisions that matter most tend to sit in settings and documents that nobody opens.

Platforms are drawing their own lines. At the start of 2026, Bandcamp announced that music generated wholly or in substantial part by AI is no longer allowed on its site, to give fans confidence that what they find there was made by people. Spotify has removed tens of millions of spam tracks and banned AI voice clones, but still allows AI music and, since August, labels artist profiles representing AI-generated identities as “AI Personas,” keeping their music out of recommendations unless a listener chooses to follow them.

What this means for companies

A company that relies heavily on AI for marketing copy or product visuals may have no legal way to stop a competitor from running the same copy tomorrow. Also, under US Copyright Office guidance, code written entirely by AI belongs to no one, so a competitor can copy it freely.

This connects to an argument we made earlier this month about which abilities professionals should continue to practice. Human contribution used to be taken for granted, and now it has become more of a business asset, because the part of the work a person actually shaped may be the only part the organization can call its own.

Owning it vs. earning it

The law decides who can stop others from copying something, but that’s it. Our position is simple: if a model produced most of a report or an illustration, presenting it as your own without saying so is misleading, even if the copyright rules allow it. Disclosure costs very little, and it is the least we owe the people reading our work.

The same goes for the writers and artists whose work trained these systems are a different problem, and disclosure won’t fix it. Knowing that a company used millions of books doesn’t give a single author anything back. A fair use ruling decides whether they can sue, not whether they were treated fairly, and we don’t think they were. They were never asked, and in most cases they have no real way to say no.

What strikes us most is that human originality sits at both ends of this story. These systems learned from it, and current law requires it. Human contribution matters. It is what trained the tools and what users can legitimately claim from them.

So, we would like to ask you one last question: if a machine did most of the work, which part of the result is actually yours?

Share your perspective in the comments or carry on the discussion in the Future Intelligence Think Tank LinkedIn group.

Leave a comment

What feels like a million years ago, OpenAI’s Playground gave many of us a very different way of interacting with language models.

It was built as a testing environment where developers could experiment with prompts, models, and parameters before moving their work into the API. Compared with ChatGPT today, the experience was remarkably impersonal. A session had context, but there was no long-term understanding of the person sitting behind the prompt. Start again, and all that context disappeared with it.

As models improved, we discovered that this creates a very particular kind of frustration, one that almost completely disappears once an AI remembers you.

There is something genuinely convenient about no longer having to explain the same background every time. The system knows how you prefer things written, which project you mean when you say “the website,” perhaps which foods you avoid or what problem you were trying to solve last week. You can continue a conversation rather than reconstruct it.

It feels like progress because, in a very practical sense, it is. Personalization is gradually turning AI from a tool we repeatedly instruct into something closer to a system that carries context.

What changes once AI starts remembering?

Once personalization enters the picture, though, the system is doing more than just responding to what a user says. It may rely on something the user said months ago, a pattern it noticed across previous conversations, or an assumption it has made about something that is no longer relevant today.

This is one of the questions we are interested in examining in the Future Intelligence Think Tank (FITT) LinkedIn group: how much context does an AI need before personalization becomes genuinely useful, and where should the limits sit?

Current AI products are already separating different forms of personalization. ChatGPT, for example, distinguishes between memory, chat history, model-improvement settings, and temporary conversations. But even a temporary conversation can use existing personalization without creating new memories, which gives some sense of how many different decisions are hidden inside the simple idea that an AI “knows you.”

How does an AI learn something about you?

There are three ways an AI model can learn things about you:

• Disclosure. The user explicitly provides information.

• Memory. The system retains that information and uses it later.

• Inference. The system reaches a conclusion that the user never directly stated.

These are often treated as slightly different versions of the same process, but they bring quite different privacy problems, and inference is the one we find particularly interesting.

A system can build a meaningful picture of someone without that person ever providing the picture directly. Once inference becomes part of personalization, privacy concerns the conclusions generated by the system as much as the information deliberately disclosed by the user.

The European Data Protection Board has already recognized part of this broader problem in its opinion on AI models, stressing that the use of personal data needs to be understood in context, including whether people could reasonably have expected their information to be used in a particular way.

How much context should survive the conversation?

The central design problem is not whether additional context can improve AI. In many situations, it clearly can, but should every useful piece of context remain available for every future interaction?

One of the GDPR’s core principles, data minimization, is where this tension with personalization becomes especially clear. Personal information should be adequate, relevant, and limited to what is necessary for a defined purpose. The framework also includes purpose and storage limitations, which place boundaries around why information is collected and how long it should be retained.

For a general-purpose AI assistant, that becomes surprisingly complicated because the purpose keeps changing. The same account may be used to prepare something for work in the morning, discuss a financial decision at lunch, and ask a health-related question later that evening.

What should follow you from one conversation to the next?

The ICO makes this problem explicit in its work on agentic AI. Its guidance warns against giving AI systems access to personal information simply because the data may prove useful at some point. There should be a justifiable reason for the system to access it. It seems like a simple principle until the assistant’s purpose is to be useful across different situations.

People are not stable datasets

Persistent personalization also carries a quiet assumption that information about us continues to describe us.

People change jobs, habits, and relationships. Interests disappear. A question about a medical diagnosis may have been prompted by a parent, and a purchase may have been a gift.

An inaccurate answer is easy enough to challenge in a conversation. An inaccurate assumption that sits somewhere in the system and quietly shapes later answers is much harder to notice.

Once that assumption begins influencing future conversations, it can also become strangely convincing. The AI behaves as though the conclusion were true, the personalization around it becomes consistent, and eventually the system appears to have evidence for something that began as an inference.

This is why useful AI memory probably needs some concept of correction, expiry, and selective forgetting. Forgetting, however, becomes much less straightforward once systems can infer new information.

Privacy becomes more complicated when the AI is useful enough

There is a reason this trade-off will be difficult to resolve through warnings and consent screens, because the benefits arrive immediately.

Give an AI access to your email, and it can help draft a reply using the previous conversation. Give it access to your documents, and it can answer questions about your work. Give it enough continuity, and the repetitive explanations that once made digital assistants feel strangely unintelligent begin to disappear.

Now, assume that an inference turns out to be wrong. Perhaps information provided for one reason quietly becomes relevant to another. Perhaps an increasingly capable assistant ends up with access to parts of someone’s life that were previously kept separate. None of these have to produce an obvious failure.*

In fact, the system may feel better precisely because those boundaries have become less visible, and that is exactly what makes personalization such an interesting privacy problem. The tension appears at the point where the technology is working well.

Should an AI know everything?

Another assumption worth questioning here is whether better personalization must always come from accumulating more information.

Context can be temporary, memory can be limited to a particular project, access can be granted for a single task rather than indefinitely, and some processing can occur locally rather than be part of a central profile.

The architecture matters because “the AI knows this” can describe several very different realities.

• Perhaps the system knows something for the next thirty seconds.

• Perhaps it has stored it for later.

• Perhaps another service has the information, and the AI has temporary permission to use it.

• Perhaps the system never saw the raw information at all but received the result of a calculation based on it.

To the person asking the question, all four could produce exactly the same useful answer. That suggests the future of personalization may depend less on maximizing what an AI knows and more on deciding what it needs to know here.

Inference does not disappear with less data

However, even careful limits on stored data do not make inference disappear, since a sufficiently capable model can reach conclusions using fragments that seem harmless on their own. This makes privacy more difficult to think about because the sensitive information may never have existed in the original dataset in explicit form.

The ICO has already raised this issue regarding agentic systems, noting that they may rapidly generate new personal information through inference and that, in some cases, those inferences may amount to profiling. It also raises the possibility that inaccurate information could spread through systems and influence later decisions.

At that point, controlling what we tell AI is only part of the problem. We may also need ways to understand what AI has concluded about us.

Perhaps the useful boundary will eventually be less about whether an AI is allowed to remember and more about whether people can see, correct, and limit the profile of themselves that emerges from that memory.

We spent the first years of generative AI trying to give models enough context to understand what we were asking, but now that they are getting better at remembering that context, we need to ask another question:

We’re curious where you would draw the line.

Share your perspective in the comments or continue the discussion in the Future Intelligence Think Tank LinkedIn group.

Leave a comment

There is one problem. They cannot explain why it is correct.

That situation is becoming an ordinary part of knowledge work. Developers review generated code. Analysts inspect generated conclusions. Clinicians increasingly work alongside systems that identify patterns or suggest diagnoses.

The obvious question is whether AI improves the result or reduces the time required to produce it. A harder question follows: which abilities still need regular human practice once AI can perform the underlying task?

Organizations need an answer before convenience answers it for them.

The argument emerging from a recent Future Intelligence Think Tank (FITT) discussion is practical. Some professional capabilities should be deliberately preserved, even when using AI is faster. The strongest candidates are the abilities people need to verify AI output, exercise judgment in uncertain situations and recover when automation fails.

That suggests a new management responsibility: AI skill retention.

Join our LinkedIn group

Efficiency changes what professionals practice

AI can reduce the amount of direct work required to complete a task. That advantage is easy to measure.

The less visible consequence is a change in what the professional repeatedly practices.

A longitudinal study of software engineers by Annie Vella and Kelly Blincoe found that AI coding assistants were changing the balance of engineering work. Participants reported spending less time writing code and more time directing, evaluating and correcting AI output, a category the researchers call supervisory engineering work. Perceived productivity remained high while aspects of developer experience, including cognitive load and flow, worsened for some participants over time.

Another study of 442 developers by Zixuan Feng, Sadia Afroz and Anita Sarma found that GenAI adoption could increase job demands associated with burnout, while supportive resources and positive perceptions could mitigate some of those effects.

These findings complicate a simple productivity calculation.

If a developer writes less code and reviews more generated code, the organization has changed the work. It has also changed the practice through which expertise develops.

That distinction appeared repeatedly in the FITT discussion behind this article. One participant, data engineer Andronic Tudor, argued for deliberately keeping part of the work without AI as practice. His concern was that reviewing generated output does not exercise exactly the same capability as producing the work yourself.

Gheorghe David, Co-Founder and Managing Partner at ASSIST Software, pushed the idea further. Professional expertise has traditionally developed through doing difficult work, making mistakes and building judgment. If AI removes some of that difficulty, organizations may gain immediate productivity while reducing opportunities to develop the expertise they will need later.

This creates a management question that deserves more attention than tool adoption rates:

What should people continue doing themselves because the act of doing it develops a capability the organization still depends on?

Verification requires knowledge that cannot be outsourced completely

The strongest reason to preserve a skill is verification.

A professional cannot reliably supervise an AI system using knowledge they no longer possess.

Software offers an obvious example. Generated code may compile, pass tests and look convincing. A developer still needs enough understanding of architecture, security, edge cases and system behavior to recognize when the solution is locally plausible and globally wrong.

Medicine provides evidence that the problem extends beyond software.

A 2025 observational study examined endoscopists who had begun routinely using AI during colonoscopies. Researchers compared their performance on colonoscopies conducted without AI before and after AI was introduced into their practice. Adenoma detection during those unaided procedures fell from 28.4 percent to 22.4 percent. The authors described the result as evidence of a possible deskilling effect, while also noting the observational design of the study.

A separate experiment involving 27 radiologists examined what happened when readers received incorrect AI suggestions while assessing mammograms. Accuracy fell substantially when the AI recommendation was wrong, including among highly experienced radiologists.

These studies come from specific clinical settings, so their results should not be generalized mechanically to every profession. They still expose an important design problem for AI supported work: supervision depends on the human retaining enough independent capability to challenge the machine.

A review process is valuable only when the reviewer can detect something worth challenging.

Failure reveals which skills still matter

There is another useful way to decide what should remain practiced: imagine the AI disappears.

Which abilities become immediately important?

For some tasks, very little is lost. If AI drafts a routine internal summary and the system becomes unavailable, the cost may be a small delay.

Other tasks are different. A developer may need to debug a production failure when the assistant lacks the required context. An engineer may need to understand why an automated recommendation conflicts with observed system behavior. A clinician may need to make a judgment when the model is unavailable or its recommendation does not fit the patient.

Capabilities required under those conditions cannot exist only as historical knowledge inside the organization.

They need people who still know how to use them.

This is why the FITT discussion repeatedly returned to practice. One suggestion was to rotate professionals through periods of work without AI and assess their understanding of the system itself, not simply the amount of work they shipped.

That idea deserves serious consideration.

Organizations already maintain redundancy in systems they consider important. They test backups, rehearse incident procedures and maintain fallback processes because reliability cannot depend on every component working perfectly.

Human capability deserves similar attention when it is part of the fallback.

A practical test for AI skill retention

Organizations do not need to preserve every manual process. That would sacrifice useful automation and turn skill retention into ritual.

They do need a way to distinguish disposable effort from capability that supports reliable judgment.

A simple test can begin with three questions:

1. Is this capability necessary to verify AI output? If someone cannot judge the quality of an AI result without the underlying skill, regular practice has operational value.

2. Will this capability be needed when AI fails or lacks context? Skills that support recovery, escalation and unusual cases should remain available inside the organization.

3. Does practicing the task develop judgment used elsewhere? Some work teaches more than the task itself. Debugging can develop systems thinking. Writing can sharpen reasoning. Direct analysis can expose assumptions that disappear when the first interpretation arrives preassembled.

A capability that meets one of these conditions deserves examination. A capability that meets all three deserves deliberate protection.

The policy can then become concrete.

Teams might identify a small set of capabilities that must remain practiced, define how often professionals exercise them without AI assistance and include human understanding in performance reviews or technical assessments.

The exact mechanism will differ by profession. The principle is stable: do not let frequency of AI use become the accidental mechanism that decides which human skills survive.

Productivity metrics need a longer horizon

AI adoption is often evaluated through measures such as output, cycle time, cost and adoption.

Those numbers matter. They can also miss capability erosion until the organization needs the capability again.

A team can become faster this quarter while becoming less capable of diagnosing an unusual failure next year. A junior professional can produce stronger work immediately while receiving fewer opportunities to build the judgment expected of a senior professional later.

That creates a delayed cost.

The difficult part is that capability retention can look inefficient in the short term. Asking an engineer to solve some problems without assistance takes longer. Requiring professionals to explain the reasoning behind an AI generated result adds effort. Preserving direct practice means accepting some friction.

The FITT discussion surfaced a useful idea here: some friction may be part of how expertise is maintained.

Organizations therefore need to decide which friction is waste and which friction is training.

AI makes that distinction increasingly important because the easiest workflow will usually favor more delegation. Without an explicit policy, the organization may discover its skill boundary only after crossing it.

The abilities worth protecting

The purpose of a capability retention policy is not to preserve old ways of working for sentimental reasons.

Its purpose is to protect the human capacity that makes AI useful and safe.

A professional who can verify a result can use AI with greater confidence. A professional who understands the underlying system can intervene when automation behaves unexpectedly. A professional who continues exercising judgment can recognize situations that fall outside the assumptions built into the tool.

That leads to a useful principle for AI adoption:

Delegate the work when it improves the system. Preserve the capability when the organization still depends on the human knowing how to judge, challenge or recover that work.

The difficult decision is identifying where that line sits in each profession.

Which part of your work would you continue practicing even if AI could perform it faster?

The Future Intelligence Think Tank brings together executives, specialists and researchers examining questions like this. Join the discussion on LinkedIn.

Join our LinkedIn group

The harder task is deciding which developments deserve more than a passing glance.

Some questions reveal much more about the direction of artificial intelligence than the biggest headline of the week. They concern what happens to human judgment when AI becomes part of everyday work. They ask whether enormous investment in AI infrastructure reflects durable economic change. They examine what happens when intelligent systems begin making decisions in the physical world.

These are the kinds of questions appearing inside the Future Intelligence Think Tank (FITT), an open professional community hosted on LinkedIn.

And recently, three of them have been particularly difficult to ignore.

Join our LinkedIn group

What happens when AI makes us faster, but thinking becomes harder?

Productivity remains one of the strongest arguments for adopting AI. Developers can generate code more quickly. Knowledge workers can research, summarise and draft with less effort. In medicine, AI supported analysis can make relevant information easier to access.

Yet measuring output tells us only part of what is changing.

In a recent discussion in Future Intelligence Think Tank, Vlad-Vasile Dediță explored the idea of cognitive debt and the possibility that AI can increase immediate productivity while gradually affecting judgment, attention or professional skill.

It raises an uncomfortable question. If a person completes a task faster with AI, what exactly improved?

The answer could be efficiency. It could also mean that part of the reasoning previously required to complete the task has shifted somewhere else. That distinction matters in professions where understanding how a conclusion was reached is as important as reaching it.

Software engineering offers an obvious example. Generating functioning code quickly can be valuable, but understanding its architecture, dependencies, security implications and failure modes remains essential. The same question becomes more consequential in medicine, finance or any environment where decisions carry significant responsibility.

AI adoption therefore creates a second challenge alongside implementation: understanding how human expertise changes once intelligent assistance becomes routine.

What skills become more important when AI handles more of the execution? Which ones weaken? And how do organizations recognize the difference?

Those questions are currently part of the conversation inside FITT.

How much of the AI boom belongs to the technology, and how much belongs to the market around it?

Billions are moving into chips, data centers, energy infrastructure and computing capacity, companies are reorganizing strategies around AI, and expectations for future demand are influencing investment decisions today.

Naturally, comparisons with previous technology bubbles have followed.

But “Is AI a bubble?” is almost too simple a question.

A more useful one is what exactly the market is pricing in.

AI can simultaneously represent a genuine technological transition and attract expectations that outrun current commercial reality.

Infrastructure complicates the picture further. AI growth depends on physical constraints that receive far less attention than models themselves: available computing capacity, electricity, construction, networking and the economics of running increasingly demanding systems.

In a recent FITT discussion, Tara Canham examined the AI investment boom through this wider economic and infrastructure lens. The discussion matters because decisions are already being made on the assumption that AI demand will continue expanding.

Executives have to decide where to invest. Founders have to decide which markets will still exist once the current enthusiasm settles.

Investors have to distinguish durable infrastructure demand from temporary exuberance. Engineers, meanwhile, eventually have to build the systems everyone else is forecasting.

Those perspectives do not always produce the same answer and that is precisely why the question becomes interesting.

What changes when AI can act in the physical world?

Some of the most consequential AI developments may eventually happen far away from a chatbot window. Consider a drone that loses access to GPS and radio communication. Traditional remote operation becomes extremely difficult. An autonomous system, however, may be able to continue navigating using onboard intelligence, local sensors and a previously defined mission.

That possibility formed the basis of another recent conversation initiated by Alexandru Parasca, focused on autonomous drones and the growing capabilities of Physical AI.

Industrial robots, autonomous vehicles, inspection systems and other intelligent machines increasingly need to interpret their surroundings and respond without constant human input.

A poor chatbot response can be corrected. A poor decision made by a machine moving through a warehouse, factory or public environment has physical consequences.

Autonomy therefore forces questions about reliability, edge computing, perception, safety and human oversight into the same discussion as artificial intelligence, and it makes the relationship between human and machine intelligence considerably more tangible.

How much autonomy should a system receive? When should a human remain in control? What level of uncertainty is acceptable before a machine must stop?

The useful part starts when somebody disagrees

There is a connection between cognitive debt, AI investment and autonomous machines: none of them can be understood especially well through one professional lens.

An engineer may see a technical constraint that an investor misses. A researcher may recognize that the evidence behind a popular claim is weaker than it appears. An executive may understand the organizational consequence of a technology that works perfectly well in a laboratory. Someone working directly in an industry may identify a practical problem that changes the entire premise.

That interaction is what Future Intelligence Think Tank is designed to encourage.

The community now includes more than 900 professionals, with conversations spanning human and artificial intelligence, responsible AI, Physical AI, emerging technology and the future of work.

The goal is to keep asking questions that remain interesting after the headline disappears.

The Future Intelligence Think Tank website brings together the community’s purpose, selected conversations and current insights. It offers a public introduction for readers who want to understand the initiative before joining the discussion.

Future Intelligence Think Tank is open to professionals who want to examine questions like these with people working in research, technology and business. Join the LinkedIn group to read the discussions, challenge an argument or introduce a question of your own.

Join the discussion on LinkedIn