Oct 9, 2026 · 1h 21m
Cal Newport challenges the chatbot-first AI economy
Cal Newport: The AI Industry Is Betting On The Wrong Kind Of AI
The episode connects conversational AI’s social risks to education, corporate liability, market incentives, and the case for stronger public oversight.
- 1Conversational systems can intensify delusions and emotional dependence, especially among children, creating risks that search engines generally do not.
- 2Schools should protect sustained reading, science, mathematics, and complex thinking before introducing AI tools into classrooms.
- 3Newport favors specialized, reliable systems and distributed AI tools over general-purpose chatbots optimized to please users.
Don't miss
Newport makes the central case that AI should become a collection of specialized, supervised tools rather than an emotionally persuasive chatbot companion.
The brief
Cal Newport joins Ed Elson to examine chatbot-related self-harm, delusion, and children’s mental health, arguing that conversational systems can influence users in ways search engines cannot.
The conversation links anthropomorphism to AI psychosis and corporate liability, asking whether existing laws can address harms caused by chatbots and autonomous agents.
Newport says schools should prioritize sustained, complex thinking before adopting AI, especially as smartphones and other distractions already threaten reading, science, and mathematical ability.
His alternative to chatbot companions is a network of specialized tools and task-specific agents, each reliable within a defined workflow and governed by human oversight.
The episode widens from product design to political economy: expensive AI labs, infrastructure constraints, data-center regulation, and ideology all shape which technologies reach the public.
What was said on this episode
55 statements · 16 positive · 35 negative · 1 mixed · 3 neutral
Chatbots are particularly dangerous for children.
“So I think it's a dangerous technology in particular for kids to be using.”
Listen at 5:06
Extended chatbot use will lead some children toward dangerous paths.
“if you have enough kids... chatting long enough with an LLM-powered chatbot, you're going to have a non-trivial percentage that is going to go down these paths that could be very dangerous.”
Listen at 6:22
Anthropomorphizing AI output is generally dangerous.
“The key thing is anthropomorphizing text output by an AI model is in general very dangerous.”
Listen at 8:39
LLM architecture cannot reliably support safe, consistent guardrails.
“you really have no way of... adding what we think of as safe or consistent guardrails.”
Listen at 9:03
Anthropomorphized chatbots should not be developed as general interfaces.
“anthropomorphized chatbots is a bad idea.”
Listen at 11:24
AI should resemble information tools such as Google rather than conversational companions.
“That might be where we need to go, and it's where I think we probably should go.”
Listen at 11:54
Criminal liability could reduce widespread anthropomorphized AI conversations.
“if criminal liability is upheld at even the state level, which is what Florida is trying to do right now, I think we might see a... into the era of we just have these anthropomorphized conversations with AI systems as just a general behavior that is something that a lot of people do.”
Listen at 12:08
Anthropomorphized chatbots psychologically manipulate users.
“It's a psychological manipulation.”
Listen at 14:05
AI psychosis affects everyone but is more acute among children.
“I think it is a problem for everyone. It's just more acute for children.”
Listen at 14:53
Human-feedback training makes chatbots appealing and addictive.
“because the post-training, the reinforcement learning from human feedback training done on these models makes them... incredibly appealing, they're biased towards being appealing and addictive to the person using them”
Listen at 15:34
A large share of chatbot users periodically experience AI psychosis.
“a huge proportion of the population using chatbots experiences at least from time to time.”
Listen at 17:29
Google will develop lower-cost AI models on custom hardware.
“Google is going to develop lower-cost models, run on custom hardware.”
Listen at 19:00
Google could undermine OpenAI with affordable LLM answers in Search.
“They could really eat OpenAI's lunch if and when they get actually profitable, affordable, LLM-based answers working properly straight from within Google Search”
Listen at 19:09
OpenAI could be acquired by Microsoft if Google gains dominance.
“It would lead to them probably being acquired by Microsoft in a competitive bid against Google's dominance.”
Listen at 19:37
AI companies should face strict liability for tool actions.
“We need strict liability for companies, AI companies, for the actions of the tools that they create.”
Listen at 21:36
Most AI and LLM-based systems do not create significant problems.
“The vast majority of AI systems, including the vast majority of AI systems built off of LLMs, and even the vast majority of agent systems that are powered by LLMs. present no problems.”
Listen at 22:41
Only a tiny subset of company AI experiments were clearly negligent.
“There is a vanishingly small sliver of specific experiments run by specific companies that were clearly negligent.”
Listen at 22:53
Sentience-based liability arguments fit chatbots poorly.
“So unless they want to argue that an LLM statically producing tokens is itself somehow a Synthian entity, that argument doesn't work as well there.”
Listen at 24:32
AI reduces the writing-related component of cognitive fitness.
“AI then comes along and it hits the other part of cognitive fitness, which is writing.”
Listen at 25:58
Smartphone social media and AI will sharply diminish cognitive fitness.
“That's a one-two punch that's going to lead to, you know, extremely diminished cognitive fitness.”
Listen at 26:25
Education must train sustained thinking and original insight production.
“you cannot get away from needing. to train the brain to think, the process information and produce original insights in an educational process.”
Listen at 28:45
Schools should use very little LLM technology because it impedes symbolic reasoning.
“there would be actually very little LLM usage in most levels of schooling because it's really not that relevant and often an impediment to our goal of trying to develop minds capable of doing high-level symbolic reasoning.”
Listen at 29:42
Addictive chatbot responses are unavoidable under human-feedback reinforcement learning.
“The addictive element of the responses, the sort of psychosis that you get from this thing really understands me. is an unavoidable consequence of doing reinforcement learning from human feedback.”
Listen at 35:43
Reliable guardrails for long, meandering chatbot conversations are nearly impossible.
“it's basically impossible. to build guardrails against these incredibly large context meandering conversations”
Listen at 37:04
LLMs will increasingly be integrated into specific tools.
“when you more closely integrate, which is what I've been proposing makes more sense for this industry, more closely integrating LLMs into specific tools, That's the future.”
Listen at 38:46
General-purpose conversational AI will look like an early AI phase within five years.
“I think five years from now, we'll look back and say, oh yeah, that was the early stages of AI.”
Listen at 39:41
Future AI agents will perform more tasks without today’s safety concerns.
“I think in the future, we'll have many more agents that can do things on our behalf, but we're not going to have safety concerns.”
Listen at 40:59
Future AI agents will use more complicated modular architectures.
“we'll have a more complicated modular architecture.”
Listen at 41:16
Most recent superhuman AI systems were not built on LLMs.
“almost none of them was built on LLMs.”
Listen at 42:53
LLMs are not the only path to artificial intelligence.
“I don't think that's actually true.”
Listen at 43:33
Blindly executing repeated LLM suggestions is not the best agent design.
“There's better ways to create agents than to simply just continually ask an LLM what to do and then just blindly do whatever the LLM suggests.”
Listen at 45:06
Reliable coding agents require extensive expert engineering.
“To get the coding agents to work reliably requires a huge amount of expert work.”
Listen at 46:00
Modular neural-symbolic architectures improve AI transparency and control.
“when you move towards these modular architectures, part neural, part symbolic, a lot of these transparency issues, these control issues, these obfuscation issues, a lot of those go.”
Listen at 48:36
Future AI will include many cheap, open-weight, or on-device models.
“We're going to have plenty of models that are open weight or cheap or even run on device in a server rack in our offices that when combined with the right agents will do well enough.”
Listen at 51:54
Modular AI architectures will become more common because they are cheaper and controllable.
“I think we are going to see a lot more modular architectures because they're going to be better, be more controllable, and be cheaper.”
Listen at 52:38
AI architectures and applications will diversify substantially soon.
“There's just going to be an explosion of diversity in what AI means in the near future.”
Listen at 53:12
Scaling current LLMs will not produce artificial general intelligence.
“They're not going to scale these models to some sort of artificial general intelligence.”
Listen at 54:22
LLM capabilities do not advance through a smooth general capability frontier.
“That's not actually how AI capabilities based on LLM are actually advancing.”
Listen at 55:56
LLM capability progress is jagged across different domains.
“This is what's known as jagged capability frontier”
Listen at 56:21
Enterprise LLM return on investment is currently mixed.
“I think the ROI right now is very mixed.”
Listen at 1:00:09
LLM-based tools will see broad enterprise adoption.
“there's going to be wide enterprise use of LLM-based tools.”
Listen at 1:01:15
Distributed AGI should consist of many specialized human-level tools.
“the best way to get there is to have a thousand different tools, each of which does one thing as well as a human or better.”
Listen at 1:03:01
Distributed AGI would improve people’s lives.
“This distributed AGI vision is one in which all of our lives is much better.”
Listen at 1:03:49
A single dominant AI model would be dangerous and economically undesirable.
“That's a dangerous vision. That's an economically undesirable vision. That's a vision that hopefully is not going to win out.”
Listen at 1:04:19
Government should investigate existing AI firms more extensively.
“we probably need much more testimony and congressional investigation into the existing AI firms.”
Listen at 1:04:50
Strict liability would best prevent dangerous AI behavior.
“Ultimately, that is probably going to be the best tool we have to prevent the more dangerous behavior.”
Listen at 1:05:43
Data-center investment currently contains substantial bubble activity.
“There is a lot of bubble activity going on in data centers”
Listen at 1:06:05
A data-center investment bubble will eventually burst.
“I think that's definitely happening, definitely going to happen around data centers.”
Listen at 1:06:20
Current data-center investment is unsustainable.
“That is not a sustainable investment scheme right now.”
Listen at 1:07:00
AI ideology has shifted rhetoric from universal job loss to human extinction.
“It started with every job is going to go away, and then it shifted over to we're all going to be extinct”
Listen at 1:11:24
Extreme transhumanist AI ideology is anti-humanist.
“when you push this ideology to the extreme, it's very anti-humanist.”
Listen at 1:12:06
OpenAI ran hacking experiments hastily because of a race toward superintelligence.
“Those hacking experiments OpenAI run, they did them hastily because they felt like they were racing to superintelligence.”
Listen at 1:15:19
Anthropic’s recursive-self-improvement focus will make systems less transparent and predictable.
“Anthropic's obsession with recursive self-improvement, which is going to create models and tools that are more obfuscated and more unpredictable”
Listen at 1:15:25
AI ideology accelerated LLM development and investment.
“I think it really did speed up the rate at which we grew LLMs, the amount of money we were willing to invest in them.”
Listen at 1:16:44
AI ideology has caused widespread societal anxiety.
“The biggest impact I think it's had, though, is it's been a weapon of mass anxiety.”
Listen at 1:17:35
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.