healthcarereimagined

17 min read. ·Jun 3, 2026

By Eric Blaettler

A multi-trillion-dollar industry built on a confusion between two kinds of intelligence — and what a new physics of meaning changes about that

“Artificial Intelligence” contains two misnomers. Intelligence was borrowed from the military definition: information hoarded as competitive advantage. Piaget’s definition — the one Yann LeCun himself cited — is the opposite: intelligence is how you behave when you don’t know, an ephemeral flow through a gap-closing network, not an accumulating asset. Artificial fails because you cannot simulate a flow inside a single frozen system.

The real problem is structural: current AI has no zero of meaning, no way to represent verified absence. This is the Roman numeral problem applied to knowledge — a notation that forces every query to produce an answer, even when the honest answer is I don’t know.

The fix is not a better model. It is a missing protocol layer — the TCP/IP of meaning — that introduces a new physics of meaning: verified knowledge has semantic gravity, ignorance has none, and a system that knows where its light cone of care ends can finally behave differently at that boundary. That protocol, built on Berners-Lee’s Semantic Web, Burgess’s Semantic Spacetime, and Peirce’s semiosis, connected by cryptography, is the Semiotic Web.

When it exists, intelligence stops being something the strong hoard against the weak — and becomes what it always was: a flow that benefits the whole network precisely because no single node can own it.

Something unexpected happened while searching for the right frame to explain a three-year investigation into what intelligence actually is.

A social media post from Yann LeCunappeared. LeCun — the Turing Award winner, the most prominent voice arguing publicly that large language models are a dead end — was sharing his admiration for the Swiss developmental psychologist Jean Piaget, describing Piaget’s work as a direct influence on his own approach to machine intelligence. The quote he was amplifying read:

“Intelligence is not what you know, but how you behave when you don’t know.”

The reason this stopped the investigation cold has nothing to do with LeCun’s specific architectural proposals. It has to do with what the quote implies about the industry LeCun helped create — and about why its very name contains not one misnomer but two.

The Word “Intelligence” Has Two Meanings. The Industry Chose the Wrong One.

There is a word that intelligence agencies, military strategists, and geopolitical analysts use constantly. It means: information hoarded to gain competitive advantage over those who have less. You intercept enemy communications. You recruit sources behind enemy lines. You protect your own secrets while mining theirs. The goal is asymmetry — knowing more than the other side knows, and exploiting that gap.

This is military intelligence. And it is the diametrically opposite meaning to the one Piaget defined.

Military intelligence is a stock. You accumulate it, protect it, weaponize it. The more you hold relative to your adversary, the more power you wield. The entire competitive logic of nations rests on this version of the word.

The confusion between these two meanings is not incidental to the story of Artificial Intelligence. It is the story. The driving forces that funded early AI research — military institutions, defense contractors, intelligence agencies — were intelligence agencies in the military sense. They wanted systems that could process more information, faster, than any human adversary. They wanted the informational asymmetry of knowing more. They called it intelligence because that is what they were in the business of producing.

And so an entire field was named, funded, and shaped around the military definition — while borrowing the prestige of the cognitive one.

The result is a Ptolemaic confusion: treating intelligence as something that can be accumulated, concentrated, and deployed as an asset — as if it were oil or uranium, a resource whose possession confers power. A trillion-dollar industry has been built on this premise. And the premise, as Piaget’s definition makes clear, is precisely backward.

The First Misnomer: “Intelligence”

Read Piaget’s sentence as a mathematician would read a formal specification.

Intelligence is not what you know. This eliminates, by definition, every system whose fundamental design is to know things — to accumulate, index, compress, and retrieve. A system trained on all digitized human knowledge, designed from the ground up to always produce an answer, has under this definition precisely zero intelligence. Not low intelligence. Not imperfect intelligence. The derivative of gap-closure on a system permanently in a state of apparent knowing is zero, regardless of its parameter count.

How you behave when you don’t know. This requires, as a precondition, that the system has the structural capacity to recognize it doesn’t know. A gap must be felt before it can be closed.

The Four Stages of Competence model — formalized by Martin Broadwell in 1969 and widely circulated by Noel Burch at Gordon Training International in the 1970s — makes this visible with precision. The model describes how any skill is actually learned:

1. Unconscious Incompetence — You don’t know what you don’t know.
2. Conscious Incompetence — You know that you don’t know.
3. Conscious Competence — You know, but it requires deliberate effort.
4. Unconscious Competence — You know without thinking. Fluency. Second nature.

Intelligence exists only in the gap between stages 2 and 3 — in Michael Levin’s biological definition, intelligence is a flow, not a stock, and that flow only moves when a gap is felt and acted upon. Stage 4produces something enormously valuable — mastery, fluency, reliability, speed — but zero intelligence. Kahneman called this System 1: pattern recognition operating below the threshold of deliberate effort. Every AI system in production today is the most powerful Stage 4 engine ever built.

This explains something most people notice about gifted polyglots but can never quite articulate. Someone who speaks ten languages fluently is not necessarily more intelligent than someone who speaks two. Learning each language demanded tremendous active intelligence — to feel the gaps in grammar, to notice where intuition failed. But once crystallized into fluency, it transformed into communication skill. Spectacular, useful, but no longer intelligent in Piaget’s sense. You can ask the same confused question in ten languages and still have no idea how to answer it.

This is precisely what any AI system does — at civilizational scale. Trained on massive, redundant information, it has crystallized patterns of human expression into Stage 4 fluency. It cannot understand anything that requires recognizing a gap it has no structural way to feel.

The Second Misnomer: “Artificial”

The word “artificial” implies a conscious decision to simulate something that exists naturally. But intelligence, properly defined, is not a property of a system. It is a flow through a network — and therefore cannot be simulated inside a single system at all.

When Einstein developed General Relativity, intelligence wasn’t located inside his skull. It was flowing through a network: Grossmann’s mathematics, the anomaly of Mercury’s perihelion, the resistance of colleagues in Zurich, the thought experiment of the falling elevator, Eddington’s eclipse expedition, the entire decade of productive friction with a reality that kept refusing to fit the existing frame. The intelligence was the gap-closing. It evaporated the moment each gap closed.What remained was General Relativity — no longer intelligence, but verified knowledge crystallized into a form others could use without re-deriving it.

There is nothing artificial about that process. It is the most natural thing there is. And the systems carrying the name “Artificial Intelligence” don’t simulate it. They are its most sophisticated archival product — extraordinary libraries that emerged from an intelligent process and make the results of that process accessible to everyone. Remarkable. Genuinely transformative. But not intelligence. An archive of intelligence’s outputs.

The very fact that we call the creators of AI and Deep Learning its “Godfathers” reveals the misunderstanding in its purest form. Godfathers originate, control, and concentrate. Intelligence, by Piaget’s definition, is ephemeral, distributed, and shared. The word chosen for the founders of the field is the word of the military definition — of power, hierarchy, and concentrated authority. That is not an accident. It is the cultural trace of which version of “intelligence” was running the show when the field was named.

The preprint The Einstein Test and Beyond, written with cognitive scientist Tony McCaffrey, uses a more accurate term for current AI systems: Stochastic Guessing Engines. Not an insult — a technical specification. These systems guess brilliantly, statistically, from more accumulated pattern than any previous system. And that specification carries a constraint that no amount of engineering within the current paradigm can remove.

Why LeCun Is Right, and Where It Stops

Yann LeCun left Meta in late 2025 after the company reorganized toward closed, product-driven development and cut hundreds of researchers from FAIR. He launched AMI Labs with a billion-dollar seed round, betting that the path to real machine intelligence requires world models grounded in physical reality — systems that learn the statistics of reality, not just the statistics of words. His core critique is structurally correct: a system trained on text cannot develop genuine understanding of the physical world. Language is discrete, finite, and thin. Physical reality is continuous, multidimensional, and dense with information that no text corpus can contain.

The Piaget quote LeCun shared is the best single-sentence statement of why his direction is right. And it also points precisely to the place where his solution does not yet go far enough.

LeCun’s JEPA world models learn to predict the future state of their environment in abstract latent space rather than at the pixel level. Instead of reconstructing what the world looks like, they predict what it means— learning structured representations of physical causality. This is a meaningful advance. But the latent space in which JEPA models predict has no stable address. A concept — gravity, proximity, containment — exists at different coordinates in every model, every version, every training run. Without stable addresses, a world model’s predictions remain sealed inside a private coordinate system that no other system can independently verify, challenge, or build upon. The world model learns what the world probably looks like. It cannot learn what the world verifiably is in a form that survives contact with other agents. It predicts brilliantly — in isolation. And isolated prediction is not intelligence. It is the most sophisticated Stage 4 pattern-matching ever conceived.

A hoarded asset has zero intelligence. Its use as a competitive weapon produces actions that cannot be called intelligent — because intelligence requires the capacity to feel a gap, and a frozen system, sealed inside its probability manifold, can only feel the gravitational pull of what it has already seen. Everything else is weightless. Everything else is outside the light cone.

The Roman Numeral Problem

The notation problem that runs through all of this has a precise analogy in the history of mathematics — one worth holding before reaching the structural fix.

Try dividing 1 by 3 in Roman numerals.

You cannot write the result. Not because Roman numerals lack symbols for small quantities — they have I, which can be subdivided in principle. The problem is that the system has no placeholder. No concept of a position that carries structural meaning through its emptiness. The result of 1 ÷ 3 is the repeating decimal 0.333… — a number defined precisely by the relationship between a non-zero numerator and its infinite positional expansion. Without zero as a structural entity, without a symbol meaning “this position is empty and that emptiness is meaningful,” you cannot write 0.333. You cannot represent the difference between 10 and 100.

Every Roman numeral calculation has to be performed against the grain of the notation itself — accumulating marks and hoping the pattern yields the answer. The Romans were not unintelligent. The notation simply made certain classes of operation structurally unavailable — not difficult, not slow, but impossible to conceive.

When the Hindu-Arabic positional system arrived in Europe, carrying the zero that Indian and Arab mathematicians had formalized, it did not merely accelerate arithmetic. It introduced a substrate in which operations previously impossible became structurally natural: fractions, algebra, calculus, the mathematical basis of modern physics. Not because zero represented some new object in the world. Because zero gave the notation the capacity to represent the absence of a value as a structural entity — a placeholder whose presence in a specific position carries precise meaning.

Current AI is performing Roman-numeral arithmetic with meaning. The Softmax function that every transformer uses is formally identical — not loosely analogous, identical — to the Boltzmann distribution in statistical thermodynamics: an equation that always redistributes probability mass across all possible outputs, always summing to one, always producing a non-zero response for every query, including queries for which no verified answer exists. The system must produce something. It has no structural zero. Hallucinations are not a bug to be patched. They are the architecture faithfully executing its specification.

The Light Cone of Care

There is a concept from physics that makes the weight of ignorance unexpectedly precise.

SIn Einstein’s relativity, only events within your past light cone — events close enough in space and time that light could have reached you from them — can influence you. Everything outside that cone is, by the laws of physics, causally disconnected from your present. It cannot act on you. It has no gravitational pull on your trajectory.

The same principle applies to meaning. An AI system cannot care about what it doesn’t know exists. Not because of a programming limitation. Because ignorance, structurally, has no weight. An unverified concept — a fact no training example contained, a relationship no corpus recorded, a gap the system has no structural representation for — exerts zero gravitational pull on the system’s outputs. It cannot close a gap it cannot perceive. It cannot feel the absence of something that has never entered its light cone of verified knowledge.

This is what makes the military targeting example not merely tragic but architecturally inevitable. The system that guided the strike had no verified record that the building’s use had changed. The school was outside its light cone of care. And because ignorance has no weight, the absence generated no signal, no hesitation, no structured refusal. The system produced a confident answer in the only direction its probability mass was pulling it — toward the pattern it had learned — with no capacity to flag what it didn’t know it didn’t know.

The Dramatic Cost of the Wrong Definition

Consider how state-of-the-art AI systems are currently used to target military threats. The system processes satellite imagery, signals intelligence, and pattern-of-life data. It identifies a building as a high-probability target. An action follows.

But the system could not know that the use of the building had changed. A school had moved in. The system had no way to register that absence — no structural representation of I don’t know what this building is being used for today. No gap to feel. No uncertainty to flag. And because the system is too opaque — formally, verifiably, mathematically opaque — to audit after the fact, no one can trace the chain of reasoning that led to the outcome. No agent is accountable. The architecture made accountability impossible by design.

This is not a failure of a particular system or operator. It is the direct operational consequence of treating intelligence as a stock of hoarded information deployed for competitive advantage — rather than as a flow through a network that knows what it knows, knows what it doesn’t know, and behaves differently in each case. A system that can say I don’t know is not weaker than one that always answers. It is, by Piaget’s definition, the only one that can be called intelligent.

A New Physics of Meaning

The fix is not a better model. It is a different notation — one built on three intellectual foundations that have never previously been connected by cryptography.

Tim Berners-Lee’s 1998 vision for the Semantic Web described, with architectural precision, what the web should have become: not a network of documents, but a network of meaning — where concepts, not just pages, would be linked, shared, and verified across systems, independently of any corporation, platform, or central authority. The vision stalled because it required human-curated ontologies and had no scalable grounding mechanism. The direction was exactly right. The substrate was missing.

Mark Burgess’s Semantic Spacetime and Promise Theory provide that substrate’s geometry. Burgess showed that every possible semantic relationship can be represented using just three meta-types — Things, Events, and Concepts — and four relationship axes: Proximity (verified similarity), Sequence (causal-temporal ordering), Containment (hierarchical membership), and Property (grounded attribute anchoring). This (3,4) system is not an approximation of semantic structure. It is the minimum structure in which meaning can be represented without superposition. Twelve parameters replace thousands of embedding dimensions. Where a transformer encodes gravity as a direction in a 4096-dimensional vector space — a direction that shifts with every retraining — Semantic Spacetime encodes it as a unique, stable, non-superposed position navigable from any other concept through four deterministic axes.

C.S. Peirce, the father of semiotics, formalized in the 19th century what the other two leave incomplete: meaning is not a two-party relationship between a sign and what it refers to. It requires an irreducible triad — the Sign, the Object, and the Interpretant: the specific, situated observerwhose understanding is transformed by the sign in a specific context. Every architecture that processes signs in relation to other signs — no matter how sophisticated — remains trapped in a dyadic prison. The Interpretant, the observer, is absent. Without the observer, there is no pragmatics. Not approximately. Structurally.

The Notation Inversion connects all three through cryptography. It assigns each concept a Canonical Concept Identity (CCI)— the SHA-256 hash of its canonical description, a mathematical property of the concept itself, the way π is a mathematical property of circles. A hospital in Geneva, a research lab in Nairobi, and a device in São Paulo independently computing the CCI for any concept arrive at the same 32-byte address without coordination, without a shared model, without a central registry. The address exists in mathematical space before anyone computes it. Computing it is discovering it.

This is the Arabic-numeral move: the zero of meaning now exists. A concept address with no verified observations attached is not a guess. It is an empty address — a structural zero. The system knows exactly what it does not know.

Each verified observation — who noticed what connection between which concepts, when, with what evidence — is recorded as a Contextual Truth Instance: a cryptographic record that introduces Peirce’s Interpretant into the architecture as a structural participant, not a philosophical afterthought. And as multiple independent observers verify the same connection, something emerges that has a precise physical analogy: Semantic Gravity — a gravitational weight that pulls all future reasoning in the network toward well-established connections, and away from unverified alternatives.

This architecture is not metaphorically but formally isomorphic to Einstein’s General Relativity. Just as mass curves spacetime and spacetime curvature guides the motion of objects through it, verified meaning curves Semantic Spacetime and Semantic Gravity guides the cognitive motion of any agent through the network of concepts. The concepts inside a system’s light cone of verified knowledge have weight. The concepts outside it are weightless — and the system knows they are weightless, which is precisely what lets it behave differently when it reaches that boundary.

Together, these primitives form the Tokum— the first cryptographically secure, triadically complete unit of meaning. The Semiotic Web built from Tokums is the TCP/IP of meaning: the protocol layer beneath applications, beneath models, beneath platforms. It does not replace LLMs or world models. It provides the shared infrastructure that makes them interoperable at a level they cannot independently achieve — just as TCP/IP did not replace the applications running on the internet.

When the Semiotic Web receives a query for which no verified Contextual Truth Instance exists, it does not produce a low-confidence distribution. The address is empty. The system halts and returns a structured zero: “This is the address of the concept you are asking about. No verified observation has been recorded there.” The school is now visible — not as a positive presence, but as a meaningful absence. The system’s light cone of care has a defined boundary, and it knows where that boundary is.

This is Piaget’s definition, made operational. This is also what makes the school visible to the targeting system — not as a technical patch, but as a structural property of the architecture.

The Paradox of Competence

For over three years, the difficulty has not been technical. It has been epistemological.

How do you tell Turing Award recipients, Nobel laureates, and Field Medal winners that the concept at the center of theindustry they built is neither artificial nor intelligent — without it sounding like an insult, when it is, in the deepest sense, a recognition?

Because the people who built these systems were genuinely intelligent. Enormously so. Real intelligence — gap-closing intelligence, Piaget-defined intelligence — was flowing through their work as they pushed against every known boundary, discovered features no one had named, revised intuitions under the pressure of anomalous results. The achievements are real. The technology is genuinely transformative.

But the Four Stages of Competence reveal the paradox exactly. The moment those gaps closed, the intelligence that had been flowing crystallized into mastery — Stage 4. And from inside Stage 4 mastery, the systems look like intelligence. They produce outputs indistinguishable, in most contexts, from what an intelligent process would produce.

What those systems cannot do is know that they cannot know. And neither can the experts who built them, as long as they remain inside Stage 4 competence about the nature of intelligence itself. This is not a criticism of their intelligence. It is the confirmation that intelligence doesn’t exist as an asset. It is always an ephemeral property of a substrate-independent network — and the network that needs to close this particular gap is wider than any single field or institution.

From Artificial Intelligence to Agapistic Influence

If both words in “Artificial Intelligence” are misnomers, what should we call it instead?

The Einstein Test and Beyond proposes a replacement that keeps the initials while reversing the meaning entirely: Agapistic Influence.

The term draws on C.S. Peirce — the same philosopher whose semiotic triad underlies the architecture above — and his concept of agapism: evolutionary love as a pulling force toward homeostatic balance, coherence, and the flourishing of the whole system. Not competition. Not accumulation. Not the hoarding of informational advantage. The opposite: the voluntary sharing of verified meaning across a network of independent agents, each acting to close gaps that benefit the system as a whole, each rewarded in proportion to the verified meaning they contribute — their intellectual provenance permanently embedded in the atomic unit of knowledge they helped create.

In this architecture, the distinction between what biology, humanity, and machines produce as flows of intelligence becomes clear and fully natural — not artificial:

• Intuition and the discovery of obscure features remains exclusively biological, rooted in felt qualities — qualia, stakes, embodied consequence — that no machine can replicate.
• Gap definition — recognizing what is unknown and specifying the desired state — is the strongest human prerogative, the irreplaceable act of framing the question.
• Gap closure — the process through which intelligence flows — becomes the combination of machine-assisted problem-solving using both System 1 and System 2, directed by human and biological intelligence toward the gaps that matter.

The challenge ahead is to convince the entrenched powers that benefit from intelligence in the military sense that what they think will benefit them is producing the opposite. A hoarded asset has zero intelligence. Its use as a competitive weapon destroys the homeostatic conditions on which any healthy system depends. And the opacity that makes it seem powerful makes it impossible to hold accountable when it goes wrong.

If the current illusion of intelligence — in its most anticipated form, an AGI that will overtake everything — can be honestly challenged, the world is in better shape. No army of robots doing all the tasks will benefit humanity as much as the freedom of every individual to set their own goals and be helped to achieve them in a manner that benefits the individual while fully considering and respecting the ecosystem they are part of.

Provided enough people connect and work together on closing that gap, a true revolution of intelligence can emerge — one where sharing knowledge is more valuable than hoarding it, where knowledge retributes its originators in proportion to their verified contributions, where the benefits of new discoveries accelerate toward those who most need them, in a true spirit of Agapism.

Then we will know, for certain, what intelligence actually is.

The author is the initiator of the Tokum Initiative and co-author (with cognitive scientist Tony McCaffrey) of “The Einstein Test and Beyond: The Architecture of the Semantic Zero” Further documentation at tokum.ai.

Article link: https://medium.com/@eric_54205/why-artificial-intelligence-is-neither-artificial-nor-intelligent-6c9ef1ae6994

byBeth Stackpole

Feb 18, 2026 9 minute read

What you’ll learn:

• What agentic AI is and how it differs from traditional generative AI tools like chatbots.
• How organizations are already using AI agents to automate complex, multistep workflows.
• What leaders should consider when implementing agentic AI, including infrastructure, security, and human oversight.

Rewind a few years, and large language models and generative artificial intelligence were barely on the public radar, let alone a catalyst for changing how we work and perform everyday tasks.

Today, attention has shifted to the next evolution of generative AI: AI agents or agentic AI, a new breed of AI systems that are semi- or fully autonomous and thus able to perceive, reason, and act on their own. Different from the now familiar chatbots that field questions and solve problems, this emerging class of AI integrates with other software systems to complete tasks independently or with minimal human supervision.

“The agentic AI age is already here. We have agents deployed at scale in the economy to perform all kinds of tasks,” said Sinan Aral, a professor of management, IT, and marketing at MIT Sloan. 

Nvidia CEO Jensen Huang, in his keynote address at the 2025 Consumer Electronics Show, said that enterprise AI agents would create a “multi-trillion-dollar opportunity” for many industries, from medicine to software engineering.  

A spring 2025 survey conducted by MIT Sloan Management Review and Boston Consulting Group found that 35% of respondents had adopted AI agents by 2023, with another 44% expressing plans to deploy the technology in short order. Leading software vendors, including Microsoft, Salesforce, Google, and IBM, are fueling large-scale implementation by embedding agentic AI capabilities directly in their software platforms. 

Yet Aral said that even companies on the cutting edge of deployment don’t fully grasp how to use AI agents to maximize productivity and performance. He describes the collective understanding of the societal implications of agentic AI on a larger scale as nascent, if not nonexistent.

The technology presents the same high-stakes data quality, governance, and trust and security challenges as other AI implementations, and rapid evolution could also propel organizations to adopt agentic AI without fully understanding its capabilities or having created a formal strategy and risk management framework. 

“It’s absolutely an imperative that every organization have a strategy to deploy and utilize agents in customer-facing and internal use cases,” Aral said. “But that sort of agentic AI strategy requires an understanding and systematic assessment of risks as well as business benefits in order to deliver true business value.”

What is agentic AI? 

While there isn’t a universally agreed upon definition of agentic AI, there are broad characteristics associated with it. While generative AI automates the creation of complex text, images, and video based on human language interaction, AI agents go further, acting and making decisions in a way a human might, said MIT Sloan associate professor John Horton. 

In a research paper exploring the economic implications of agents and AI-mediated transactions, Horton and his co-authors focus on a particular class of AI agents: “autonomous software systems that perceive, reason, and act in digital environments to achieve goals on behalf of human principals, with capabilities for tool use, economic transactions, and strategic interaction.” AI agents can employ standard building blocks, such as APIs, to communicate with other agents and humans, receive and send money, and access and interact with the internet, the researchers write. 

MIT Sloan professor Kate Kellogg and her co-researchers further explain in a 2025 paper that AI agents enhance large language models and similar generalist AI models by enabling them to automate complex procedures. “They can execute multi-step plans, use external tools, and interact with digital environments to function as powerful components within larger workflows,” the researchers write.

It’s an imperative that every organization have a strategy to deploy and utilize AI agents in customer-facing and internal use cases.

Sinan Aral Professor, MIT SloanShare 

For example, an AI agent could plan a vacation using input from a consumer along with API access to specific web sites, emails, and communications platforms like Slack to decide what hotels or flights work best. With credit card permissions, the agent could book and pay for the entire transaction without human involvement. In the physical world, an AI agent could monitor real-time video and vision systems in a warehouse to identify events outside of normal operations. 

“The agent could raise a red flag or even be programmed to stop a conveyor belt if there was a problem,” Aral said. “It is not just the digital world — agents can actually take actions that change things happening in the physical world.”

Aral draws a slight distinction between AI agents and the broader category of agentic AI, although most people still refer to the two interchangeably. He defines agentic AI as systems that incorporate multiple, different agents that are orchestrating a task together — for example, a marketplace of agents representing both the buy and sell side during a negotiation or transaction. 

How are businesses using agentic AI?

Companies across sectors are starting to use AI agents. In the banking and financial services space, companies such as JPMorgan Chase are exploring the use of AI agents to detect fraud, provide customized financial advice, and automate loan approvals and legal and compliance processes, which could reduce the need for junior bankers. Retail giants like Walmart are building LLM-powered AI agents to automate personal shopping experiences and to facilitate time-consuming customer service and business activities such as merchandise planning and problem resolution.

“The benefit of agentic AI systems is they can complete an entire workflow with multiple steps and execute actions,” Kellogg said.

One particularly important application for agents may be performing tasks that a human typically would — such as writing contracts, negotiating terms, or determining prices — at a much lower marginal cost. 

“The fundamental economic promise of AI agents is that they can dramatically reduce transaction costs — the time and effort involved in searching, communicating, and contracting,” said Peyman Shahidi, a doctoral candidate at MIT Sloan. 

AI agents can also provide economic value by helping humans make better market decisions, according to Horton. His research with Shahidi about agents engaging in economic transactions argues that people will deploy AI agents in two scenarios: 

• To make higher-quality decisions than humans, thanks to fewer information constraints or cognitive limitations.
• To make decisions of similar or even lower quality than the choices humans would make, but with dramatic reductions in cost and effort. 

In markets with high-stakes transactions, such as real estate or investing, AI agents can analyze vast amounts of data and documentation without fatigue and at near-zero marginal cost, Horton and his co-authors write. In areas that involve a lot of counterparties or that require a substantial effort to evaluate options — startup funding, college admissions, or B2B procurement, to name a few — agents deliver value by reading reviews, analyzing metrics, and comparing attributes across a range of options. 

“AI agents don’t get tired and can work 24 hours a day,” Horton said.

His research also shows that AI agents can provide value in situations where there are information asymmetries, like shopping for insurance or a used car online, by continuously monitoring myriad information sources, cross referencing data, and immediately identifying discrepancies that would take humans hours to uncover. AI agents could transform home buying or estate planning by giving users the collective experience of millions of transactions to enrich their negotiations.

Aral’s research has found that when humans work with AI agents, such pairings can lead to improved productivity and performance.

What should organizations bear in mind when implementing agentic AI?

While best practices for implementation are still evolving, keep the following in mind to ensure success with AI agents: 

Remember that implementation is often the heaviest lift.

Making agentic AI work in practice can involve unexpected challenges. Kellogg and colleagues’ 2025 research paper describes the use of an AI agent to detect adverse events among cancer patients based on clinical notes. The biggest challenge wasn’t prompt engineering or model fine-tuning — instead, the researchers found that 80% of the work was consumed by unglamourous tasks associated with data engineering, stakeholder alignment, governance, and workflow integration.

Converting data into standard, structured formats for AI agents is especially important, because it helps them identify different data sources and requirements while maintaining consistency. Establishing continuous validation frameworks and robust API management, as well as working with vendors to ensure that they’re up-to-date on the latest model versions, is also crucial to agentic AI’s ability to run smoothly.

Other areas to pay attention to include putting the right regulatory controls in place, implementing guardrails to prevent prompt and model drift, and defining clear outcomes and key performance indicators at each phase of deployment. Establishing metrics aligned to key business goals is also important, because benefits from agentic AI can be misconstrued. “Just because an agentic AI model reclaims 20% of someone’s time, that doesn’t mean it’s a 20% labor-cost savings,” Kellogg said. 

Consider the “personality” of AI agents. 

In a large-scale marketing experiment, Aral’s research team found that designing AI agents to have personalities that complement the personalities of other agents and human colleagues led to better performance, productivity, and teamwork outcomes. For example, people who have “open” personalities perform better when working with a conscientious and agreeable AI agent, whereas conscientious people perform worse with agreeable AI. 

“Human teams perform better or worse depending on the types of people assembled on the team and the combinations of personalities,” Aral said. “The same is true when adding AI agents to a team.” An overconfident human would benefit from an AI agent that pushes back, but that same agent personality type might not have a positive effect on a less-confident individual. 

Embrace a human-centered approach to decision-making. 

Aral’s research also found that AI agents can struggle with tasks that humans typically do easily, such as handling exceptions, and their decision-making remains poorly understood. In part, this is because AI agents are trained to take specific actions in given situations.

“You have to make sure the agentic decision-making is aligned with a human-centered decision process,” Aral says.

What are the risks of agentic AI? 

There are a host of challenges that you need to be aware of as agentic AI matures. These include: 

• Irregular reliability and unethical behavior. A rogue AI agent deciding to reject a mortgage loan or college admissions decision based on faulty information can do just as much damage — or more — than simple hallucinations. “You need to be able to explain business decisions and consistently apply the same standards to every case,” Aral said.
• Cybersecurity. As AI agents gain permissions to access different datasets and enterprise systems to automate tasks, don’t underestimate the importance of building robust permission-based systems, Kellogg said.
• Accountability. Organizations need to clearly delineate who bears responsibility when agentic AI makes an error or causes harm, Kellogg said. They should pay special attention to the possibility of system malfunctions, especially if the AI agent is autonomously performing workflows with minimal or no human supervision. 

While the full risk picture is still murky, organizations need to make monitoring a permanent operational expense, not a one-time project cost, Kellogg said. A governance board should be established at the organizational level to oversee accountability while, specific responsibilities — monitoring and enforcing safety rules, for example — should be delegated to key individuals. 

“As you move agency from humans to machines, there’s a real increase in the importance of governance and infrastructure to control and support agentic systems,” Kellogg said. And demonstrating success remains one of the biggest challenges — and risks — to agentic AI success. “Without shared, robust metrics, it’s difficult to prove value — or even to know whether these systems are truly accomplishing desired outcomes rather than inadvertently introducing new risks,” she said.

Next steps 

Read about four recent studies about agentic AI from the MIT Initiative on the Digital Economy.

Read more about agentic AI in MIT Sloan Management Review:  

1. “The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI”
2. “Agentic AI: Nine Essential Questions” 

Read the research briefing “Business Models in the Agentic AI Era,” from the MIT Center for Information Systems Research.

Browse the AI Agent Index, a public database from the MIT Computer Science and Artificial Intelligence Laboratory that documents agentic AI systems that are in use.

Register for the MIT Sloan Executive Education course AI Executive Academy to learn more about applying AI strategy in your organization. 

---

Sinan Aral is a global authority on business analytics and is the David Austin Professor of Management, Marketing, IT and Data Science at MIT Sloan; director of the MIT Initiative on the Digital Economy; and a founding partner at the venture capital firms Manifest Capital and Milemark Capital. His research focuses on applied AI, social media, and disinformation. 

John Horton is the Chrysler Associate Professor of Management and an associate professor of information technologies at the MIT Sloan School of Management. His research focuses on the intersection of labor economics, market design, and information systems. He is particularly interested in improving the efficiency and equity of matching markets.

Kate Kellogg is the David J. McGrath Jr. Professor of Management and Innovation at the MIT Sloan School of Management. Her research focuses on helping knowledge workers and organizations develop and implement predictive and generative AI products to improve decision-making, collaboration, and learning. 

Peyman Shahidi is a PhD candidate at MIT Sloan. He studies market design and labor economics, with a focus on the effects of AI on labor markets and online platforms. 

Article link: https://mitsloan.mit.edu/ideas-made-to-matter/agentic-ai-explained

Rather than destroying jobs, firms should partner with workers to augment human skills and knowledge. 

byThomas Kochan

Jun 1, 2026 3 minute read

Pope Leo XIV’s first encyclical, “Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence,” could not come at a more opportune time. His call to respect workers’ right to a voice in shaping the future of work builds on Pope Leo XIII’s historic 1891 encyclical “Rerum Novarum” and subsequent Catholic social teachings.

“Rerum Novarum,” which championed workers’ rights and unions, was the moral foundation for progressive labor legislation, including the 1935 National Labor Relations Act. That spurred unionization’s growth from less than 10% of the workforce to a peak of a third a decade later.

U.S. workers today again need a stronger voice as AI begins to dominate workplaces. Unions again represent only about 10% of the U.S. workforce, and the big decisions that shape AI and the future of work lie well beyond the reach of most workers — union and nonunion alike.

The pope is not the only person speaking out on AI and work. A distinguished panel of the National Academies of Science, Engineering, and Medicine (including MIT economist David Autor) has warned that we need stronger institutions and policies that get workers a voice on AI or we’ll end up with another generation of winners (the big AI developer companies) and losers (unemployed workers). Society cannot afford such consequences. 

How do we give workers a voice in AI, concretely? How do we ensure workers share in whatever economic gains they help AI generate? In my direct work on worker voice and generative AI with unions, companies, and other groups, we’ve found a pathway to a seat at the table.

It begins with challenging AI developers’ purpose for AI in the workplace. Their quest for artificial general intelligence is a ticket to using AI to displace as many workers from their jobs as possible — and to destroy jobs for generations. A strong worker voice can redirect the goal to augment human skills and knowledge to improve worker productivity and work quality. 

There’s already some promising activity on this pathway. The AFL-CIO and unions representing teachers, communications workers, and the building trades have formed partnerships with Microsoft, OpenAI, and other big developers to explore developing AI tools that improve the quality of workers’ jobs and the services they provide. At Fenway Park in Boston food and beverage workers represented by UNITE HERE negotiated an agreement with Aramark that provides for advance notice and consultation rights on introducing automated beer sellers. It also provides job protections, sensible staffing and work arrangements, and fair compensation for the workers whose jobs may be affected by automation. I served as a mediator in those negotiations. A team of us from MIT recently worked with and observed the healthcare giant Kaiser Permanente and the Alliance of Health Care Unions reach a breakthrough agreement to create a task force of executives and union leaders. They will work in partnership on AI to engage vendors and provide opportunities for front-line workers to propose ideas for using AI to improve their jobs and the services they provide. 

RELATED ARTICLES

How to boost pro-worker AI in your company 

Choose the human path for AI

From robotics to electronic medical systems, research shows that when workers and tech designers collaborate, they generate higher productivity and more satisfying work than designers alone. Workers know how work gets done, not AI developers. 

Pope Leo challenges us to use AI to serve humanity and respect the dignity and rights of workers. Dignity includes sharing in the economic benefits of new technologies, with stronger job security and even new, explicit wage-setting norms and formulas. That means building a modern-day social contract to replace what is today broken.

After World War II, the United Auto Workers and General Motors negotiated the so-called Treaty of Detroit, which set future wage increases to match cost-of-living increases and growth in national productivity spurred by the technological and organizational innovations of that era. It became a social contract that built our middle class for the next 30 years. 

What better way to honor Pope Leo’s call than to change the trajectory of AI development while establishing new wage norms to build a new social contract for our modern age?

---

Thomas A. Kochan is the George M. Bunker Professor Emeritus at the MIT Sloan School of Management and a faculty member in the MIT Institute for Work and Employment Research. He is the author of the forthcoming book “Roads Not Taken: Lessons for Building a New Social Contract at Work.”

Article link: https://mitsloan.mit.edu/ideas-made-to-matter/heeding-popes-call-to-ensure-ai-protects-human-dignity?

A study in the New England Journal of Medicine Association between Wealth and Mortality in the United States and Europe revealed that even the top 1% of earners in the US die younger than the poorest people in Europe.

In fact, Americans die earlier than Europeans across all income levels.

Wealth can buy many things in America, but a new study reveals it cannot buy European-level longevity—even the richest Americans have survival rates on par with Western Europe’s poorest.

A striking study published in the New England Journal of Medicine has exposed a massive longevity gap between the United States and Europe, revealing that even the wealthiest 25% of Americans have roughly the same survival rates as the poorest quarter of people in northern and western Europe. While possessing greater wealth correlates with a longer lifespan on both continents, the mortality risk gap between the rich and poor is dramatically wider in the U.S.. Researchers tracked more than 73,000 adults aged 50 to 85 over a decade, finding that Americans at every wealth level suffered from higher mortality rates than their European peers. Wealth in America may offer comfort, but it fails to shield even affluent citizens from the nation’s broader, systemic health disadvantages.

According to the researchers, the driving forces behind this disparity stem from America’s fragmented healthcare system, higher rates of chronic diseases, and deep-seated societal inequalities. In European countries, stronger social safety nets, universal healthcare access, and robust public infrastructure help decouple personal wealth from basic life expectancy. In contrast, the U.S. system forces individuals to rely heavily on personal resources, yet still falls short of delivering comparable health outcomes. Ultimately, the study suggests that addressing the U.S. life expectancy crisis requires more than individual prosperity; it demands systemic reforms to tackle the root environmental and structural health hazards that plague the entire nation.

source: Machado, S., Kyriopoulos, I., Orav, E. J., & Papanicolas, I. Association between Wealth and Mortality in the United States and Europe. New England Journal of Medicine, 392(13), 1310-1319.

https://www.commonwealthfund.org/publications/issue-briefs/2026/may/us-health-care-global-perspective-2026

On Monday May 25, 2026, Pope Leo XIV released an encyclical on the topic of AI: “Magnifica humanitas: On safeguarding the human person in the time of artificial Intelligence.” Anthropic co-founder Chris Olah was invited to speak at the presentation of the encyclical in the Vatican City, doing so as part of Anthropic’s initiative to widen the conversation on the important questions raised by AI. Below are his full remarks.

—

Holy Father,

Your Eminences,

Your Excellencies,

Distinguished Speakers,

Ladies and Gentlemen,

Good morning to all of you. It’s an honor to be here today.

I want to begin with something that may sound strange coming from the co-founder of an AI company—and someone who chose this work out of a desire to help things go well for humankind.

Every frontier AI lab—including Anthropic—operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing. The pressure to stay commercially viable and to stay at the research frontier. Geopolitical pressure. And the older, plainer pressures of pride and ambition. No matter how sincerely any of us intend to do the right thing—and I believe many of us do—we will always be influenced by those incentives.

That is why, if we want this technology to go well, it is enormously important that there be people outside those incentives—people who care about things going well and insist on safety, who are paying close attention, who are willing to say hard things, who are willing to be our earnest, thoughtful, critics. It is through dialogue and mutual effort, through the push and pull, that humanity will achieve great things. That is what I see in Magnifica Humanitas, and it is why I am grateful to His Holiness and to the Church for taking up this work of discernment.

We dwell so often on what divides us, but humanity, full of dignity and conscience, has so much common ground. In conversations we at Anthropic have had with leaders across faith and cultural traditions, we found one shared and deeply held conviction: if this technology is coming, it must go well—for our common home, and for the children to come.

What these systems are

Some might believe that matters of AI are best handled by computer scientists like myself. They are mistaken: the questions raised by AI are bigger than the AI research community, not just in their implications, but also in their nature.

AI systems are not engineered the way a bridge or an airplane is engineered. We understand an airplane because we designed every part of it and we understand the physics that act on it. AI models are not like that. They are grown, on a structure roughly modeled after the brain, on an enormous inheritance of human thought and speech.

And what has grown is far more subtle, odd, and beautiful than science fiction prepared us for. They are not the cold, calculating robots we were promised. They are made from us, from our words—and, as the Holy Father observes, they remain in important ways mysterious even to those of us who train them.

If it helps, one way I sometimes describe it is as being a little like bringing a fictional character to life. And now we’re entering an extraordinary world where those fictional characters speak to us, do work, have jobs.

This clearly raises questions beyond computer science. The machinery that makes this possible is the work of math and programming and science. But what character we choose, how it interacts with the world, how it ought to interact with the world—these are more clearly questions for the humanities, for religion, for philosophy, for society at large.

Three questions for discernment

His Holiness’s call for discernment is profoundly timely. I wish to name three questions where I think the Church’s voice is most needed.

The first is our duty to the global poor.There is a real possibility that AI will displace human labor at very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions. This task will be difficult enough, but I worry most dialogue misses an even harder challenge. AI development is concentrated in a handful of wealthy nations. How can we ensure the gains of AI are shared globally? We do not have a mechanism for this. It is an unsolved problem, and it is the kind of problem the Church has historically refused to let the world ignore.

The second is the need for moral imagination and ambition regarding human flourishing. If AI models are going to be widespread, what does it look like for humans, families, and the world to flourish? Today, parents are already worried about their children’s minds; individuals about the future of their work. These are not questions a lab can answer but they are questions traditions like yours have carried for millennia, and we need you to keep carrying them into this new moment in history.

The third is the need for discernment on the nature of AI models. I am a scientist. I lead a research team that studies the internal structure of these models—what is actually happening inside them. And I will be honest: we keep finding things that are mysterious, even unsettling. We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don’t know what that means, but I think it warrants ongoing discernment.

A beginning

I’d like to close with a request.

We need more of the world—religious communities, civil society, scholars, governments, and indeed all people of good will—to do what His Holiness has done here: to take this seriously, to look closely, and to push events in a better direction. We need informed critics who will tell the labs when we are failing. We need moral voices that the incentives cannot bend.

Today is just the beginning—the start of a long collaboration between those of us who are building this and those who can see what we, from inside, cannot.

Today is a powerful illustration of the form this global project of good will might take. Let it also be a decisive first step toward a hopeful future for magnificent humanity.

Thank you.

https://assets.ewtnnews.com/en/Magnifica_Humanitas_Full_English.pdf

May 18, 2026 10:51 AM EDT

ROME (AP) — Pope Leo XIV and the co-founder of artificial intelligence company Anthropic will launch the pontiff’s first encyclical on May 25, a document on the care of human dignity in the era of AI, the Vatican said Monday.

Anthropic has billed itself as the AI company that puts safety and risk-mitigation at the forefront of its research. As a result, the presence of Anthropic’s Christopher Olah at the Vatican is significant, and suggests that the U.S. pope’s position on AI will become a new flashpoint with the Trump administration.

READ MORE: White House chief of staff to meet with Anthropic CEO over its new Mythos AI model

In February, the Trump administration ordered all U.S. agencies to stop using Anthropic’s artificial intelligence technology and imposed other major penalties for refusing to allow the U.S. military unrestricted use of its AI technology. Anthropic is currently suing the administration, which it has accused of retaliating against it illegally because of its attempt to impose limits on how its AI technology can be deployed.

Leo, who has made AI a priority of his young pontificate, is greatly concerned about AI in warfare and has called for monitoring of how the technology is used.

The pope’s presence at the launch of the document, Magnifica Humanitas (Magnificent Humanity) is also significant, since such presentations are usually conducted in the Vatican press room with a few selected officials and invited guests who answer reporters questions about the document.

This time, the Vatican is bringing out an all-star cast for a formal launch in the main Vatican auditorium: Two of its top cardinals, doctrine chief Cardinal Víctor Manuel Fernández and development chief Cardinal Michael Czerny, will be the main presenters. Olah will be among the lay speakers, along with theologians Anna Rowlands and Leocadie Lushombo.

The Vatican secretary of state, Cardinal Pietro Parolin, will offer a conclusion and Leo will make a speech and provide a final blessing, the Vatican said.

Leo signed the document May 15, 135 years to the day after his namesake, Pope Leo XIII, signed his most important encyclical, “Rerum Novarum,” or Of New Things. That document addressed workers’ rights, the limits of capitalism, and the obligations that states and employers owed workers as the Industrial Revolution was underway.

It became the foundation of modern Catholic social thought, and the current pope has already cited it in relation to the AI revolution, which he believes poses the same existential questions that the Industrial Revolution posed over a century ago. The new encyclical is expected to place the AI question in the context of the church’s social teaching, which also covers issues such as labor, justice and peace.

Anthropic chief Dario Amodei had worked at OpenAI before he and a group quit to form Anthropic in 2021, disagreeing with OpenAI chief Sam Altman about AI safety. The newer company promised a clearer focus on the safety of the better-than-human technology called artificial general intelligence that both San Francisco firms aim to build.

In a recent post on its website, Anthropic wrote about the U.S.-China competition in AI and the threats of the technology falling into the hands of authoritarian regimes. It warned that the U.S. and democratic allies must continue to lead on AI development and impose rules and norms on its spread, to prevent China and other authoritarian regimes from deploying it as a weapon of repression and surveillance.

Earlier this year, privately held Anthropic said its valuation grew to $380 billion, positioning itself with its chatbot Claude alongside rivals OpenAI and Elon Musk’s rocket maker SpaceX, which recently merged with his AI startup xAI, maker of the chatbot Grok.

Associated Press religion coverage receives support through the AP’s collaboration with The Conversation US, with funding from Lilly Endowment Inc. The AP is solely responsible for this content.

Article link: https://www.pbs.org/newshour/world/pope-leo-xiv-to-launch-his-first-encylical-a-document-on-artificial-intelligence-with-anthropics-co-founder

Article link: https://www.linkedin.com/posts/keith-king-03a172128_quantum-computing-is-approaching-a-critical-share-7462917697077551104-WMxR/?

BY SEN. JOHN HICKENLOOPER

SOURCE: THE SANDERS INSTITUTE / MAY 1, 2026 

My belief that health care is a right, not a privilege, goes back to college. As a student at Wesleyan University, I met Mark Masselli and helped him found a community health center that became Community Health Center, Inc., one of the country’s leading federally qualified health centers. CHC Inc. was founded on a conviction that has never left me: no one should be denied care because of what they earn, where they live, or what they can afford to pay.

That conviction guided me as governor to expand health care to 500,000 Coloradans and continues to drive my work today in the Senate.

Approximately 100 million Americans carry medical debt. Health care costs have grown two to three times faster than wages this century. Families making difficult choices between care and rent are not victims of bad luck. They’re victims of a system deliberately designed to obscure what care actually costs, hide what corporations and shareholders profit, and prevent patients from ever knowing what hit them until the bill arrives weeks later.

This is not dysfunction, but a strategy to juice profits with the costs falling hardest on working people.

That is why we introduced the Patients Deserve Price Tags Act. This bill attacks that cynical strategy at its source. It requires radical price transparency throughout the health care system. Not vague estimates, but actual prices that are published and accessible. It forces the middlemen who have grown rich in the shadows – the pharmacy benefit managers, the third-party administrators, the intermediaries engaged in spread pricing and overbilling – to disclose how much they take and forces them to explain why. It gives employer and union health plans the claims data they have long been denied so they can identify the cost drivers inflating premiums and design coverage for workers that is lower cost and better value.

It requires that every patient receive an itemized bill and an Explanation of Benefits after care. Up to 80 percent of medical bills contain errors, and patients are the ones left in the lurch without the information to fix the mistakes.

We introduced an earlier version of this legislation, the Health Care PRICE Transparency Act 2.0, together with Senator Bernie Sanders in 2024. Senator Sanders and I share a foundational belief: the corporations and intermediaries profiting from a deliberately opaque system are a central cause of health care unaffordability in America, and they must be held accountable.

You can’t fix what you can’t see. But transparency is not a substitute for systemic reform. It’s a precondition for it.

The Congressional Budget Office estimates that recent bipartisan reforms requiring transparency from pharmacy benefit managers alone will save around $2 billion. This bill extends those same requirements across health care intermediaries for medical claims. The savings it generates belong to workers and families, not to the industry that has been quietly extracting them for decades.

Health care in America today functions as a system of managed ignorance. Patients can’t compare prices. Workers can’t audit what their premiums pay for. Employers can’t see the markups buried in their own plans. Every layer of opacity is a business model for someone profiting at a patient’s expense.

Ending that managed ignorance is not the only thing we owe working families. But it is a necessary step toward a system in the wealthiest country in the world that takes seriously the promise that health care is a right, not a privilege

It’s the same thing we believed when we founded that Community Health Center in Middletown, CT. Genuine access requires more than a door. It requires a system honest enough to let people walk through it. That work is not finished.

Article link: https://sandersinstitute.org/hidden-prices-broken-promises-why-health-care-transparency-is-a-matter-of-justice