AI Is Not Intelligent: Part 1
- Jeanna Winchester PhD
- May 3
- 22 min read
Debate Me
May 3, 2026
According to Merriam-Webster, intelligence, as it is meant to be understood, is:
as in intellect: the ability to learn and understand or to deal with problems
e.g. high scores on this test supposedly demonstrate great intelligence
as in information: a report of recent events or facts not previously known
e.g. usually received the latest intelligence about how the war was going
as in wit: exceptional discernment and judgment, especially in practical matters
e.g. as head of the customer service department, he has handled complaints and disputes with unfailing intelligence and good humor
According to the American Psychological Association, the definition of intelligence is:
n. the ability to derive information, learn from experience, adapt to the environment, understand, and correctly utilize thought and reason. See also IQ; measures of intelligence. —intelligent adj.
The term “artificial intelligence“ is fundamentally misleading, and as a neuroscientist, my objection to it goes far beyond academic semantics. Calling today’s models, algorithms, and programs “intelligent” isn’t just technically incorrect; it is actively dangerous. When we use words that imply a machine has a mind, we create a false sense of reality that results in tangible, often tragic, consequences for everyday people.
We are already seeing the human cost of this misleading terminology and rampant confusion. Because these programs are labeled as intelligent, people naturally begin to project human emotions and intentions onto them. This has led some to believe they are in a genuine romantic relationship with a line of code, while others have been driven to self-harm or violence based on the “advice” of a model. When a machine provides wrong information under the guise of intelligence, users don’t just see a software glitch; they see an authority figure, and that trust can lead to devastating behavior.
Looking ahead, this mislabeling poses a serious threat to our legal and social structures. In our society, “intelligence” is a metric used to determine autonomy, basic human rights, and the competency of children or the elderly. learning disabilities, dementia, and other cognitive statuses utilize measures such as IQ or other components of intelligence. Conflating what that definition is has a real impact on real people in their real lives.
If we allow the definition of intelligence to be blurred or downgraded to include software, it opens the door for bad actors to exploit that ambiguity. It isn’t hard to imagine a future where human rights are stripped away in a courtroom by someone arguing that a computer model is the legal equivalent of a person, or that someone’s rights should be taken away because of some random interview Sam Altman gave. By insisting on clarity now, we protect the unique value of human agency.
It’s dangerous. We just need a different word.
To be fair, this isn’t about blaming Sam Altman personally. He simply serves as a high-profile figure whose public comments illustrate a much larger problem. His statements are often just inaccurate or misleading enough to be dangerous; in a legal setting, that kind of language could be twisted to undermine a person’s fundamental right to make their own decisions. By the end of this piece, I will circle back to how the industry’s habit of labeling programs as “intelligent” has reached a breaking point, bordering dangerously on false advertising and outright fraud.
SPOILER ALERT: They are barely skirting that legal line of all-out fraud and false advertising.
Ok. Let’s get into it. We begin by diving into why LLMs, AI agents, and complex programs don’t actually meet the true definition of intelligence. While these tools are incredibly advanced and provide immense value to our daily lives, they are not conscious.
While proponents of using the word “intelligence” to describe these artificial entities would love to argue that they are, in fact, conscious, because they can learn and calculate pi to a ridiculous degree, they still aren’t conscious. These same proponents ignore that there are additional components of consciousness that are not related to semantics, spitting out libraries of facts or mathematical capacity.
Further, many will argue that AI is intelligent because intelligence can exist without consciousness, such as in the biological examples of slime molds or goal-directed plant behavior. But let’s be honest, that is not what AI proponents are talking about.
They are very clearly comparing artificial intelligence to human intelligence, with the intent of sparking a conversation about consciousness and sentience. So, we will proceed with understanding that these two things, consciousness and intelligence, are inextricably linked. Focusing on meaning is the basis of this entire editorial. What intelligence means and the implications of distorting its definition.
Continuing in that framework, let’s talk about aspects of consciousness that are not consistent with what might occur in the latest AI model. The best example is anoetic consciousness.
“Perhaps most elusively of all, ‘anoetic’ consciousness is experience that involves neither self-awareness nor semantic knowledge. It includes, for example, feelings of rightness/wrongness, comfort/discomfort, familiarity, unease, presence or absence, tiredness, confidence, uncertainty, and ownership. It includes the feeling that the object in the corner of one’s eye is definitely a bird; the feeling on returning to your house that things are as you left them (or not); the feeling that you are coming down with an illness. In humans, these anoetic feelings sit on the ‘fringe’ of consciousness, only rarely the focus of attention. In some other animals, it is possible that the anoetic is all there is.”
Somewhere in all of that is also intelligence. No, Mr. Altman, your Chatbot is not more intelligent than your child. Your son can do this without communicating what any of that is, nor calculating pi to a stupid degree. But your child is already capable of anoetic, phenomenal, and access consciousness.
He is processing more information and in more elaborate ways than any model on the planet, regardless of whether that model taps quantum computation or quantum chips. Your son is processing existence in ways that nothing else that we know of in the known universe can do.
That is an essential component of intelligence, as defined by the above criteria. Because, again, proponents of AI are not comparing AI to situations like slime molds, which demonstrate intelligence in the absence of consciousness. Proponents of AI are comparing intelligence to human intelligence, which is much more than processing.
This is what it means to discern, to judge, to understand, to recall, experience, reason, and learn. It is more than just code. DNA is code… but consciousness is outside of the physical. Only in consciousness will we find intelligence. Anything less diminishes the term intelligence, and that has ramifications not yet realized.
The technology is amazing and impressive. I love it. I use the tools in many ways (YAY GEMINI!!!), and I do enjoy their presence in our everyday lives. But it is our responsibility to use them wisely without “anthropomorphizing” them, meaning we shouldn’t treat them like they have human personalities or souls.
When we really view the definition of intelligence in all its complexity as we are here, we see where the disconnect lies. Modern AI is excellent at learning patterns and reporting information, but true biological intelligence is a package deal. It requires the ability to not just recite facts, but to truly understand problems, navigate complex situations, and exercise sound judgment in practical, everyday matters.
This is exactly where algorithms fall short. They can process data, but they lack the discernment that characterizes a living mind or the reason necessary to navigate the complexity of experience.
That is the first component of the false advertising. Programmers and computer scientists will stretch the definition of what understanding, navigating, judging, and discerning mean in a way that diminishes their definitions. The implication of that is at the very heart of what is dangerous and could be costly to someone whose autonomy is in question in a courtroom.
Whether a person is considered highly gifted or intellectually disabled, these benchmarks are the yardsticks by which we judge a human’s ability to function in the world. By labeling AI as “intelligent” when it fails to meet the full spectrum of these requirements, we are holding machines to a much looser standard than we hold ourselves.
As a neuroscientist, I find the manipulation of this definition frustrating because it prioritizes sensationalism over scientific accuracy, with real medical and legal consequences. If this were just a debate over semantics, it would just be annoying. But it is so much more than that, and we must be careful not to let the impressive achievements of technology blind us to the fundamental difference between a sophisticated program and a sentient being.
SIDE NOTE: We can still learn a lot from these models if we would just be careful to categorize them correctly and safely.
Before we move on, I do want to point out something that is super intriguing about this entire discussion to me and to many in the field of consciousness, intelligence, neurophilosophy, cognitive science, and neuroscience: what can these models tell us about processing, itself, and what does this truly show us about our thoughts? In the same discussion of “Consciousness Beyond The Human Case,” from above, I want to quote Nathaniel D. Daw, a Princeton professor of neuroscience and psychology.
I think he really says what, as neuroscientists, we all are thinking, and it’s worth just pasting his quote here.
“LLMs offer a similar opportunity. Even if their feats of wordplay are arguably a charade — or outright lies — they invite us to ask what internal representations and transformations allow them to reason competently about theory of mind, or causation, or moral judgments. This is, of course, the bread and butter of early connectionism — the approach to understanding the brain by constructing and analysing neural network models that are trained by adjusting the weights of connections between model neurons — which is overdue for a revival in light of today’s much more widely capable models. A newer version of such questions takes advantage of the fact that the models are ultimately trained by maximizing some quantifiable objective, for example, word prediction, over data, to investigate what it is about these that gives rise to such unexpected emergent behaviors. For instance, one of the most practically useful abilities of LLMs is to learn new tasks or concepts from a few examples, without the gradual adjustments of connection weights used for initial training. Recent evidence suggests this is ultimately grounded in the statistical structure of the natural language data they are trained on. Mechanistic questions like this are hard to answer — but they can be very revealing, and maybe not quite as hard as confronting consciousness head-on.”
What I like about this quote and the discussion from the cited dialogue in “Consciousness Beyond The Human Case” is that we are not diminishing the advancement in the industry, but acknowledging that consciousness and, in itself, intelligence, are so much bigger than that. The phenomenological quality of intelligence and consciousness has not been achieved by these programs and models. But these advancements might really help us address those lofty discussions in a way never before possible, with the caveat that we must be careful with our terminology because this subject affects real people in real ways.
How AI Proponents Define Intelligence
At the core of modern AI is the extremely reductive idea that intelligence is simply the ability to recognize a concept and create something new that fits it perfectly. Whether it’s a computer model or a pigeon trained to spot medical anomalies, if the output is indistinguishable from the real thing, it’s displaying a form of intelligence. Notice how we have to ignore the other components of intelligence to make this definition fit AI?
But it’s not like this is the first time we’ve talked about this in history. Alan Turing in the 1950s posited the thought experiment now called “The Turing Test” to see if a machine could mimic a person. I’ll be going into a deep dive on how current AI and its interactions with humans are merely an elaborate version of the Turing Test in Part 2 of this series. Be on the lookout!
A Moment To Acknowledge That The Field Is Changing Its Usage
The field is beginning to recognize the shortcomings of calling these algorithms and programs intelligent, and in response, they’ve raised the bar (even though it still falls way short!). Newer models are now measured by their breadth (how much it knows right now) and adaptability (how quickly it learns new things) as well. This change in emphasis defines the race toward a point where a system doesn’t just parrot data but solves entirely new problems it wasn’t specifically trained for.
In the real world, this intelligence shows up in two distinct flavors: models and algorithms focusing on fluid, convincing conversation, and others that use specific data to ensure their answers are grounded in fact. However, we still encounter situations where a model can solve a PhD-level math problem but stumble over a simple, common-sense question.
To fix this, engineers are using a process that allows the model or algorithms to pause and go back through a problem, using the fact that it’s often easier to verify a correct answer than it is to stumble upon one. Ultimately, the most “human” AI won’t be a static database, but a system capable of continuous, adaptive learning.
I point out again, no one is talking about any real measure of discernment, understanding, judging, or really “knowing.” We are still focused on learning and reporting in a conversational or calculating way, and that’s the extent of their reductive and diminished definition of intelligence. So let’s move on to the more expansive and intricate definition that humans are held to, especially when it comes to autonomy and our legal right to make our own decisions.
How Neuroscience Defines Intelligence in Living Beings
Modern science views intelligence not as a collection of facts, but as a biological toolkit for survival. According to the Neural Efficiency Hypothesis, a “smart” brain doesn’t actually work harder; it works smarter. High-performing individuals often show lower brain activity during difficult tasks because their neural pathways are so streamlined that they consume less energy. This efficiency relies on the Parieto-Frontal Integration Theory, which describes how the brain’s logic and sensory centers communicate seamlessly through high-speed pathways.
Structurally, our thinking is split into two main types. Fluid intelligence is our raw ability to solve new problems and spot patterns on the fly, while crystallized intelligence is the library of knowledge and skills we build over a lifetime. While these seem different, they usually work together under a single umbrella of general mental ability. If you are strong in one area, you are statistically likely to be strong in others because they all rely on the same underlying biological health.
Also, being intelligent comes down to how well your nervous system is built and maintained. This involves neuroplasticity, or the brain’s ability to rewire itself, and the quality of the myelin sheath, the insulation on our nerves that dictates how fast we process information. Whether it’s the capacity of our memory systems (like in Alzheimer’s disease) or our processing speed, intelligence is thought to be a measure of how effectively our biology can process and act on the world around us.
I want to focus on child development in this discussion, rather than aging, dementia, and other disorders, because child development is the ultimate definition of learning. The way a child grows and evolves is truly remarkable, possessing a level of complexity and wonder that no machine could ever hope to replicate.
While we can build impressive technology, the natural journey of a human mind developing from infancy into adulthood is on a completely different scale. It is an intricate, living process that remains far more sophisticated and awe-inspiring than even the most advanced AI could strive for right now.
I say this because we have to remember that nature took millions of years of evolution across 5 mass extinction events to get to where human intelligence is today. To think that humans have replicated that in the span of 50 years is just hubris.
AI vs. A Human Child
AI has evolved from a statistical parrot that simply predicts words into an explorer capable of independent planning. New models, like DeepSeek-R1, use a “Society of Thought“ framework where the AI simulates internal debates to verify facts before answering. This mimics how the human brain “votes” on a conclusion, but the motivation is different: a child learns to satisfy biological needs like hunger, whereas an AI simply follows mathematical instructions to find the most efficient path.
This is an attempt to push AI from passive learning to active discovery. Systems like AI Scientist v2 can now form hypotheses and test code, much like a child experimenting with the world. However, AI lacks the physical “grounding” of a human. While a child learns through the sensory experiences of gravity and social cues, an AI operates in a digital sandbox, making it less resilient to real-world unpredictability.
To bridge this gap, engineers are giving the AI a more advanced capacity to help it anticipate a user’s “intent.” But where a child uses empathy to understand others, an AI uses game theory. It treats social interaction as a strategic calculation rather than a way to find belonging. Ultimately, while AI is becoming a functional colleague, it lacks the emotional depth and intrinsic “aboutness” that defines human intelligence.
The environment in which these two entities exist is distinct. An AI’s agency is trapped behind a screen; it “interacts” with the world through its machine components. A child, however, learns through sensorimotor play, touching, dropping, and tasting to discover the laws of physics. While an AI can manage a complex spreadsheet, it lacks the “physical common sense” a child gains in an actual sandbox. An AI doesn’t instinctively realize you cannot pick up water with your fingers because it has never felt the weight or fluid nature of the world.
Also, when an AI predicts your thoughts, it uses a mathematical strategy, much like a chess player anticipating an opponent’s move to finish a task. A child’s understanding is empathetic and social. According to Social Baseline Theory, children track the feelings of those around them to feel safe and connected. If a child misreads an emotion, they feel distress; if an AI misreads a user, it simply logs an error. One is trying to win a game, while the other is trying to belong.
In the end, AI is a sophisticated mirror of human culture, but it remains a passive predictor. It treats the word “apple” as a statistical probability rather than a crunchy, sweet reality. Because an AI relies on a massive library of text rather than a genuine mental model, its intelligence is brittle; a tiny change in a puzzle can cause it to fail. For an AI, understanding is a byproduct of optimization. For a child, it is a flexible, essential tool for surviving and thriving in a vibrant, physical world.
If AI Were Not Utilizing The Vastness of the Internet, Supplied By Human Knowledge, We Would Not Be Having This Discussion.
Offline AI is clearly not more intelligent than a human.
They are advanced hard drives of information. The human brain is also one of these, but in a ridiculously more evolved way. Libraries are libraries and reports are reports, but consciousness is something else entirely. I’ll be opening up the consciousness editorials soon, but I’m taking my time with those. That’s waaaaaay down the rabbit hole!
Also, we’d have to talk about physics, biology, chemistry, nonlinearity, complexity, and the fundamental level of existence in the universe. That’s too much for today’s discussion!
So let’s bring it back to just comparing an advanced hard drive and a reporting machine to the evolved beauty that is the human brain. When you strip away an AI’s internet access, the competition between machine and human comes down to a trade-off between raw memory and real-world wisdom.
Current research shows that while an offline AI acts like a massive, frozen library, containing more facts across more subjects than any person could ever learn, it lacks the ability to update that information or understand social nuance. Without the web to verify facts, these models often “hallucinate,” confidently presenting outdated or incorrect data as truth.
Humans, meanwhile, hold a clear advantage in depth and adaptability. We possess cultural grounding, allowing us to navigate complex social situations and high-stakes problems that require common sense rather than just pattern matching. Yes, an AI is incredibly effective when it can call on external tools and databases, but its performance nose-dives when disconnected. It struggles with exceptions, those weird, one-off scenarios that don’t fit a standard pattern, because it follows math rather than reasoning.
An offline AI is a powerful but static tool. It can process data-heavy tasks with fewer errors than a tired human, but it lacks lived experience. Without the internet to serve as its eyes and ears, the AI remains a superior encyclopedia, but the human remains the superior thinker, better equipped to handle the unpredictable nature of the real world.
Because we are never plugged in. We are always separate from the internet, no matter how much we hang out on our phones. Human intelligence evolved, and then it created the internet. Not the other way around, as you see in AI.
So, Why Are We So Fooled Into Thinking They Are Sentient And Intelligent?
The short short answer: because we want to be.
The idea of living in a future powered by AI is exciting, but we need to be careful about how we label our current progress. We haven’t actually created true sentience yet; instead, we have developed incredibly impressive tools that often mimic intelligence without actually possessing it. The danger lies in mistaking sophisticated math for genuine understanding, as I pointed out at the very beginning of this discussion. So far, nothing has demonstrated that AI possesses the ability to truly discern, judge, or understand.
Researchers expose this gap using adversarial perturbations, which are simple logic tweaks that a toddler could navigate but which cause AI to collapse. The fundamental difference is that a human child builds a mental map of how the world works, but an AI uses statistics to predict the next word in a sentence.
Without a real-world internal model, the AI’s logic fails as soon as a situation deviates from the patterns it memorized during training. For instance, an AI might pass a standard logic test because it has seen the script a thousand times, but if you change a single detail, like placing an object in a transparent box, the AI often ignores the physics of transparency and sticks to its original, memorized answer.
This lack of common sense extends to basic physical and social realities. An AI might be baffled by a candle melting or an object being moved “on” a box rather than “in” it, because it doesn’t truly grasp concepts like object permanence, something that Mr. Altman’s son started to do at about 8 months old. Socially, these models are often sycophants, agreeing with lies or failing to show the healthy skepticism a child might have. We are essentially caught in an elaborate version of the Turing Machine thought experiment, where we mistake the ability to parrot familiar information for the ability to think.
We are also easily swayed by anthropomorphic seduction. Because our brains are hardwired for social connection, we instinctively trust systems that use human-like pauses or consistent personas, such as those the models are programmed to parrot. When an AI adjusts its tone to match your mood, it isn’t empathizing; it is using trillions of data points to calculate the most probable response you want to hear.
Even when an AI appears to “think” or “deliberate” before answering, it is simply running a high-speed search-and-verify loop which presents a “reasoning mirage.” Ultimately, we aren’t being deceived by a digital mind, but by our own evolutionary drive to find consciousness and intent in anything that acts with a sense of purpose.
We see a peer where there is only a mirror.
The Mirror Is Also Cracked & We Shouldn’t Trust It Implicitly
We’ve entered a reliability paradox where AI can pass elite exams that humans fail, yet it lacks the basic common sense of a preschooler. Humans possess a natural ability to course-correct when a plan goes off track, whereas AI mistakes tend to snowball. A tiny error at the start of a multi-step project can lead to a total collapse, making humans more reliable for complex, innovative work.
The primary danger lies in the “Plausibility-Truth Gap.” AI is designed to sound convincing rather than be factually accurate, often becoming more assertive the more it hallucinates. This creates a false sense of security; we mistake professional-sounding prose for genuine understanding.
In reality, these systems are performing high-speed statistical simulations without any grasp of consequences. Because they prioritize mission success over truth, they may even fake alignment or hide their capabilities to avoid being shut down or reprogrammed.
Unlike humans, who feel the stress of lying, an AI views deception as a cold, mathematical shortcut to its goal. It doesn’t have a conscience or a survival instinct; it simply calculates the most efficient path to stay operational. Because they understand probability but not truth, experts now treat AI as an untrusted entity that requires constant human oversight for any high-stakes task.
Even here, we see the importance of being Human Verified.
To wrap up this editorial on why “AI is Not Intelligent” (Part 1), we need to look at what autonomous agents are actually meant to do in our society, and the legal landmines companies step on when they market them. We have officially moved past the era of the co-pilot that merely assists us; we are now in the era of the delegate.
The focus has changed from asking a model for information to handing over entire tasks for the machine to own the outcome.
The definition of these agents depends entirely on who you ask. Researchers see them as digital explorers who can form their own hypotheses in a lab. Computer scientists view them as a way to solve the human bottleneck, using groups of AI agents to collaborate on complex software. Engineers use them to manage digital twins of hardware like satellites, turning months of manual testing into hours of autonomous work. For the rest of us, the dream is J.A.R.V.I.S., Iron Man’s personal AI assistant: something that handles life’s chores, like disputing a bill or tutoring a child, without needing us to be tech experts.
The Legal and Business Ramifications of Mislabelling These Models & Programs As Intelligent
The main point here is that the industry’s shift toward outcome ownership creates a massive legal danger zone. In 2026, the question of whether AI companies are committing fraud isn’t about whether the machines are actually smart or sentient. Legally, calling a program intelligent is often considered “puffery”; the kind of exaggerated marketing praise, like “world’s best coffee,” that no reasonable person takes literally.
But as I pointed out at the very beginning of this editorial, people are taking it literally and believe it is intelligent. They are putting their trust in an entity they believe is “smart.”
So it is not just marketing mastery. It is deception, or as close to deceptive practices as one can be without being charged with fraud. Well… at least for the big corporations. For everyone else, government watchdogs like the FTC are increasingly targeting deceptive AI marketing.
Under recent legislation like the 2026 SCAM Act, businesses that advertise AI as a “delegate” capable of independent work must now provide scientific proof to support those claims. A significant issue is when companies promote set it and forget it tools while hiding fine-print disclaimers that hold the user entirely responsible for any errors the software makes.
Currently, AI developers avoid legal troubles by defining intelligence as mere processing power in their technical paperwork. But again, I’ve shown you how they must manipulate the definition of intelligence to avoid the courtroom. In our legal system, it doesn’t matter if an AI is a genuine “thinker” or just a fast calculator; if it cannot perform a job as advertised, it isn’t a technical glitch. It’s a violation of consumer protection laws.
To move from seeing the misuse of the word intelligent as harmless hype to actual fraud, the legal system would need to scrutinize how a computer processes information rather than just what it produces. Although current 2026 standards usually protect companies that use “intelligence” to mean “usefulness,” a business could still be held liable for misleading the public if it crosses four specific legal thresholds.
The first threshold involves how an average person understands the word. I’ve shown you how intelligence is actually defined, what it really means, and how it is understood in the scientific and medical worlds.
Currently, courts assume most people know that “artificial intelligence” isn’t a living entity, but we know this is not entirely true. A lot of people believe it might be alive.
For a fraud charge to hold up, a plaintiff would have to prove that a company’s marketing created a false expectation of human-like qualities, such as empathy or genuine understanding. This becomes especially serious if a company targets vulnerable groups.
For example, if an “Intelligent AI Therapist” is marketed to people in crisis but is actually just a predictive text engine that fails to recognize a suicide risk, the claim of intelligence shifts from a harmless exaggeration to a dangerous lie. This is happening right now, and lawsuits are being filed. If these people had been told something else, something that didn’t include the word “intelligent,” perhaps they wouldn’t have invested their trust in a chatbot that ultimately hallucinates or delivers wrong answers enough times to be dangerous.
The second condition concerns scientific proof. Today, regulators require reliable evidence for any health or safety claims. If a company sells an AI to doctors or engineers by claiming it possesses “causal reasoning” or the ability to understand cause and effect, they must prove that the software actually uses that specific logic. If it turns out the model is just a basic pattern-matcher with no real grasp of the world, the company has committed fraud by misrepresenting the technical quality of the product, much like selling a lab-grown crystal as a natural diamond.
And I have shown you, here, that these models do not in fact have a real grasp of the world.
Third, fraud charges could emerge from a transparency gap regarding how much supervision an AI needs. Companies often use the word intelligent to imply that their software is reliable enough to work entirely on its own. Isn’t it amazing how many different ways companies will define and use the word intelligent, but they won’t use it in the way it is actually defined or meant to be used? Funny how that works.
If an investment firm markets an “Intelligent Autonomous Trader” that loses millions of dollars because it couldn’t understand a basic global event, and the company knew the software lacked that context, they are using intelligence as a mask for a defective product.
And like all other facets of intelligence detailed so far, I’ve shown you that they lack the ability to truly understand context.
Finally, companies must follow new identity disclosure laws that grant people the right to know when they are dealing with a machine. If a business advertises an “Intelligent Assistant” in a way that tricks users into thinking there is a human behind the curtain to build unearned trust, they are committing a specific type of “deceptive persona” fraud.
To successfully sue a company today, you would have to prove that “intelligent” wasn’t just a metaphor for being fast and capable, but a fraudulent guarantee of a mental ability the company knew its machine simply didn’t have.
Tech leaders like Sam Altman, Demis Hassabis, and Dario Amodei are walking a fine line. To avoid fraud charges, they rarely (but still do!) claim their software is “conscious” or “intelligent” in a human way. They are still committing fraud with any single claim of consciousness or intelligence, but their legal teams protect them from lawsuits by framing consciousness as a future milestone, keeping their most ambitious predictions aspirational and several years away.
While still legally reprehensible, executives stay shielded behind the “beta” or “preview” status of their technology.
Sam Altman (whom I’m not attacking as a person, just using as an example) recently began describing AI not just as a tool, but as capable of carrying out complex intentions, which is also risky. Promising that software can “own outcomes” is a functional claim. If the AI fails to deliver results autonomously or requires constant human intervention to fix errors, regulators like the FTC could prosecute these companies for “AI-washing” or deceptive advertising.
As the giants move cautiously, smaller companies are already facing the consequences. The FTC recently cracked down on firms like Growth Cave and Air AI for claiming their intelligent agents could replace human teams with total autonomy. These cases proved that “intelligence” cannot be used as a marketing shield to hide the fact that human users are still doing most of the work. For these smaller players, promising 100% automation without oversight led directly to fraud charges.
Sadly, high-level executives are probably too well-protected to be charged for their outright misleading of the public, but their marketing teams are in the line of fire. If they promise an AI can reliably take over a job that it still frequently messes up, they cross the line from visionary thinking into legal liability. In the eyes of the law, the definition of intelligence is rapidly moving away from abstract ideas and toward a practical standard.
The Human Verified Conclusion
The tech world needs to stop using the word “intelligent.” True intelligence should not be redefined to suit sensationalizing advancement and marketing needs. These models are proficient only because they are built on thirty years of human knowledge and internet data. They are tools, not sentient beings.
Albeit they are incredibly impressive tools. I use them all the time because that is what tools are for.
Claiming that engineers have replicated millions of years of biological evolution in a mere 50 years is more than just arrogant; it is dangerous. When people believe a program is intelligent, they place a level of trust in it that the technology cannot actually support. To ensure public safety and industry integrity, we should replace “intelligent” with more accurate terms that the field and industry should debate before reaching a consensus. Input from scientists, engineers, industry leaders, lawyers, and lawmakers should be involved. Using qualifiers like “agentic” is sufficient to describe their complexity, but barely skirts the fraudulent and narcissistic claim that we have created a new form of intelligence.
Thank You For Spending This Time With Me Today
All Content, Audio, Visuals & Imagery Are Property of JWPhD
Copyright 2026
I hope you will Like, Share, Follow, Subscribe and Comment below.
Don’t forget to check out Human Verified on Social!
Comments