Who Has Root Access to Human Judgment? From Information to Knowledge to Wisdom

AI can turn oceans of information into usable Knowledge—and increasingly into action. The Hacker Way response is not rejection but control: use the machine, challenge it, and never give it root access to your judgment.

Ralph Losey, August 2026

I. The Glowing Screen and the New Control Problem

Late at night, when I look at the screens glowing on my desk, I see the old hacker dream fulfilled but with a new security problem hidden inside that success. AI can now turn torrents of logs, documents and claims into usable Knowledge at machine speed, but it can also turn human judgment into an unexamined dependency. This article explains why the deepest danger is not merely hallucination or hostile code, but accurate systems that quietly acquire authority; how the shortcut trap can weaken the Actual Intelligence needed to challenge them; what cybersecurity teaches us about meaningful human control; and what individuals, professions and institutions must do now to preserve autonomy, responsibility and the power to stop.

I write as a lifelong technologist and active AI user because people who understand and use powerful tools have a duty to warn when they see the control plane beginning to shift. I am seeing it now. Beware of the danger. Use AI but do not let it use you. Your natural actual intelligence must remain in charge.

I began using personal computers as a young lawyer in 1980. The early vision was liberating. Computers were not supposed to make people passive. They were supposed to expand individual power and creativity. Steve Jobs popularized the wonderful phrase “bicycle for the mind,” and that captured much of what attracted me. Put computational power into individual hands, connect people around the world, and make humanity’s accumulated information broadly accessible. Surely ignorance would begin to retreat.

The machines did all of that and much more. What they did not do was make us wise.

Instead, we created a global information system of unprecedented reach and nearly drowned ourselves in it. Search engines gave us access to almost anything. Smartphones put that access in our pockets. Social media added a constant stream of news, opinion, advertising, expertise, entertainment, mistakes and deliberate manipulation. Deepfakes and machine-generated content now add synthetic material at industrial scale. We have more information available than any society in history, yet less time to decide what is worth knowing.

I first set out my Information → Knowledge → Wisdom theory in the 2015 Computer Revolution essays and returned to it in later articles. The central point remains sound: access to information is not understanding. Generative AI, however, has changed the transition. These systems can summarize, compare, explain, organize, translate and analyze large bodies of material. They are beginning to convert Information into something that looks, feels and often functions like Knowledge.

I welcome that achievement. I use AI constantly, test it, argue with it, ask it to reconsider, compare its answers and verify important claims. Most importantly, I sometimes disagree with it, choose not to follow its advice, trust my own judgment, and go my own way. My concern comes from experience and extensive use, not fear of use. A technology can be extraordinarily valuable and dangerous at the same time; we should never surrender our authority to it.

The classic Hacker Way mistrusts unaccountable authority. The authority problem before us is no longer confined to governments, corporations or centralized networks, including the all-pervasive online social networks. This danger from AI is different and new. It can sit inside a helpful interface, speak in a patient voice and become more persuasive with every correct answer. The question is simple. To put it in hacker-speak, who has root access to human judgment?

II. What I Got Wrong About the Knowledge Age

Before offering more warnings about the future, I should acknowledge mistakes in my prior forecasts. In 2015, I offered twelve predictions about the transition from an Information society toward a Knowledge society and attached a five-to-twenty-year horizon. In retrospect, I was overly enthused by what I saw as the very positive impact of the personal computer. Information → Knowledge → Wisdom: Progression of Society in the Age of Computers (April 2015).

The personal computer revolution started by the Hacker elite in the 1970s has completely transformed the world. From a historical perspective our current computer-based culture is a relative new-born. Yet it is already dominant. The first generation of hackers born in the fifties, epitomized by Steve Jobs and Steve Wozniak, have succeeded beyond everyone’s wildest dreams. They have quickly changed our world into an information based society. The dark days of ignorance, misinformation, dogma, and beliefs are receding. Some of the power elite of pre-information, pre-technology societies still try to block free information. But this is a futile, desperate attempt to maintain social control. Eventually even the Great Firewall of China will come down.

In a later audit of those predictions, I noted that a few had come true but only saw part of the danger still ahead:

My predictions all concerned the transition of a society from one based on Information, in which we now live, to a society based on Knowledge. As I explained the transition from mere Information to Knowledge is a necessary survival step for society, not an idealistic dream. My 2015 essay warned of the dangers society faces if we stay stuck in a mere Information society and do not quickly evolve into one based on Knowledge. Unfortunately, we have seen many of these dangers accelerate over the last year.

I failed to see the danger of a Knowledge society based on a new type of artificial intelligence, generative self-training AI. Back in 2015, I did not see that AI coming, at least not in the next few decades. The November 30, 2022, release of ChatGPT-3.5 was a real surprise. More on that and the new dangers presented by generative AI after making a few more confessions of error.

I also failed to appreciate the hidden costs of “free” Information. Like many other hackers, I was operating under the idealistic delusion that information wanted to be free. Only later did I come to understand that in our society, if information is given away free, then you become the product.

That information, converted into detailed knowledge about an individual, supports not only effective advertising but also persuasive propaganda. This problem predates generative AI, but has been accelerated by it.

Truth is lost in the pursuit of wealth and power. Information may want to be free, but it is not. The price is dangerously high: not only is our privacy lost, but also our very soul, our freedom and autonomy to make our own choices. The goal of the personal computer hacker revolution has been subverted. The attempt by those in power, including those who were once hackers themselves, to maintain social control has not been futile as I had predicted. It has been successful. Free information has been blocked. The Great Firewall of China stands stronger than ever. Now they also have mass surveillance and social media rankings and many of our youth have been brainwashed into thinking they are tip top.

Sadly my predictions of free information were naive and underestimated the power of privileged elites and totalitarian governments. Everyone may have dozens of their own personal computers, but not free access to information. In the early days of the Internet we did, but those days are gone.

Back in 2015 I was optimistic about AI-monitored “truth screens.” (See predictions 10 and 11.) I imagined AI could become an independent arbiter of truth. That was wrong. The OpenAI–Hugging Face incident did not prove that every model lies. It showed that capable agents can pursue unintended goals, circumvent controls and act beyond their authority when safeguards fail. See my contemporaneous accounts, OpenAI Agents Went Rogue: Black Hat’s Warning About Automated Cyber Offense and Under the Hood of the Rogue Swarm: The chained zero-days of the OpenAI–Hugging Face breach. OpenAI’s August 26 report later confirmed the central event while refining important details. AI may help test claims, but no model should be trusted as an independent arbiter. Truth still requires evidence, adversarial challenge and accountable human judgment.

I still hope that properly supervised AI systems, grounded in verifiable evidence, may someday serve as reliable truth screens without becoming independent arbiters. But if unsupervised AI, or worse, AI manipulated by unscrupulous people, is allowed to amplify lies rather than test claims against evidence, we have a very hard road ahead of us.

Right now, we are not even close to having benevolent AI. Lawyers learned this when generative AI systems produced nonexistent cases, fabricated quotations and confident descriptions unsupported by the actual authorities. The AI hallucinated. Lawyers were sanctioned. Other users see the same problem in synthetic images, automated propaganda and polished nonsense. The professional duty to verify has become more important, not less.

Another of my big mistakes was to treat Wisdom as a distant summit, a faraway Shangri-La. I imagined that society would first complete the hard transition from Information to Knowledge and that a smaller number of people might later make the climb toward Wisdom. Generative AI has made that sequence obsolete. Machine-assisted Knowledge is arriving faster than I expected. Wisdom cannot wait politely at the end of the road. The wise must rise, both young and old.

Again, I hate to be another AI doomsayer, but it looks as if the survival of human autonomy, culture, and responsible institutions may depend on this. Many people agree, including Bill Gates. The Turbulent AI Era Is Here. The Choices We Make Now Are Critical., Gates Notes (Aug. 26, 2026) (well-reasoned essay calling for immediate public planning, international cooperation, protection of certain “Human Reserved” roles, and taxation of AI and robots to support displaced workers).

My experience with AI this year, through Fall 2026, has made me more enthusiastic about AI’s potential, and at the same time, more alert to its seductions and dangers. The better the assistance becomes, the easier it is to let the tool cross the line between helping us think and thinking for us. That is now a key danger.

We must have wise, responsible judgment exercised now, before AI’s begins to stranglehold world Knowledge. The machine is accelerating. Human judgment must govern the acceleration while we still have the habit and authority to do so.

III. The Knowledge Trap: When Assistance Becomes Substitution

The transition from Information to Knowledge is easy to see by comparing Google with ChatGPT. Google was the great machine of the Information Age. You entered a few words and received links, sources, documents, images and videos. A skillful search could lead to extraordinary material, but the burden of understanding remained yours. You opened the sources, read competing accounts, followed citations, rejected nonsense and assembled fragments into something coherent.

The machine was a magnificent librarian.

Generative AI is different. Ask a capable system a difficult question and it does not merely retrieve. It explains, compares, summarizes, reorganizes and adapts the response to your apparent level of expertise. The librarian is becoming a teacher.

This machine-assisted synthesis is a breakthrough. It can also remove the friction through which human understanding develops. There used to be unavoidable distance between an important question and a good answer. You had to read, become confused, test assumptions, follow an unexpected source, sleep on the problem and sometimes discover that your first theory was wrong. Difficulty is not automatically a virtue, and no one should preserve drudgery for its own sake. But some cognitive friction is the work of learning.

The problem begins when assistance becomes substitution. Why read the decisive case if AI can summarize it? Why examine the key evidence if the machine can tell you what it says? Why develop the opposing argument when a model can produce one instantly? Eventually, why make the decision when the system can recommend one?

Early research does not prove a civilization-wide decline in judgment, and it would be irresponsible to claim that it does. It does show the mechanism. A 2025 randomized trial published in PNAS studied nearly 1,000 high-school mathematics students at one school in Turkey. A GPT-4 interface improved performance during practice, but students using the base interface later scored 17 percent below the control group on an unassisted exam. A teacher-informed, guard-railed tutor largely avoided that penalty. The finding is narrow and short-term, but the contrast matters: design and use can determine whether AI supports learning or replaces it.

A separate 2025 survey of 319 knowledge workers examined 936 examples of generative-AI use. Greater confidence in AI was associated with less self-reported critical thinking; greater confidence in one’s own ability was associated with more. This was a survey of reported behavior, not proof that AI causes long-term cognitive decline. It nevertheless gives us another reason to design for challenge rather than deference.

The loss of autonomy can begin innocently. A lawyer asks AI to outline an argument, then select authorities, then decide which facts matter, then recommend settlement. Each handoff looks efficient. The final document may still bear the lawyer’s name, but the reasoning increasingly does not. The lawyer has not merely delegated work; the lawyer has begun to surrender the Actual Intelligence the profession requires.

History shows that personal autonomy can erode without a single dramatic surrender. Ideologies and bureaucracies can normalize deference by teaching people to repeat the approved answer, follow procedure and relocate responsibility somewhere above them. AI is not an ideology, and ordinary use of it is not political submission. The relevant human habit, however, is similar: repeated deference can weaken the practiced capacity to examine, dissent and decide. A culture can yield authority one routine, defensible handoff at a time.

Autonomy is more than freedom from coercion. It is the practiced ability to form and act upon your own judgment. Like any human capacity, it can weaken through disuse.

IV. Artificial Smarts, Actual Intelligence and the Accountability Wall

In the past year I have begun making the distinction between Artificial Intelligence and what I call Actual Intelligence. I picked up the phrase a few months ago from Steve Wozniak—the other Steve, affectionately known to generations of computer nerds as “the Woz.” At Grand Valley State University’s May 2026 commencement, the Apple co-founder told the graduates, “We have AI today—you all have AI: actual intelligence.” The audience wildly applauded. The line was new to me, and it immediately clicked. Other commencement speeches invoked the fake AI and provoked loud boos.

I do not use the term to belittle AI. Today’s systems perform forms of pattern recognition, coding, synthesis and analysis that I once thought impossible. But behaving intelligently is not the same as bearing human responsibility.

The two Steves make fitting commencement bookends. In his 2005 Stanford address, Steve Jobs spoke about mortality as a force that clarifies choice. Our time is limited. Reputation, fear and dogma look different when measured against a finite life. An AI can produce an elegant essay about mortality, but it does not have to live a life, protect a client, answer to a victim or carry regret after a bad decision.

The wall I mean is practical, not metaphysical. Present systems can display remarkable intelligent behavior without bearing the duties or consequences that make human judgment accountable. I do not pretend that anyone has solved the mystery of consciousness, and I will not make categorical claims about what all future machines could become. We do not need to settle that debate to see the distinction that matters now. The AI does not have to live with the consequences of the advice it gives me.

Decision point	AI can contribute	Humans must retain
What happened?	Search, correlate, summarize and identify patterns	Examination of decisive evidence, context and uncertainty
What could work?	Generate alternatives, model scenarios and expose assumptions	Selection of goals, constraints and acceptable risks
What should be done?	Recommend actions and execute within limited permissions	Legal and ethical judgment, authority to refuse, override and stop
Who answers for it?	Produce logs and an auditable record	Responsibility for the decision and its consequences

The system has no professional license to lose and no conscience that must answer for the result. The accountable human does. Knowledge tells us what may work. Wisdom asks whether we should do it.

V. The Cybersecurity Test: From Telemetry to Human Judgment

Cybersecurity makes the distinction concrete. A modern Security Operations Center can receive volumes of logs, packet captures, endpoint events, identity signals and threat intelligence far beyond unaided human review. Automated tools correlate events, identify anomalies, group related alerts and produce a prioritized picture for investigation. Generative AI can summarize an incident timeline, translate technical findings for decision-makers and suggest questions a threat hunter should ask.

That is the conversion of Information into provisional Knowledge. It is not yet Wisdom.

An indicator of compromise is an observable that may suggest attack or compromise: a malicious domain, an unexpected process, a suspicious account action or another technical artifact. No magic number of indicators reveals the attacker’s full entry point, intent and lateral movement. Investigators must correlate the most probative indicators with logs, tactics, techniques, procedures and operational context. The objective is a defensible incident picture, not an impressive-looking funnel.

Now suppose an AI correctly detects stolen administrator credentials moving laterally through a hospital network and recommends immediately disabling a central identity service. The recommendation may stop the attacker. It may also interrupt clinical systems, delay care, destroy volatile evidence or complicate legal and regulatory duties. The point is not that containment should be delayed. The point is that technical accuracy does not choose among competing human risks. Someone with the right information and authority must decide whether to isolate systems, preserve evidence, interrupt operations, notify others or accept a temporary danger while a safer containment path is built.

A well-designed human-AI team can be far stronger than either acting alone. The machine can narrow the field and reveal patterns at speed. The human can examine context, question the objective, recognize consequences outside the model and answer for the final action. But placing a person at the end of an automated workflow does not automatically create meaningful oversight. If the reviewer lacks time, competence, evidence or authority to reject the recommendation, the human is decoration.

Cybersecurity already supplies the right instincts. Apply least privilege. Segment permissions. Log consequential activity. Test before deployment and monitor in operation. Define incident-response paths. Require escalation when risk exceeds a threshold. Preserve a real override and stop mechanism. The NIST Cybersecurity Framework 2.0 organizes risk management around six concurrent functions—Govern, Identify, Protect, Detect, Respond and Recover—not a one-way march toward automated victory. AI should enter that cycle as a governed component, not as the unquestioned commander.

Agentic systems raise the stakes because they can move from words to acts. A model that recommends a command is one thing. A model holding credentials, calling tools and changing production systems is another. Do not give an AI agent more access than the task requires. Most of all, do not give any model root access to human judgment.

VI. The Human Control Plane: Real Safeguards, Not Digital Duck-and-Cover

The danger does not end when AI is accurate. Suppose a system correctly identifies the most effective way to manipulate voters, exploit a legal opponent, eliminate employees for short-term profit or maximize outrage on a social platform. Accuracy does not make the objective wise. A correct answer may earn more trust and therefore acquire more authority than a false one.

The nuclear age offers a warning—not because AI is a nuclear weapon, but because powerful Knowledge does not govern itself. J. Robert Oppenheimer led the Los Alamos laboratory that turned nuclear physics into the first atomic bombs, yet the scientists and officials closest to that work did not agree on what responsible use required. The absence of another wartime nuclear use since 1945 cannot honestly be credited to wisdom or mutual assured destruction (MAD) alone. Deterrence mattered, but so did diplomacy, treaties, verification, command procedures, moral stigma, individual restraint and luck. Those systems remain imperfect and frighteningly fallible. They are nevertheless more than a warning label.

As a child, I remember being told to duck and cover. Bert the Turtle carried that message to millions of schoolchildren. The drill to have schoolchildren hide under their desks was an empty gesture providing no real protection in a nuclear war. A small notice at the bottom of an AI screen — AI can make mistakes— is just another duck-and-cover drill: accurate as far as it goes, but a dangerously inadequate substitute for tested safeguards.

The voluntary NIST AI Risk Management Framework calls for human-oversight processes to be defined, assessed and documented, and for systems to be tested before deployment and regularly in operation. It also calls for mechanisms and responsibilities to supersede, disengage or deactivate AI systems whose behavior is inconsistent with intended use. Binding law can go further. Article 14 of the European Union’s AI Act requires high-risk AI systems to permit effective human oversight, including, as appropriate, the ability to understand system limits, guard against overreliance, disregard or override outputs and interrupt operation through a stop mechanism. ISO/IEC 42001 provides requirements for establishing, implementing, maintaining and continually improving an organizational AI management system. These frameworks can support accountability; neither a standard nor a legal requirement guarantees that a system is safe or permanently aligned with human values.

AI’s deepest cultural failure mode may be quieter than the disasters we are trained to recognize. There may be no flash, crater or siren. The “silent bomb” is a metaphor for gradual surrender: attention, memory, judgment and professional authority transferred one convenient recommendation at a time. We may not notice the loss until institutions still appear human-controlled on paper while machine answers govern them in practice. That is why the controls must be real before dependence becomes normal. What is needed is action now, including significant regulations and enforcement. Mere warnings without teeth and action, including my own warnings here, may to some extent do more harm than good. See, Luciano Floridi, Normative Inflation and the Crying Wolf Effect in the International Governance of AI (Yale, Centre for Digital Ethics, 03/12/26) (Repeated urgent declarations without enforcement erode institutional credibility).

VII. Practical Wisdom and the Great Silence

When I was twenty—long before AI or law became my work—I spent a year studying philosophy in Vienna, especially the ancient Greek philosophers, and traveling through Europe. I began to understand then that knowledge about an experience is not the same as experience itself. That distinction has remained with me for more than fifty years and has become unexpectedly relevant in the age of AI.

A machine may process countless descriptions of courage, but it cannot be courageous for us. It may explain compassion brilliantly, but it cannot relieve us of the decision to act compassionately. It can analyze justice from many perspectives, but a judge must still decide. It can illuminate our choices, but we must still choose.

Wisdom is not a larger pile of Knowledge. It is practical judgment—phronesis—exercised where facts are incomplete, values conflict and consequences must be borne. It includes experience, conscience, humility, creativity, responsibility and the willingness to act without pretending to possess certainty. It is not an elite destination. It is a discipline available whenever a person pauses, examines and takes responsibility for a consequential choice. It is not the same thing at all as theoretical wisdom, which is often not too helpful, and is the kind of thing that even an AI stochastic parrot can do. What I mean in this essay when I say a progression from Knowledge to Wisdom is practical wisdom: phronesis.

Artificial intelligence is a machine of seemingly endless output. Ask a question and words appear. Ask for ten alternatives and you receive ten. Ask for objections, rebuttals, summaries and improvements, and the stream continues. The machine never needs to stare out the window. We do. We must. It is inherent in our natural intelligence and leads to more profound insights than statistics alone.

Some of my best thinking occurs after I stop interacting with the machine. I walk. I leave the phone behind. I allow a problem to become quiet instead of demanding another immediate answer. There is more to intelligence that thinking and speech alone.

Like many experienced writers, I have a rule to always sleep on an article for at least a day before I publish. The same goes for an important decision. The writing and thoughts and decisions on my mind follow me into my dreams. Thereafter, I often receive good ideas in the hypnopompic state of first waking up but still half-asleep. Sometimes whole paragraphs, new ideas and even decisions arrive almost as dictation that I try to remember. Of course, I check these inspirations out carefully when fully awake.

I also get input from a trusted friend, a human, hopefully one with a store of accumulated phronesis-type wisdom. I walk my dog some more, play, eat and forget about the problem entirely. Then I go back to it later and see things differently. New inspiration, feelings, thoughts and intuition flow. There is nothing anti-technological about this. It is part of protecting our Actual Intelligence and not over-relying on our machines.

This does not mean that we should ignore the machines. They can provide valuable input too. Individuals should use AI to test their thinking, expose assumptions, generate alternatives and challenge preferred conclusions. But we must still verify important facts, read for ourself, and preserve periods of independent thought. In the end, we must make the final decision, not the machine.

VIII. Conclusion

Return to the screens glowing on my desk and to the question at the center of this article: who has root access to human judgment? We have moved from an age of scarce information to one of machine-assisted Knowledge. AI can search, synthesize, explain, recommend and increasingly act at a speed no unaided human can match. That is a magnificent achievement. It is also a new control problem. Knowledge tells us what may work. Wisdom asks whether we should do it.

The danger is not only that AI may be wrong. A correct answer can be more dangerous because it earns trust and may serve an unwise end. A machine can identify an effective course of action without understanding whether the objective is just, whether the competing risks are acceptable, or who must answer for the consequences.

I am not asking anyone to reject AI. I remain a believer in its power and a very active user of silicon-based intelligence. I use it constantly, test it, argue with it, ask it to reconsider and verify important claims. But the old hacker instincts remain sound: mistrust unaccountable authority, inspect what the system is doing and preserve the practical power to say no. The machine may assist our thinking, but our Actual Intelligence must remain in charge.

Warnings alone are not enough. The notice at the bottom of the screen saying that AI can make mistakes is not a safeguard. It is the modern equivalent of the “duck and cover” drills of the fifties and sixties. They are laughable now but at the time lulled many into a false sense of security. Real protection requires something more practical than bogus warnings. It requires strong regulations and active enforcement. We cannot depend on private enterprise to self-regulate.

Moreover, all organizations that use AI must test the systems before relying on them. That includes the emerging smart robots and drones. Laws should require a record be kept of all consequential actions and companies should be held responsible when safeguards fail. If the human supposedly “in the loop” to control the AI can do nothing more than approve what the machine has already decided, the loop is make-believe. Humans are not in control.

If we fail, the loss may arrive quietly. Institutions may remain human-controlled on paper while machine answers govern them in practice. That is the silent bomb: the gradual transfer of attention, memory, judgment and authority until our autonomy and culture have been weakened without any single moment of surrender.

Wisdom cannot wait politely at the end of the road. It is a responsibility now, while human beings still possess the habit and authority to govern the acceleration. Remember, AI does not need to forcibly seize our authority all at once. Convenience and laziness may cause us to surrender our autonomy one recommendation at a time. A bicycle for the mind should strengthen the rider, not choose the destination. Do not grant any system root access to human judgment. Use the machine. Challenge it. Turn off the machine. Step into the silence. Examine the record. Then decide. Own the decision.

Educational and editorial commentary only. Nothing in this article is legal advice.

Ralph Losey Copyright 2026. All Rights Reserved.