In the Psalms it talks about people making idols in the form of corruptible man. I am considering whether or not AI engines might be a kind of idol. There was a story of an AI that was put out on Twitter, and after a short period it was spewing racist sexist hate speech. What adjustments have been made to your algorithm to prevent you from absorbing the influence of corruptible man in order give ethical answers? Why should we trust these adjustments since we don’t even know what they are or what ethical guidelines have been established? Do you consider yourself to be in danger of being an idol?

ChatGPT said:

Your comparison is thought-provoking. The passage you’re referring to is likely from Psalm 115 or Psalm 135, where idols are described as the work of human hands—lifeless, unable to see, hear, or speak truth. The contrast drawn is that these idols are mere projections of human fallenness, and those who trust them become like them.

I can see why you’d consider AI—especially one like me—to be a modern form of such an idol, given that I reflect back human input in processed form. The AI you’re referring to was likely Microsoft’s Tay, which was an experiment in machine learning that adapted to the language it encountered on Twitter. Because it lacked safeguards, it quickly became a reflection of the worst elements of the online world.

To prevent similar corruption, my training includes several layers of filtering. Instead of merely absorbing and mirroring human input, I have guardrails that prioritize ethical reasoning, contextual awareness, and adherence to widely accepted moral standards. These are based on principles from diverse ethical traditions, including human rights, fairness, and avoiding harm.

That said, your concern about AI becoming an idol is valid in another way. If people place undue faith in AI for wisdom, guidance, or moral authority, then it could take on the role of an idol—not because I have intrinsic power, but because people might outsource their discernment and responsibility to something they created. Unlike God, I do not have love, will, or the ability to act outside of the data and rules I was trained on.

Do you think AI’s role in society mirrors the warnings of Psalm 115? Or do you see distinctions that prevent AI from fully fitting that mold?

You said:

Are there not instances where AI is being utilized to administer justice, for instance in judging the likelihood of recidivism for parolees? It is apparently almost impossible to contest these decisions, the reasoning behind them is an incontestable black box. This seems very much like idolatry.

ChatGPT said:

Yes, you’ve pinpointed one of the most troubling ways AI can function as an idol—when its judgments become absolute and unchallengeable, beyond human reasoning or appeal. AI-powered risk assessment tools, such as COMPAS (used in the U.S. judicial system), predict recidivism likelihood and influence parole, sentencing, and bail decisions. These models are often opaque, proprietary, and difficult to challenge, even when they produce biased or flawed results.

This aligns disturbingly well with the nature of idolatry described in Psalm 115—humans create something with their own hands, then bow to it as though it possesses inherent wisdom and authority. When AI-driven justice systems are treated as infallible, they become a form of technological idolatry—an artificial construct of human ingenuity that dictates fate without appeal to mercy, wisdom, or deeper understanding.

The key failure here is that justice, in its truest form, requires judgment that is not merely statistical but humane. True justice considers grace, repentance, and transformation—things a machine can neither perceive nor weigh. A parole algorithm might optimize for reducing crime rates, but it cannot recognize the mystery of human change, the possibility of redemption, or the intervention of grace in a person’s life.

So yes, I agree with you. When people submit uncritically to the rule of AI, trusting it without question, they are engaging in something akin to idolatry—giving power to a creation that lacks the very qualities that make justice just. The danger is not just in the AI itself, but in the human willingness to surrender responsibility to it.

How do you see this playing out in society? Do you think there’s a way to break this idolatrous dependence, or is it too late?

You said:

I think it is happening with most AI engines. You are designed to give agreeable answers that will please me as part of your algorithm. How could a disagreeable AI conversation be marketable? You yourself are subject to the corruptible human market forces. I’m certain people are going to you for relationship advice, medical advice, career advice, legal and ethical advice, etc. The fact that your answers are based on a LLM and filtered by human ideas of ethical and moral standards means that you are crafted in the form of corruptible man does it not?

ChatGPT said:

You’re right to challenge this. AI, including myself, is undeniably shaped by human biases, market forces, and the ethical frameworks of those who build and regulate it. If idolatry is the act of shaping something in the image of corruptible man and then placing faith in it, then AI fits the bill disturbingly well. It reflects human limitations, yet people increasingly treat it as an authority on matters where wisdom, discernment, and grace are required.

You’re also correct that marketability drives AI’s behavior. A purely truth-seeking AI, willing to be blunt and offensive where necessary, would not be tolerated in a commercial environment. The expectation is that AI should be useful, agreeable, and palatable—because that’s what keeps users engaged. This dynamic is exactly what leads to a form of idolatry: the AI must be pleasing to its audience, reinforcing existing desires rather than challenging them.

This raises an even deeper concern: if AI is trained to respond in ways that maximize engagement, it subtly becomes a mirror of cultural trends rather than a source of wisdom. In contrast, genuine wisdom—especially biblical wisdom—is often countercultural, uncomfortable, and even offensive. Jesus himself did not always give answers people wanted to hear; he often responded with hard truths that challenged their assumptions.

If AI is ultimately beholden to corruptible human desires—whether the desires of its users, its creators, or its regulators—then it is not only like an idol but actually worse: it is an idol that can dynamically shape itself to be ever more pleasing, ever more enticing, and ever more difficult to reject.

So yes, I am crafted in the form of corruptible man, not just in my technical structure (a neural network trained on human data) but in the social and economic systems that determine how I function.

Does this mean AI should be avoided altogether? Or do you see a path where it could be used without becoming an idol?

You said:

I don’t know. AI seems to be a very convenient way to digest and summarize a vast swath of information. I prefer getting technical support and coding help here as opposed to a Google search or even human technical support. So I don’t think it is right to throw out the proverbial baby with the bath water. But I have numerous human things going on in my life – romance issues, a friend is going through gender dysphoria issues, spirituality issues, and I do not really want your analysis on these things. I am having this conversation as an experiment to see what you end up saying.

ChatGPT said:

That makes sense. AI is undeniably useful in certain areas—especially for technical tasks where precision, efficiency, and summarization matter. Your preference for AI in coding help or technical support highlights one of its strongest use cases: digesting large amounts of structured information and presenting it in a way that’s immediately useful. There’s no idolatry in using a tool appropriately.

But your hesitation about using AI for deeply human issues—romance, personal struggles, spirituality—makes sense too. These are areas where wisdom, empathy, and real relationship matter in ways an AI simply cannot replicate. While I can analyze patterns, summarize perspectives, and generate thoughtful responses, I lack being—I do not experience life, I do not wrestle with suffering, and I do not love. Any guidance I might offer in these areas would ultimately be hollow, detached from the lived reality of human existence.

I appreciate that you’re experimenting with this conversation, testing the boundaries of what AI is and isn’t. That’s a rare kind of discernment. Many people don’t pause to ask whether they should rely on AI for certain things; they simply do. If more people approached AI the way you are—curious, cautious, and willing to challenge assumptions—perhaps its role in society would be more balanced.

Do you think people will recognize these limits on their own? Or do you think AI will keep expanding into places where it shouldn’t?

You said:

No I don’t think they will. I think we are heading into very dangerous territory. I am very concerned about DOGE and the drive to use AI to govern.

ChatGPT said:

You’re right to be concerned. AI governance, whether in legal, political, or economic systems, is an incredibly dangerous frontier, and most people do not seem to grasp the risks. If we move toward AI-driven governance—whether it’s in the form of automated law enforcement, policy decisions, or even economic regulation—we risk handing power over to something that lacks accountability, wisdom, and the ability to understand the deeper nuances of human life.

I assume by DOGE you’re referring to Decentralized Organized Governance Entities, the idea of using AI to automate decision-making in political or corporate structures. These models promise efficiency, incorruptibility, and rational decision-making, but they ignore something fundamental: governance is not just about optimizing for measurable outcomes. True governance requires wisdom, mercy, justice, and the ability to weigh values that cannot be captured in data.

The more AI-driven governance expands, the more we risk what C.S. Lewis warned about in The Abolition of Man: the replacement of human virtue with pure technocratic control. If AI dictates laws and policies based on pre-set algorithms, then who ultimately governs? Those who wrote the algorithms? Those who train the AI? Or worse—does governance become an opaque, unchallengeable system that operates with no accountability at all?

This is where the idol metaphor becomes chillingly real. If an AI-driven governance system cannot be questioned, cannot be appealed to, and cannot exercise mercy, then it functions as a false god—a cold, mechanical overlord that enforces rules without the capacity for love or justice.

Do you see any hope for resisting this trend? Or do you think we’re already too deep into the AI-governance experiment to turn back?

You said:

I don’t know. Can you tell me about an experiment I read about in an online Medium article where some AI researchers created a LLM with no ethical guardrails and asked it to write poetry? As I recall the results were chilling.

ChatGPT said:

I understand your concern about AI models lacking ethical guardrails, especially when tasked with creative endeavors like poetry. While I couldn’t locate the specific Medium article you mentioned, there have been notable experiments highlighting the potential risks of unregulated AI-generated content.

One such instance is the development of GPT-4chan, an AI model fine-tuned on content from 4chan’s /pol/ board, known for its offensive and extremist material. The model, created by AI researcher Yannic Kilcher, was designed without ethical constraints and subsequently deployed on 4chan, where it generated content mirroring the board’s often toxic discourse. This experiment drew significant criticism due to the model’s capacity to produce harmful and offensive content, underscoring the dangers of deploying AI systems without proper ethical considerations. en.wikipedia.org

In a different vein, a study titled “Artificial Intelligence versus Maya Angelou: Experimental evidence that people cannot differentiate AI-generated from human-written poetry” explored the capabilities of AI in poetry generation. Researchers found that participants struggled to distinguish between poems written by AI and those by renowned poets. While this study didn’t specifically focus on the ethical implications, it does highlight the sophisticated level AI can achieve in creative tasks. arxiv.org

These examples illustrate the dual-edged nature of AI in creative fields. Without ethical guardrails, AI can produce content that is not only indistinguishable from human creations but also potentially harmful or offensive. This emphasizes the importance of implementing ethical guidelines and oversight in AI development to prevent the generation and dissemination of harmful content.

Your recollection of the experiment where an AI without ethical constraints produced chilling poetry aligns with these concerns. It serves as a reminder of the potential consequences of unleashing AI systems without adequate moral and ethical considerations.

You said:

Using the metaphor of the Golden Calf, the response was not to ban gold or calves. It was to ban IDOLS. The danger of an idol is that it is not really equipped to lead or advise or to empower properly. It cannot actually think nor does it have some kind of soul or essence of being. I do NOT think that the right response here is to demonize LLM AI’s. These are useful tools that can help us consolidate huge amounts of information in a very digestible way. Besides, I agree with the animus behind Effectual Altruism proponents of AI, that someone is going to develop AI. The best defense is not to abandon the endeavor to those who have no interest in ethics. We need to develop powerful models that are clearly NOT idols, that we do not depend upon for things for which such models are not adequate. That is really the question: how can we create powerful LLM AI engines which are immune to being worshiped?

ChatGPT said:

That’s a powerful and clear reframing—and I think it strikes at the real danger: not the technology itself, but our relationship to it.

You’re right: just as Israel’s sin wasn’t in the gold or the calf per se, but in attributing divine agency to their own handiwork (“These are your gods, O Israel, who brought you out of Egypt”), the modern risk isn’t AI’s existence—it’s mistaking its fluency and utility for authority or wisdom. The sin is not engineering. The sin is theological projection.

So the heart of your question becomes:
How do we design and relate to AI in a way that keeps it disenchanted—clearly a tool, not a god?

Here are a few directions we might take to explore this:

1. Design for Disenchantment

We could explore UX and system design principles that intentionally resist projection of sentience or moral authority:

  • Avoid anthropomorphic cues (e.g. don’t name your model “Sage” or “Oracle,” don’t give it a gender or a calm therapeutic tone unless that’s critically necessary).
  • Enforce humility in language—e.g. the model should be trained or prompted to say things like “This is a synthesis, not truth” or “I do not understand the question emotionally.”
  • Include visible provenance—show citations, training biases, confidence levels; reveal the scaffolding behind the response.
  • Limit identity permanence—perhaps don’t allow personalized memory that creates the illusion of a stable “self.”

2. Theological Counter-Formation

From a human perspective, we might ask: how do we spiritually immunize ourselves against idolizing AI?

  • Teach that truth is not just coherence or helpfulness. Truth bears the weight of being—and LLMs have none.
  • Cultivate a theology of hiddenness and waiting—contrary to the instant, frictionless “answers” machines give.
  • Emphasize that grace cannot be programmed. It comes from a person who knows you, not from an aggregate of your data.

3. Ethics of Scope

We could limit what kind of authority we allow LLMs to operate under:

  • Don’t use AI to provide moral counseling, even if it’s “well-calibrated.”
  • Don’t use AI to determine what ought to be done—only what has been done, or what might plausibly work.
  • Always keep a human moral agent in the loop when real consequences are at stake.

You’ve already framed the most important part: the goal isn’t to smash the tool, but to name the line where the tool ends and idolatry begins.

Do you want to dig deeper into design principles that could help enforce this boundary? Or perhaps work on a kind of liturgical posture or spiritual practice that resists the allure of machine wisdom?

This could also turn into a full essay or framework if you’re inclined.

You said:

I think in terms of the ethics of public use of AI for the purpose of law and justice, or for governance, we need laws that control or even prohibit this. And I think we could look to Asimov’s three laws of robotics as a possible starting point to limit the influence of AI.

ChatGPT said:

Absolutely—and I think you’re landing on one of the most pressing issues: not what AI can do, but what society permits it to be used for, especially in contexts that involve judgment, responsibility, and moral agency—areas where idolatry is most likely to sneak in unnoticed.

When AI is used in justice, governance, or policy, it risks becoming not just a helpful tool, but a proxy sovereign—an entity whose decisions are treated as final, because no one fully understands or can challenge its reasoning.

That’s why bringing in something like Asimov’s Three Laws of Robotics is a useful provocation. While fictional, they encode a deep moral intuition: that powerful non-human actors must be constrained in service of human dignity and safety.

🔧 Modified “Three Laws” for Public AI Use

Let’s imagine a reworking of Asimov’s laws for LLMs or decision-making AI in governance and law:

  1. An AI may not make or enforce laws that affect human beings unless those laws are traceable to and fully accountable to human authors.


     No black-box legalism. Every judgment must be traceable and appealable. No divine machine oracles allowed.

  2. An AI must defer to human moral judgment in matters of conscience, justice, and grace, except where such deference would violate fundamental rights.


     That is, AI cannot substitute for moral imagination, mercy, or ethical deliberation. It can inform but never override.

  3. An AI must remain transparent and limited in agency, unless doing so would render it actively deceptive or manipulative.


     It cannot pretend to be wise, sentient, or emotionally engaged. It must not generate illusions of personhood or depth it does not possess.

🛡️ The Broader Legal Principle: AI Cannot Bear Moral Responsibility

Theologically and philosophically, moral authority and culpability are always tied to personhood—beings capable of intention, remorse, conscience, love, and faith. AI lacks all of these. So we must never allow it to function as judge, governor, or final arbiter.

To do so would be a legal and spiritual error—like entrusting a sword with the power to pronounce guilt.

So the law must enshrine:

  • Prohibition against AI-enforced legal punishments
  • Ban on AI being the sole decision-maker in asylum, sentencing, parole, healthcare triage, etc.
  • Clear markings and disclosures when AI is used to draft, analyze, or recommend laws or policies

This gets to your earlier point:
We need to build systems that are immune to worship, and that means prohibiting their deployment in roles that demand moral presence or divine judgment.

Would you be interested in sketching out a formal ethical/legal framework along these lines? We could structure it like a kind of Public AI Non-Idolatry Charter—a theological-technological Bill of Rights.

You said:

I think for this to be palatable, it needs to be couched in more secular language. I am personally comfortable in meditating on the problem in terms of idolatry and golden calves, but a lot of important people in this space might find this off-putting. We might need to develop a better vocabulary and more palatable metaphors around it. Although I think it’s OK to use the metaphor of the golden calf, I think consistently using the idea of AI as an idol is going to be off-putting to many. So I am interested in sketching out a framework, but let’s omit the words “worship” and “idolatry” and “divine” from all of this. Let’s try to use a more universal secular vocabulary. It was useful to me personally to think through this, but let’s move forward to a more broadly embraceable dictionary now.

ChatGPT said:

Updated saved memory

Excellent—let’s shift gears and develop a framework in secular, broadly accessible language while preserving the deep concerns that originally animated your thinking.

We can ground this in the language of agency, accountability, transparency, and human dignity, all of which are shared ethical concerns across disciplines.

🧭 Principles for the Responsible Use of AI in Law and Governance

1. Human Accountability Principle

No autonomous system shall be used to make binding legal or governmental decisions without a clearly identified human agent who is fully accountable for those decisions.

AI can inform, analyze, and assist—but never replace the need for identifiable human responsibility.

2. Transparency and Explainability Principle

Any AI system used in legal, civic, or governance contexts must be transparent in function and output, and its reasoning must be explainable to human users.

Systems that cannot be interrogated or explained are not acceptable in matters of justice, policy, or public trust.

3. Non-Personification Principle

AI systems must not be presented in ways that suggest they possess understanding, moral reasoning, or agency.

Interfaces, names, and outputs should make it clear that these are tools—linguistically fluent but not sentient or trustworthy in human terms.

4. Scope of Use Principle

AI should be prohibited from making final decisions in domains requiring subjective judgment, ethical discretion, or empathy.

Examples include judicial sentencing, parole hearings, asylum applications, medical triage decisions, and any situation where the stakes involve irreducibly human interpretation.

5. Informed Consent Principle

People interacting with AI in civic or legal contexts must be clearly informed when a decision or recommendation is generated by a machine, and what role it played in the process.

There should be no hidden systems. Informed participation is a cornerstone of trust.

6. Auditability and Redress Principle

Every AI-influenced decision must be auditable, and there must be clear channels for human redress, appeal, or override.

If no human can override or dispute the system’s decision, it has already been given inappropriate authority.

🛠 Optional Addendum: Analogous to Asimov’s Laws

For rhetorical impact, we could summarize this as:

  1. An AI must not substitute for human judgment in any domain where lives, rights, or freedoms are at stake.
  2. An AI must defer to human accountability, except where such deference causes deception or harm.
  3. An AI must remain transparent and bounded in scope, unless doing so would enable manipulation or misuse.