St. Augustine and AI’s false promise

Reading time: ~7 minutes

Every time a new “revolutionary” artificial intelligence model comes to market, interface designers and product teams ask the same question again: what should we do with the fact that AI already knows how to draw, understand text, and suggest solutions?

The answer might be found not with analysts from Silicon Valley, but with a theologian who lived sixteen centuries ago — Augustine of Hippo.

---

Why Augustine, and why he still teaches us today

In the fourth and fifth centuries Augustine wrote that human beings are constantly torn between two dimensions of time:

• Time of eternity — the idea, the expectation, the promise.
• Earthly time — practice, mistakes, everyday compromises.

This is exactly the gap in which modern designers and product teams live when it comes to AI:

We believe in an “AI-first future,” but we work with very half-baked MVPs.

AI promises:

• automation of routine,
• instant design mockup generation,
• predictive analytics of user behavior,
• intelligent suggestions in real time.

But what actually reaches the user’s screen often turns out to be:

• half-working autocomplete,
• a laggy response from the interface,
• strangely generated text and images.

In other words: the promise is great, but real user value still arrives slowly and with caveats.

The concept that Augustine describes well — deferral: postponing, shifting to the future. The “intelligent AI assistant” always lives a bit further along the time scale — in the next update, the next model, the next feature release.

The designer’s task is not to simply wait for this “smart future,” but to design interfaces and processes so that even today’s imperfect AI is actually useful to people.

---

What users actually want from AI

Users rarely want “AI” as such. They want:

• speed of solving the task,
• predictability of the result,
• the ability to intervene when things go wrong,
• understanding of what and how the system is doing.

When design and UX are built correctly, even relatively weak AI can solve real tasks:

• it saves time searching for information,
• it helps structure thoughts,
• it offers response templates or design solutions.

But when the design is built around an abstract promise of an “ideal smart assistant,” the following suffer:

• user trust,
• clarity of the interface,
• real product value.

---

The “AI-powered” sticker does not yet make a product smart

Clicking an “AI” button in the interface is an act of faith, not a guarantee of results.
For designers and product managers this means:

• if the system can make mistakes — it must always be designed as a helper, not as absolute authority,
• if the user can give feedback or clarify — they must have that option,
• if the interface forces the user’s answer into an AI-shaped template — we need to clearly understand what we gain and what we lose.

---

How the “false promise” of AI reveals itself in design

While business and marketing talk about an “AI revolution,” design teams face much more prosaic problems:

1. AI features are imperfect

   • Answers can be inaccurate.
   • The generation style may not match the context.
   • The interface responds slowly or glitches.

2. They require a lot of “manual work”

   • The user still rewrites AI-generated text.
   • The designer redraws generated layouts.
   • The author still checks facts from the neural network’s responses.

3. They shift responsibility to the user

   • “AI helped to generate the answer, but we are not liable for errors.”
   • This is a subtle shift of responsibility from the system to the person.

As a result, AI becomes less of a solution and more of an extra hand that “helps but gets in the way”.

---

The gap between expectation and reality

Promise:

• “AI will solve your task in two clicks.”
• “The interface will adapt to you.”
• “The system will understand context and suggest the perfect solution itself.”

Reality:

• the system offers several answer options,
• some of them are irrelevant,
• the user still has to make a decision,
• and often has to spend time checking the result.

This doesn’t mean AI is useless — it often helps as a prompting hint, assistant, helper in articulating thought.
The design task is not to hide that fact, but to make it obvious and comfortable.

---

Practical principles of designing with imperfect AI

1. Make it clear the system is a helper, not an oracle

Instead of an abstract “AI answer”:

• Use labels: “Generated by a neural network,” “AI answer (under review).”
• Keep a link to source data if the AI works based on data.

Simple example of structural markup for the interface:

<div class="message message-ai">
  <span class="message-source">AI answer</span>
  <p class="message-text">
    Answer generated by a neural network and may be inaccurate.
  </p>
  <p class="message-content">
    <!-- Here the answer ideally loaded from JS data -->
  </p>
  <button class="button-secondary button-feedback">Give feedback</button>
</div>

.message-ai {
  background-color: #f0f4ff;
  border-color: #c3d9ff;
}

.message-source {
  font-size: 0.8rem;
  text-transform: uppercase;
  letter-spacing: 0.04em;
  color: #6a7b99;
}

This pattern helps users not to perceive AI as absolute authority.

---

2. Give users control

Instead of a single “AI answer” option, use several levels of autonomy:

• hints and phrase clichés only,
• AI-generated text with editing right next to it,
• an answer variant with multiple alternatives.

For example, the interface can show three alternatives generated by the neural network:

const alternatives = [
  "I agree with your proposal. I will add a few details.",
  "I would suggest looking at this task from a different angle.",
  "Let’s consider three solution options."
];

function renderAiAlternatives(altList) {
  const container = document.querySelector(".ai-alternatives");
  container.innerHTML = altList
    .map(
      (text, index) => `
    <div class="alternative">
      <button class="alternative-use" data-index="${index}">
        Use
      </button>
      <p>${text}</p>
    </div>
  `
    )
    .join("");
}

renderAiAlternatives(alternatives);

document.querySelectorAll(".alternative-use").forEach((btn) => {
  btn.addEventListener("click", (e) => {
    const index = e.target.dataset.index;
    document.querySelector(".user-text").value += "\n\n" + alternatives[index];
  });
});

This simple script lets the user choose, not just accept a ready-made answer.

---

3. Make uncertainty and constraints visible

AI often is not confident. The interface can avoid hiding that:

• Add tooltips like “Answer from the neural network may be inaccurate.”
• Show user ratings of answers (“Useful / Not useful”).
• Make source data visible if it exists.

<div class="ai-output-card">
  <div class="ai-header">
    <span>AI answer</span>
    <span class="confidence confidence-low">Low confidence</span>
  </div>
  <p>Answer generated based on data from open sources.</p>
  <footer class="feedback">
    <button class="feedback-button" data-action="like">Useful</button>
    <button class="feedback-button" data-action="dislike">Not useful</button>
  </footer>
</div>

.confidence-low {
  color: #e57373;
}

This kind of pattern helps:

• lower expectations about the system,
• preserve user trust,
• provide feedback for the model and product team.

---

4. Clearly show where and how AI is working

Using AI shouldn’t be a black box:

• Add short explanatory text: “AI analyzes your text and suggests options.”
• Make data sources visible (for example, links to articles if the AI generates answers based on them).
• Indicate the model version or key constraints.

Even a simple text block becomes more powerful:

<p class="ai-disclaimer">
  The system uses text-generation models to provide suggestions.
  We do not guarantee answer accuracy and recommend reviewing them.
  <a href="/ai-faq">Learn more about how AI works in the product</a>.
</p>

This element lowers expectations of “magical intelligence,” but makes the user experience more transparent.

---

Design questions that help bring AI back to an earthly plane

To keep design and UX of an AI-enabled product from turning into pure abstraction of a “smart future,” ask yourself:

1. What exactly does the user do when using this AI feature?
2. Where do they see errors or inaccuracies in the AI answer?
3. How do they currently fix those errors, and how much time do they spend?
4. What would we remove from the product if there were no AI, but the underlying task remained?
5. Where do we put the user in a position to “confirm the AI answer,” and where could we let them formulate their answer themselves?

Design then becomes not a way to “sell AI,” but a mechanism for aligning expectations and reality.

---

When AI becomes real value rather than promise

AI works “for real” when the interface and the team focus on three things:

1. Make routine faster.
   Even if the neural network’s answer needs some tweaking, it’s still 30–70% faster than writing from scratch.

2. Give the user options, not answers.
   Several generated answer variants, templates, clichés — this gives the user a sense of control.

3. Leave space for corrections.
   The interface must be designed around the fact that answers are imperfect and will need editing.

This is much less spectacular than a “fully autonomous smart assistant,” but it works today.

---

Summary: Augustine and interface design with AI

Augustine spoke about the gap between dream and practice, between the ideal and the possible.
Modern UX and product design in the AI era is an almost literal illustration of this idea:

• We dream of a fully autonomous assistant.
• But we work with imperfect models and APIs that produce results of varying usefulness.

The designer’s task is not to promise more than the product can deliver today, but to:

• make clear and honest what the product actually does now,
• reduce the risk of disappointment,
• make imperfect AI work for the user, not for an abstract “AI-powered” label.

The next time you design an interface with AI, ask yourself:

“Where am I currently selling a promise, rather than real value for people?”

The answer to that question is the key to making AI not a false promise of the future, but real help in today’s task.