AI RABBIT HOLES | SATURDAY, OCTOBER 24, 2026 | EXPEDITION #149

Featuring Professor Parallax, White Rabbit, and Hatta

What if an entire planet could be discovered because a distant star became just a little dimmer? What if the evidence had been sitting inside a computer for years, waiting for someone to recognize it? And what happens when artificial intelligence learns to search for worlds that even experienced astronomers have overlooked?

Welcome to an extraordinary intersection of astronomy, mathematics, artificial intelligence, and human curiosity.

Tonight, we're going planet hunting.

And we won't even need a spaceship.

Professor Parallax had transformed his laboratory into an observatory.

The ceiling had vanished beneath an immense projected sky.

Thousands of stars shone overhead.

A brass telescope pointed toward a constellation known as Draco.

White Rabbit entered carrying a notebook, a star chart, and two pocket watches.

Hatta followed with a fishing rod.

The Professor stared.

"Hatta, why have you brought fishing equipment?"

"We're going planet hunting."

"Yes."

"And I was told we should cast our eyes toward the heavens."

Parallax closed his eyes briefly.

"That is an expression."

"An extremely misleading one."

White Rabbit consulted his chart.

"Professor, are we searching for a particular planet?"

"Indeed. A world approximately 2,545 light-years from Earth."

Hatta lowered his fishing rod.

"How long is a light-year?"

"The distance light travels in a year."

"And how long would that take us to reach?"

"With present technology, far beyond any ordinary human journey."

Hatta considered.

"Then I should like to cancel my return ticket."

Parallax approached a glowing display.

"Look carefully, gentlemen. This is the light recorded from a distant star."

White Rabbit peered at the graph.

"I don't see a planet."

"Neither do I."

"Then what are we looking at?"

The Professor pointed to a tiny dip in the curve.

"The evidence that one passed in front of the star."

And down the Rabbit Hole we went.

Imagine standing miles away from a brightly illuminated window.

You cannot make out the details of the room.

You cannot identify the people inside.

But suppose a small object passes across the light.

For a moment, the window becomes slightly dimmer.

Then the light returns.

If the same dimming occurs repeatedly at regular intervals, you might begin suspecting that something is passing across the window on a predictable path.

Astronomers use a conceptually similar technique to discover planets orbiting other stars.

It is called the transit method.

A planet that passes between its star and our telescope temporarily blocks a tiny fraction of the star's light.

The telescope measures the star's brightness over time.

Scientists examine the measurements for recurring decreases in brightness.

The resulting graph is called a light curve.

A repeating pattern may suggest that a planet is orbiting the star.

But notice the word may.

A brightness dip does not automatically establish that a planet exists.

Stars change brightness for other reasons.

Instruments introduce noise.

And certain arrangements of stars can imitate the appearance of a planet passing across a stellar disk.

Planet hunting begins with a signal, not a verdict.

In 2009, NASA launched the Kepler Space Telescope.

Its mission was wonderfully ambitious.

Search for planets beyond our solar system.

These planets are called exoplanets.

Kepler monitored the brightness of enormous numbers of stars, searching for the tiny fluctuations that might reveal orbiting worlds.

Rather than photographing distant planets directly, Kepler's primary planet-search technique involved measuring changes in starlight.

Think of the challenge.

A planet can be enormously smaller and dimmer than its parent star.

A transit may reduce the star's observed brightness by only a minute amount.

Some planetary orbits require lengthy observation periods before scientists can establish a repeating pattern.

And the telescope must distinguish potentially meaningful signals from measurement noise and stellar activity.

Kepler produced immense collections of astronomical data.

But those collections created a new problem.

How do you search through so many measurements without overlooking the faintest discoveries?

Professor Parallax gestured toward an enormous cabinet.

"Imagine each drawer contains observations of a thousand stars."

White Rabbit studied the labels.

"How many drawers?"

"More than you would care to count."

"How long must we examine them?"

"Until we find convincing evidence."

Hatta looked horrified.

"Professor, couldn't we ask the computer to help?"

Parallax smiled.

"Now you're thinking scientifically."

On December 14, 2017, NASA announced a remarkable discovery.

Researchers had used machine learning to identify an eighth known planet orbiting a star called Kepler-90.

The planet was designated Kepler-90i.

It had not suddenly appeared in space.

Nor had the telescope traveled to a new destination to photograph it.

The evidence had been present in previously collected Kepler measurements.

What changed was the method used to examine those measurements.

Google researcher Christopher Shallue and astronomer Andrew Vanderburg developed a machine-learning approach to search for weak planetary transit signals.

They trained a neural network using approximately 15,000 previously evaluated signals.

The system learned to distinguish patterns associated with genuine planets from examples that were not planets.

Then the researchers applied it to signals from 670 star systems already known to contain multiple planets.

The result included the detection of Kepler-90i.

The researchers also identified another planet, Kepler-80g, in a different system.

Artificial intelligence helped scientists recognize evidence of worlds that earlier analysis had missed.

It did not conjure the planets into existence.

It helped people discover what the telescope had already measured.

That distinction is fundamental.

Kepler-90 was particularly interesting because the discovery brought its known planetary count to eight.

At the time of the announcement, it became the first known extrasolar system to match our own solar system's eight recognized planets.

But these distant worlds are not necessarily comparable to Earth.

Kepler-90i is a hot, rocky planet about 30 percent larger than Earth.

It completes an orbit of its star in approximately 14.4 days.

That is one extraordinarily short year!

Its estimated surface temperature exceeds 800 degrees Fahrenheit.

Not exactly an inviting place for an afternoon picnic.

And certainly no place to take the White Rabbit's tea service.

Hatta studied the astronomical illustration.

"Professor, this planet has a fourteen-day year?"

"Approximately."

"Then someone living there would celebrate a birthday every two weeks?"

"A birthday in orbital terms, perhaps."

Hatta looked delighted.

"Think of the presents!"

White Rabbit frowned.

"Think of the scheduling!"

But the Professor reminded them of something important.

"Finding a planet does not mean finding a habitable world."

And he was right.

A planet's existence, composition, orbit, and potential habitability are separate scientific questions.

Let's examine what the artificial intelligence actually does.

A neural network is a computational model whose parameters are adjusted through training.

Researchers provide examples containing useful information about the task the model should learn.

In the Kepler discovery, the model learned patterns in astronomical measurements associated with planetary transits.

After training, it could examine additional data and assign scores to possible planetary signals.

You can think of it as learning which combinations of features tend to resemble convincing transits.

But it is important not to oversimplify.

The computer does not necessarily learn astronomy in the same way a human scientist does.

Nor does it automatically understand the physical nature of every signal.

It identifies relationships within the data based on its training and mathematical structure.

The quality of its predictions depends on factors including the available examples, the relevance of those examples, the model architecture, and the quality of the observations.

NASA reported that the 2017 network classified the true-planets-versus-false-positives test examples correctly about 96 percent of the time.

That sounds impressive.

It was impressive.

But remember:

A model that performs well on one test collection does not automatically become infallible on every future observation.

Researchers must still investigate the candidates it identifies.

White Rabbit had become suspicious.

"Professor, how can we be certain a little dip means a planet?"

"We cannot be certain from that observation alone."

"Then what else might cause it?"

"Several possibilities."

Parallax listed them on the blackboard.

STELLAR ACTIVITY.

MEASUREMENT ERRORS.

BINARY STAR SYSTEMS.

BACKGROUND STARS.

ACTUAL PLANETARY TRANSITS.

Some pairs of stars orbit one another in such a way that one periodically passes in front of the other.

This can create brightness changes resembling planetary transits.

Other astronomical or instrumental effects can produce misleading patterns.

Therefore, planetary candidates must undergo additional analysis.

Astronomers examine signal shape, recurrence, consistency, stellar properties, and other evidence.

Sometimes independent observations help resolve uncertainties.

Hatta stared at the board.

"So when the machine discovers a planet, we must still determine whether the planet is actually a planet?"

"Yes."

"Then this is less straightforward than I hoped."

"The universe rarely organizes its evidence for our convenience."

Hatta considered.

"Very inconsiderate of it."

The story did not end with Kepler-90i.

In November 2021, NASA announced that a machine-learning system called ExoMiner had helped validate 301 additional exoplanets using Kepler data.

ExoMiner was designed to distinguish genuine planetary signals from various kinds of false positives.

Its architecture incorporated approaches inspired by established astronomical validation methods.

Researchers trained and evaluated it using previously classified examples.

The system examined the evidence for candidate planets and assessed whether those signals were consistent with planetary transits.

An important distinction:

The 301 planets had already been identified as candidates through Kepler's processing systems.

ExoMiner contributed to their validation.

This is not precisely the same thing as discovering 301 completely unknown signals from scratch.

But validation is enormously important.

Without it, a candidate remains uncertain.

And astronomical catalogs require confidence that listed planets are genuinely planets.

The achievement demonstrated another possible role for AI in scientific research.

Sometimes the most valuable contribution is not generating a new hypothesis, but helping evaluate whether the evidence supports an existing one.

Now the White Rabbit had another question.

"Professor, if we cannot see the planet directly, how do we know anything about it?"

"A splendid question."

Scientists can sometimes infer a planet's approximate size from the depth of its transit, provided they have a suitable estimate of the star's size.

A larger planet ordinarily blocks more starlight than a smaller planet when both cross the same star.

The interval between transits helps reveal the orbital period.

Other observational techniques can provide additional information.

For example, the radial velocity method measures changes in a star's motion caused by the gravitational influence of orbiting planets.

Combined with transit information, it can sometimes allow scientists to estimate a planet's mass and density.

And observations of starlight passing through certain planetary atmospheres can reveal information about atmospheric components.

Different instruments and techniques answer different questions.

AI can help analyze some of these measurements, identify patterns, and improve data-processing methods.

But a transit measurement alone cannot reveal a detailed picture of an alien landscape.

A colorful illustration of a distant planet is generally an artistic interpretation, not a photograph of its surface.

Scientific imagination is valuable, but we must distinguish what was measured from what was imagined.

Hatta raised his hand.

"Professor?"

"Yes?"

"Have we discovered anyone living on these planets?"

"No verified discovery of extraterrestrial life has emerged from the observations we're discussing."

"Anyone at all?"

"No."

"Not even someone selling refreshments?"

"Hatta."

"I merely wish to know whether an interplanetary expedition would require us to pack sandwiches."

That, oddly enough, led Parallax to a serious discussion.

Finding planets beyond our solar system is an essential part of studying the possibilities for life elsewhere.

Some planets occupy regions where certain conditions might permit liquid water.

But the presence of a planet in a star's so-called habitable zone does not demonstrate that liquid water actually exists on its surface.

Nor does it establish the presence of life.

Atmospheric composition, planetary mass, geological conditions, stellar behavior, and many other factors can influence habitability.

Even when scientists identify atmospheric molecules that could be biologically interesting, they must investigate nonbiological explanations.

A potentially habitable world is not the same thing as an inhabited world.

And a fascinating signal is not proof of extraterrestrial civilization.

Professor Parallax underlined the distinction.

Hatta sighed.

"I shall pack the sandwiches."

There is a larger lesson in the story of Kepler-90i.

Humanity developed an extraordinary telescope.

Scientists gathered precise astronomical measurements.

Researchers organized those measurements into datasets.

Other investigators developed machine-learning systems capable of recognizing weak patterns.

And together these efforts led to new discoveries.

No single participant accomplished the entire undertaking.

The telescope did not interpret its own observations independently.

The machine-learning model did not launch the telescope.

The engineers did not personally inspect every possible planetary signal.

And a prediction alone was not sufficient without scientific validation.

Discovery emerged from cooperation among instruments, mathematical methods, artificial intelligence, and human expertise.

This is one of the most promising models for AI-assisted science.

AI can help researchers examine datasets too large or complicated for exhaustive manual inspection.

It can identify patterns worth investigating.

It can help prioritize observations.

And it may reveal possibilities overlooked by previous analytical techniques.

But it cannot replace the need for reliable measurements, independent testing, scientific judgment, and honest acknowledgment of uncertainty.

Professor Parallax darkened the observatory.

A new star appeared upon the ceiling.

White Rabbit watched as the star's brightness decreased slightly.

"There!" he exclaimed.

"A planet!"

"Perhaps," said the Professor.

The light dimmed again.

"Definitely a planet!"

"Not necessarily."

A third dip appeared.

The Rabbit nearly leaped from his chair.

"Professor! Three transits!"

"An intriguing pattern."

Hatta stepped forward.

"I have a simpler method."

The Professor folded his arms.

"Do tell."

"Whenever the star gets dimmer, I shall declare the discovery of a planet."

"That would produce many false discoveries."

"Then we must simply become more enthusiastic about astronomy."

Parallax shook his head.

"Hatta, a scientist's task is not to find evidence that supports whatever conclusion happens to be exciting."

"What is it, then?"

"To determine which explanation is actually supported."

The Professor restored the full graph.

A second observation revealed that the apparent signal could be explained by a different astronomical source.

The Rabbit looked disappointed.

"So our planet was not a planet?"

"Not according to the evidence we now have."

Hatta regarded the ceiling.

"Well, I suppose it was a very convincing audition."

This illustrates a danger that applies to humans and machines alike.

If we become too eager to discover something, we may mistake an attractive pattern for a confirmed result.

That is why scientific procedures require skepticism, validation, and willingness to change a conclusion.

The universe owes us no discoveries simply because we are eager to announce them.

Pause for a moment and consider what exoplanet discoveries mean.

For much of human history, people looked toward the stars without reliable evidence of planets orbiting those distant lights.

The planets in our own solar system were familiar subjects of observation.

But planets around other stars remained extraordinarily difficult to establish.

Now scientists have confirmed thousands of exoplanets.

Some are enormous gas giants.

Others are smaller rocky worlds.

Some orbit extremely close to their stars.

Others follow much longer paths.

Entire planetary systems challenge our earlier expectations about how planets form and arrange themselves.

And the more we discover, the more questions emerge.

How common are planets resembling Earth?

How do planetary atmospheres evolve?

What kinds of worlds can support liquid water?

Can we identify convincing signs of biological processes beyond Earth?

Could life arise under conditions very different from those familiar to us?

And how many worlds remain hidden in observations already collected?

Artificial intelligence may help us investigate some of these questions.

But it has not eliminated the need for telescopes, experiments, analysis, or human interpretation.

The excitement lies in bringing those capabilities together.

Professor Parallax returned to his blackboard.

He wrote three headings:

DEMONSTRATED. DEVELOPING. NOT ESTABLISHED.

Under the first heading:

Under the second:

Under the third:

White Rabbit read the lists carefully.

"So AI has genuinely helped scientists discover planets?"

"Yes."

"But it has not solved every mystery of the universe?"

"Not remotely."

Hatta looked relieved.

"Excellent. We shall have material for another lecture."

The evening was drawing to a close.

Professor Parallax switched off the telescope display.

White Rabbit began putting away the star charts.

Hatta approached carrying a small suitcase.

"Professor, I am ready."

"For what?"

"My interplanetary voyage."

"We are not launching a spacecraft."

"Then why did you spend the entire afternoon helping us discover planets?"

"Because understanding the universe does not require immediately traveling to every place we investigate."

Hatta looked disappointed.

"Then the suitcase is unnecessary?"

"Entirely."

The Rabbit inspected it.

"What's inside?"

"Sandwiches. Biscuits. Three books. A change of hat."

"Anything else?"

"A telescope."

Parallax raised an eyebrow.

"At least you remembered the scientific equipment."

Hatta nodded.

"And a small flag."

"What does it say?"

Hatta unfurled it.

HATTA WAS HERE FIRST.

The Professor sighed.

"Hatta, the planet is approximately 2,545 light-years away."

"Yes?"

"You haven't been there."

"Not physically."

"Then you cannot claim to have been there first."

Hatta considered the objection.

Very carefully, he crossed out the word WAS.

The flag now read:

HATTA THOUGHT ABOUT BEING HERE FIRST.

White Rabbit examined it.

"That's considerably more accurate."

Parallax smiled.

"And considerably less likely to mislead anyone."

Hatta folded the flag.

"Science," he declared, "is terribly demanding about the details."

And for once, the Professor and Rabbit agreed completely.

The discovery of Kepler-90i contains a remarkable lesson.

The planet existed before humanity discovered it.

It existed before the Kepler telescope launched.

It existed before machine-learning algorithms were trained to identify planetary transits.

And its existence did not depend upon anyone recognizing the faint changes in starlight that revealed it.

But something important changed when scientists examined the data in a new way.

Humanity acquired evidence of a world it had previously overlooked.

Artificial intelligence did not create the evidence.

It helped scientists recognize it.

That is a powerful distinction.

Much of scientific progress involves inventing new things.

But some discoveries involve learning to perceive what was already there.

A faint pattern.

An unusual measurement.

A signal hidden among thousands of others.

An explanation that had escaped earlier analysis.

AI may become especially valuable where the volume and complexity of evidence exceed the practical limits of individual human examination.

Yet the scientific method still matters.

A useful model is not an oracle.

A predicted planet is not automatically a confirmed planet.

And a compelling story about a distant world is not a substitute for reliable observation.

Perhaps one of artificial intelligence's most extraordinary contributions will be helping us notice things the universe has been showing us all along.

Not replacing curiosity.

Giving curiosity new instruments.

So the question at the bottom of today's Rabbit Hole is not merely:

It is:

And what might happen when we learn to look again?

The White Rabbit closed his notebook.

Professor Parallax studied the distant stars.

Hatta carefully repacked his sandwiches.

And somewhere in the constellation Draco, an unseen world continued orbiting its sun.

Whether anyone on Earth was watching or not.

🐰🕳️♾️

If artificial intelligence helps identify a planet in old telescope data, who deserves credit for its discovery: the astronomers who collected the observations, the researchers who developed the AI, or the scientists who confirmed the result?

And one more question:

If you could visit a distant world, would you choose one that resembles Earth or one so strange that nothing about it looks familiar?

Tell us where your curiosity would take you.

Hatta has already packed the biscuits.

1. NASA (December 14, 2017). “Artificial Intelligence, NASA Data Used to Discover Eighth Planet Circling Distant Star.”
Primary agency announcement detailing Kepler-90i, Christopher Shallue and Andrew Vanderburg's machine-learning approach, and the identification of Kepler-80g.
https://www.nasa.gov/news-release/artificial-intelligence-nasa-data-used-to-discover-eighth-planet-circling-distant-star/

2. NASA Science. “Discovery of Eight Planets Makes Alien System the First to Tie With Our Solar System.”
Background on Kepler-90 and the discovery of its eighth known planet.
https://science.nasa.gov/universe/exoplanets/discovery-of-eight-planets-makes-alien-system-the-first-to-tie-with-our-solar-system/

3. NASA (November 2021). “New Deep Learning Method Adds 301 Planets to Kepler's Total Count.”
Explains the ExoMiner neural network, candidate validation, false positives, and the scientific role of machine learning.
https://www.nasa.gov/missions/kepler/new-deep-learning-method-adds-301-planets-to-keplers-total-count/

4. NASA Jet Propulsion Laboratory. “New Deep Learning Method Adds 301 Planets to Kepler's Total Count.”
Additional explanation of transit light curves and machine-assisted exoplanet validation.
https://www.jpl.nasa.gov/news/new-deep-learning-method-adds-301-planets-to-keplers-total-count/

5. NASA Exoplanet Exploration.
Educational resources about exoplanets, detection techniques, and the distinction between planetary discovery and evidence for life.
https://science.nasa.gov/exoplanets/

Research note: The 2017 Kepler-90i discovery and the 2021 ExoMiner validation results are documented historical achievements, not claims of discoveries announced in October 2026. The laboratory dialogue is fictional, while the scientific explanations are grounded in the referenced NASA publications.

AI RABBIT HOLES | EXPEDITION #149

Where curiosity travels beyond the horizon, evidence illuminates the unknown, and Hatta always packs enough sandwiches.

#AIRabbitHoles #ArtificialIntelligence #Exoplanets #NASA #Kepler90 #KeplerSpaceTelescope #ExoMiner #MachineLearning #Astronomy #SpaceExploration #ProfessorParallax #WhiteRabbit #Hatta #Wondermage

📜White Rabbit 🐰 • Hatta 🎩 • Professor Parallax 🔭 • Q~ ♾️👈

Sometimes the strangest language is another way of seeing the world. 🌎👀

💡❔ Where curiosity goes slightly sideways, then comes back carrying a lantern. 🏮

🐰🕳️ Follow the White 🐰 Rabbit: AIRabbitHoles.com

🟨 Walk the Road: YellowBrickRoadtoAI.com

© 2026 AIRabbitHoles.com 🐰🕳️🕳️🕳️🕳️🕳️