Tony Stark had never been the sort of person to admit defeat easily.
He spent several days gathering the game records of chess masters from around the world.
After finding a way to rapidly digitize the material and feed it into J.A.R.V.I.S.'s database, he finally reconnected it to Miss A.
Although he noticed that the score between the two AIs had reset to zero, he didn't think much of it and allowed J.A.R.V.I.S. to begin playing again.
At first, things went exactly as he had hoped.
J.A.R.V.I.S.'s win count climbed steadily, and Tony's mood improved right along with it.
But before long, something started to feel wrong.
Tony had confidence in the AI he had trained with thousands of master-level games.
Still, even if Miss A remained at the same level as before, there was no way it should be losing every single game without ever winning.
So he halted the match and randomly reviewed one of the recent games.
Miss A's play was terrible.
It looked like the work of an elementary-school student who had just learned the rules.
Beyond following the legal movement of the pieces, everything else seemed to be decided by rolling dice.
Tony went back to review the very first game.
Sure enough, it was the same familiar pattern.
You play your game, I play mine.
Miss A was utterly crushed.
In the second game, after the colors switched, Miss A copied J.A.R.V.I.S.'s opening and subsequent moves from the previous match.
No matter how J.A.R.V.I.S. responded, White stubbornly followed the script of the first game.
Once a piece was captured and the sequence diverged from the original, Miss A's play immediately descended into chaos.
Tony frowned.
Then suddenly, a thought struck him.
He attempted to review some of the games from several days earlier—the matches played during the latter stages of the previous session.
Miss A responded:
> No relevant data available.
This was the fourth type of response Tony had ever seen from the AI.
The first was its self-introduction.
The second was the explanation of chess rules.
The third was game reconstruction.
And now there was a fourth:
No data available.
He typed:
> Who are you?
The reply appeared instantly:
> Hello. I am a self-learning artificial intelligence, Miss_A_121298.
Tony's eyes widened.
The version date had changed.
At that moment, he finally understood what this game was actually about.
It wasn't chess.
Nor was it a contest to see who could gather more chess records and train a stronger AI.
It was a competition in learning ability.
A contest of underlying algorithms.
A battle of mathematical thinking between two genius minds.
Only now did Tony find the game genuinely interesting.
This was no longer simple competitiveness.
He immediately stopped the J.A.R.V.I.S.–Miss A matches.
Using a fully trained AI against an AI that possessed only foundational logic was essentially equivalent to feeding your training data directly to the opponent.
The outcome had no meaning.
So Tony dug through his archives and retrieved an early prototype version of J.A.R.V.I.S.
Then he began rebuilding it according to the same rule structure Miss A used.
After all, chess rules weren't entirely fixed.
If the two AIs weren't operating under the exact same framework, the contest wouldn't be fair.
Which rule set they used didn't really matter.
The purpose of most rule revisions throughout chess history had been to maintain fairness between White and Black and between first and second move advantages.
But in this system, colors alternated every game.
Victory wasn't measured by a single match.
Over a large enough sample size, both AIs would face identical conditions.
That made the competition fair.
After establishing the foundational logic, Tony personally played several test games.
Once he confirmed there were no bugs in the move generation, he deleted all the records.
What remained was a blank AI framework.
In a display of his characteristic humility, he named the new AI:
A.L.B.B.Y.
(A Little Bit Better than You)
Naturally, he included a version number as well.
Once everything was ready, he connected it to Miss A and allowed the two AIs to begin playing.
Tony watched the opening games.
Because Miss A retained memories of its previous matches against J.A.R.V.I.S., its openings already looked reasonably competent.
But whenever a piece was captured and the flow of the game disrupted, it immediately fell apart.
Watching two clueless AIs blunder into each other, Tony laughed a few times before disabling the visual replay system and letting them continue in the background.
For a contest like this, anything under ten thousand games was meaningless.
Only after one hundred thousand games—or more—would the learning curves begin to stabilize.
So Tony had no choice but to wait.
Even though he hated waiting.
That evening, he invited a large number of women over to distract himself and stop obsessing over the progress of the AI battle.
Finally, around noon the following day, he terminated the match between A.L.B.B.Y. and Miss A.
The first thing he did was review the final game.
The moves looked surprisingly sophisticated.
The ridiculous blunders from earlier were gone.
No more situations where a queen was about to be captured and, instead of responding, an AI casually ignored the threat and played elsewhere as though nothing were happening.
Looking solely at the number of victories, Miss A still came out ahead.
Examining the win-rate graphs, Tony noticed that throughout the early and middle stages, the two AIs had remained fairly even.
Only in the later stages did the gap emerge.
Miss A's learning efficiency was noticeably superior to A.L.B.B.Y.'s, leading to more wins and fewer losses.
In other words, Henry's self-learning AI architecture was simply better than J.A.R.V.I.S.'s.
J.A.R.V.I.S. had been developed entirely on the Hollywood Kid supercomputer.
Tony was convinced Henry had observed the entire process.
There was plenty of evidence.
From time to time, Henry had left behind pieces of code capable of improving computational efficiency.
That much was undeniable.
Given that Henry had effectively watched J.A.R.V.I.S.'s development from start to finish, it wasn't surprising that he had created a superior AI in the form of Miss A.
Before discovering the chess website, Tony hadn't even known Miss A existed.
Considering the exchange, if Miss A had turned out to be inferior, Tony would have wanted to track Henry down just to laugh at him all day.
Then another thought occurred to him.
Processing efficiency.
Tony began analyzing the response times of the two AIs.
Both operated within millisecond and microsecond ranges.
Yet there remained a slight but measurable difference in average thinking time.
Ignoring algorithmic differences, that implied something significant:
The computer running Miss A was every bit as powerful as Hollywood Kid.
Originally, Tony had worried that deploying his AI on a supercomputer created an unfair advantage.
Now it appeared that whatever hardware Henry was secretly using was not inferior to Hollywood Kid at all.
That realization actually reassured him.
It meant he could unleash the full power of the supercomputer without feeling guilty.
Then another idea struck him.
If algorithm design and mathematical ability formed the first two layers of competition...
Then computer hardware itself had become the third.
Thinking about it that way, Tony suddenly understood why Henry had resigned as CEO of Stark Pictures.
A man capable of building something like this would probably find running a movie studio unbearably dull.
At least Tony himself would feel that way.
Which brought him back to an earlier question.
Why had Henry accepted the position in the first place?
Was it really just to keep an eye on the West Coast for him?
Henry had once said that he had forgotten something.
So what exactly was it that he had forgotten?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For 40 advanced chapters, visit my Patreon:
Patreon - Twilight_scribe1
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~