I, for one, embrace our new artificial intelligence rulers.
-Fresh Structure Deck: Blue-Eyes Max
This deck revolves around a resilient “Blue-Eyes” monster as its trump card. Numerous “Blue-Eyes” support cards such as “The White Stone of Ancients” and “Blue-Eyes Jet Dragon” possess effects tied to your Graveyard. Monitoring your Graveyard’s status is crucial. Once the stage is set and you unleash formidable creatures like “Blue-Eyes Chaos MAX Dragon,” victory is within reach!
Included cards:
- Blue-Eyes White Dragon (UR)
- Blue-Eyes Alternative White Dragon (UR)
- Blue-Eyes Chaos MAX Dragon (UR)
Release Date: February 7th (subject to time zones)
Fresh Selection Pack: Journey to Tomorrow
Included cards:
- Diabellstar the Black Witch (UR)
- Snake-Eyes Flamberge Dragon (UR)
- Cosmic Quasar Dragon (UR)
- T.G. Glaive Blaster (UR)
- Emperor Charles the Great (UR)
Release Date: February 7th (subject to time zones)
Fresh Selection Pack: Name of Champions
Included cards:
- Lady Labrynth of the Silver Castle (UR)
- Destiny HERO – Destroyer Phoenix Enforcer (UR)
- Baronne de Fleur (UR)
- Dingirsu, the Orcust of the Evening Star (UR)
- Accesscode Talker (UR)
Release Date: February 7th (subject to time zones)
Moreover, in reference to the A.I. system, a brief presentation was streamed on the second day of the event (commencing around 47 minutes).
Ken’ichi Kataoka (Master Duel/Duel Links Senior Producer)
Mr. Hattori (Programmer)
Q: Can you provide a simple overview of this A.I. project?
- Mr. Hattori: This initiative involves a system that employs Python for Master Duel. Essentially, with knowledge of A.I. and the Yu-Gi-Oh! OCG, individuals can create a Duel A.I. capable of playing Master Duel.
Q: By “individuals,” do you mean even those outside Konami?
- Mr. Hattori: Indeed, that is correct. Initially, we aim to witness a Duel between two A.I.s crafted by individuals or groups external to the Master Duel production team.
The visual on the screen essentially illustrates the creation of a specialized interface for the A.I. to directly engage with the game (top image). The text in the center of both images delineates that the A.I.’s actions, probability assessments, and more contribute to a feedback system. The lower image elucidates the possibility of utilizing a console simulator to conduct Duels rapidly with a text-only interface.
The two sections on the right essentially detail the organization and transmission of information from the Duel and board to the A.I., which in turn responds with its inputs and requests. The A.I. agent (coded in Python) functions as an A.I. framework / trained model.
Q: This appears significantly distinct from the typical Master Duel screen; could you provide an overview of what we are observing?
- Mr. Hattori: Certainly. Notably, at the bottom-left corner, a percentage is visible. This percentage reflects the winning odds based on the A.I.’s moves. While the bottom-left box displays the overall estimated winning odds, the bottom-right section showcases the top 5 moves. Consequently, we focus on the A.I. selecting the #1 move. Presently, the winning odds are low, but a move is anticipated to substantially enhance the chances of winning, warranting close attention.
(The A.I. utilizes The Melody of Awakening Dragon to search for Blue-Eyes Chaos MAX Dragon, facilitating its Ritual Summon and elevating the winning odds to around 80%)
Q: At first glance, it seemed like a Duel between two CPUs, but it’s more than that, correct?
- Mr. Hattori: Indeed, these are not CPUs but external programs operating within Master Duel. In essence, machine learning is employed to educate the A.I.. Notably, the A.I. does not leverage user data; instead, it learns through repeated Duels without prior knowledge of the rules.
Q: Through these repetitions, approximately how many Duels has the A.I. engaged in?
- Mr. Hattori: It has undergone training by Dueling the Solo Mode CPU over 100,000 times.
Q: So, devoid of explicit rule instruction, the A.I. has evolved through over 100,000 Duels?
- Mr. Hattori: Yes. “It has probably played more Duels than humans!”, “That’s some tough training!”
Q: The on-screen information is quite comprehensive and intriguing!
- Mr. Hattori: Indeed, we believed it would captivate viewers to witness the A.I.’s thought process displayed on screen.
Ken’ichi: While we specialize in game development, A.I. expertise is not our forte. Thus, we found it intriguing to explore the potential of A.I. development by experts in the field and the resulting Duels between such A.I.s. This project was initiated with the question, “Who can create the most formidable A.I.?” Additionally, the prospect of A.I.s Dueling human players is enticing, prompting curiosity about the outcomes of matches between “the world’s #1 A.I.” and “the champion of the World Championship.”
Q: Can individuals external to Konami participate in this endeavor as well?
Ken’ichi: Absolutely. However, access to this system for widespread A.I. creation will not be immediate. Currently, [protocols/rules/terms of service] are being formulated, with plans to post positions for A.I. Duel creators on the official Master Duel website within the year.
Mr. Hattori: Initially, we anticipate engagement from entities proficient in A.I. development. Ideally, competitions could ensue among universities or similar entities.
“That scenario sounds fascinating, with each university crafting its unique A.I. and subsequently pitting them against others!”
Ken’inchi: Indeed, A.I.s fashioned by universities offer intriguing perspectives.
“Yes, their distinct characteristics reflect the creators’ individuality.”
Ken’ichi: While the preceding dialogue primarily caters to A.I. enthusiasts, benefits extend to all Duelists.
Mr. Hattori: As demonstrated in a previous video, please observe:
Mr. Hattori: By leveraging the A.I.’s winning odds estimation function, a feature akin to the renowned advantage meter in go/shogi can be developed.
Q: So, as depicted in this image, the turn player holds the advantage, as indicated on screen, correct?
Mr. Hattori: Precisely.
“Thus, at a glance, viewers can discern the player with Mystical Space Typhoon is favored to win!”
“For tournament commentators, a quick glance at the screen reveals the leading player.”
“[…]and there might be instances where a player with a mere 5% chance of winning draws a card and swiftly alters the odds!”
Q: Are there additional features to highlight?
Mr. Hattori: Indeed, we are enthusiastic about a functionality that predicts the identity of face-down cards. Let’s refer to the next slide…
Mr. Hattori: Here, we witness the A.I. attempting to anticipate the face-down card. [Briefly discusses potential inaccuracies in predictions, varied A.I. responses, etc.]
“Hence, the A.I. speculates: Could that face-down card be Ring of Destruction?”
“It seems to have encountered RoD excessively.”
“With whom did it engage in those 100,000 Duels?”
“This is how the A.I. acquires distinctive traits based on its opponents.”
“It evokes the realm of VRAINS, where Dueling data analysis is feasible.”
Ken’ichi: The essence of Master Duel lies in the enjoyment of both participants and spectators. Therefore, we strive to enhance the presentation of the game’s highlights in the simplest manner possible, fostering a broader player base. The evolution of Master Duel is geared towards Duelists and their supporters. Stay tuned for the future of Yu-Gi-Oh digital games!
A live demonstration ensues. Both Ken’ichi and Mr. Hattori caution that despite the A.I.’s extensive training of over 100,000 Duels, it is still considered “young” and “developing” in A.I. terms, potentially leading to unexpected moves. Mr. Hattori emphasizes that when the A.I. intends to attack and encounters an opponent’s face-down card, its decision-making factors in potential face-down cards. Once it deems the card non-threatening (e.g., Eternal Soul instead of Mirror Force), it proceeds with the attack. He also notes that the A.I. possesses a rudimentary understanding of the opponent’s Deck to make such assessments. When questioned about teaching the A.I. about specific Decks to enhance its adaptability, Mr. Hattori confirms the feasibility of such training methods. The A.I. employs Ring of Destruction during the End Phase, prompting inquiries about the learned nature of this move. Mr. Hattori explains that through numerous Duels, the A.I. has developed a preference for maneuvers leading to victory. Although unfamiliar with Yu-Gi-Oh rules, the A.I. comprehends the advantageous outcomes of certain actions.
(The Duel concludes prematurely due to time constraints)
