A new virtual celebrity DApp game, “Gene A.I.dols” (pronounced “Gene Idols”), was announced on February 14, 2019. In a nutshell, Gene A.I.dols is a blockchain-based DApp game in which players create and collect their own unique virtual idols (celebrities).
At first glance, it may look like a collectible game, but under the hood are a multiple cutting-edge technologies such as AI-based facial image generation and speech synthesis, and even the achievement of true ownership via the blockchain. We asked developers Yoshikazu Nishimura and Takuya Obata how the project was conceived and what role AI and blockchain play.
Recording idols’ “genes” on the blockchain
— The DApps game project Gene A.I.dols has been announced. What are the features of the game and the gist of the project?
Yoshikazu Nishimura: Gene A.I.dols are so-called “virtual idols” but the game is unique in that while the idols are virtual, they seem like they could exist in real life – they each have their own appearance and voice, they can talk with users, and new idols can be created from existing ones.
Each idol has its own artificial genes, so no two idols share the same face or voice. Furthermore, combining the artificial genes of two idols can generate a new idol. As the word “gene” implies, the artificial genes are the source data when generating idols. Making these into tokens on the blockchain gives users complete ownership which they can freely transfer or trade.
The game-like elements are what idols you get at the beginning and what kind of idols you can create by combining them. In addition, the idols are not just still images – they make facial expressions and move, and we even plan to make them able to interact with users. Players get to enjoy acquiring their favorite idols, watching them, talking with them, and forming a collection.
*1 ICOVO AG also operates Blockchain Insight.
— This DApp uses a blockchain, but how will it be offered?
Takuya Obata: We’ll first release it as a web application for browsers. Since it’ll run on Ethereum, users will need a wallet like MetaMask, and as it is premised on the use of Ethereum, we envision the target users to be those with some experience using cryptocurrency. However, since the idols are unique to the user and we expect users will want to take them with them and to show others, we intend to make it available on smartphone as soon as possible.
In order for a wider audience to enjoy the game, we need to do everything we can to eliminate the cryptocurrency hurdle, e.g. making it possible to use common credit cards, which we want to do eventually. Gene A.I.dols will progress through various project stages that include upgrades and additional features, but the first phase is the release scheduled for this spring.
Ownership and extensibility based on tokenization
— How did you come up with this project?
Nishimura: There are already virtual idol DApps like Crypko and CryptoKanojo. Each has its own features, but I was most interested in the ability to generate character images with AI and create new characters by combining them.
As an engineer, I’m also involved in blockchain for ICOVO AG and AI for alt Inc. We came up with Gene A.I.dols by combining these two technologies. What makes it different from Crypko and CryptoKanojo is that the idols’ visuals are not illustrations but rather realistic images of women, almost real photographs, and their voices are unique and can interact with users.
Obata: In terms of using non-fungible blockchain tokens*2 for the idols, there is a precedent called CryptoKitties that has cat characters. There are two major reasons for converting items in collectible games into non-fungible tokens.
*2 Non-substitutable tokens guaranteed by the blockchain to be one-of-a-kind. Standardized as ERC 721 on Ethereum.
The first is that recording something as a token on the blockchain means that it will never disappear. Even if the game stops being offered or the operating company goes bankrupt, it can continue to exist as long as there’s the blockchain. For the user, this is true ownership, and being able obtain this never-before-seen, definite sense of ownership is something unique to DApps.
It’s difficult to make everything decentralized, but it’s still possible to transcend the boundaries of the game. For example, we could see a game emerge in which the idols ride on the CryptoKitties cats in races. Even if people don’t want to make a new game, users could create idol contests, or add market features that allow users to trade or rent idols, as this is also part of the concept.
Idols based on the artificial genes of users
— It really seems like the most important element in Gene A.I.dols is the artificial genes. What exactly are these?
Nishimura: The artificial genes are the parameter information required to generate the idols, which is tokenized and put on the blockchain. Each face, voice, and conversation pattern is a vector with hundreds of dimensions, so there’s considerable diversity. When creating a new idol from two idols, the originality comes from adding randomness to the characteristics inherited from the two. Although a new idol may resemble an original idol to a certain extent, it will never be exactly the same.
Regarding the generation algorithm, I think it’s better not to lay it out – it’s more stimulating to users’ imaginations that way, and verifying and researching can be part of the fun. We’re also thinking about making it possible for users to generate artificial genes from their own facial images and voices and combine that with the idol genes to create new idols. Artificial genes would allow humans in the real world to enter the virtual world, which would make the game even more fun.
Realistic non-existent girls generated through deep learning
— You mentioned that you’re using AI in Gene A.I.dols – specifically how is it being used?
Nishimura: It depends on how you define AI, but the AI in Gene A.I.dols is machine learning and deep learning. For example, we’re using a deep learning mechanism called a GAN (Generative Adversarial Network) to generate the facial images. Using a GAN, we’re able to quickly enhance the accuracy and quality of the generated images by making the side that generates the image and the AI (neural network) side that identifies it compete with each other. We first used actual photographs of women’s faces to train the Gene A.I.dols’ facial image generation, then generated high-quality facial images using the GAN. Looking at the sample images, they may look like real women, but they are all AI-generated, non-existent girls.
The idols’ voices are also speech models generated through machine learning to sound natural. Previously, in order to freely generate speech from someone’s voice using speech synthesis, that person had to first read a large number of sample sentences, data that was then analyzed to create a speech model. However, for Gene A.I.dols, we use a technology called speaker adaptation, which can generate speech models from a very small amount of speech data, to easily create numerous unique speech models. The idols have unique voices as well as faces, and that information is also recorded in the artificial genes. When idols are combined, their voice qualities are passed on as well.
— You mentioned that dialogue will also be possible, but does this mean that each idol has something like a personality that changes based on its environment and can be passed on?
Nishimura: Implementation of a dialogue function is farther down the road. We can see the technology for a dialogue engine, but we’re still considering things like how to position it within the idol’s personality, how advanced to make it, and whether to incorporate elements like change and growth due to environment. For the moment, we’re planning to have them already possess some sort of personality when they’re born.
Aiming for a spring release with full bodies and VR compatibility in the future
— What are you excited about in the future?
Nishimura: Although still in development, it’s becoming possible to generate facial images. We plan to report our progress on the official website up until the release. There are a lot of things still in the planning stage that haven’t been finalized yet, but ultimately we’d like them to have not only faces but also bodies. Having full bodies increases expressiveness with options like dancing and changing clothes, and makes it easier to add VR compatibility or turn them into Vtubers. The hurdles to development are high, but if we get a lot of users, we’ll be able to continue, so once it’s released I hope everyone gives it a try.
Obata: Since it’s a DApp game, we also have high expectations for third-party expansion and user-led planning. Of course, as the original developers, we’re also thinking about planning campaigns to stir up excitement, so stay tuned for that. Finally, even though the theme at this time is idols, I think this model has a lot of development potential as a platform, for example replacing the idols with existing popular characters. If there’s any company out there that owns IP content and is interested in DApp games, we’d love to work with you.
Interview date: February 1, 2019
Editor & Photographer: Makoto Nakazato
CTO and Co-Founder, ICOVO AG / CTO, alt Inc.
Nishimura researched DNA computing while studying at Waseda University, then received an M.S. in Computational Molecular Biology from the University of Southern California. He entered Oracle Corporation Japan as a consultant in 2005, and then in 2011 became a director at Link and Partners. In 2013, he established Comps Co., Ltd., taking the position of representative director. In addition to developing and consulting on academic projects such as natural language processing and machine learning, it spun off G. U. Lab, which specializes in blockchain research and is actively engaged in developing and providing a mobile wallet called Tachyon Wallet and Ethereum development ecosystem tools. In November 2017, Nishimura became the CTO of alt Inc., followed by co-founding ICOVO AG in Zug, Switzerland, and becoming its CTO as well in March 2018. He received the Best Innovation Award at the Global Blockchain Summit 2016 for his proposal of a blockchain technology-based location information platform, and has co-authored “Blockchain Application for Beginners – Smart Contract Development on Ethereum” (Shoeisha, 2017).
COO, ICOVO AG
Obata graduated from the University of Electro-Communications’ Graduate School of Informatics and Engineering in 2011. After developing industrial robots at Canon Inc., he became a freelance engineer in 2014, participating in projects such as dialogue engines and recommendation engines. In 2016, he established WLTC, Inc. for blockchain-related development and, among other things, has done back-end development for exchanges and developed smart contracts. In July 2018, he became the COO of ICOVO AG, where he is working hard to improve the health of the cryptocurrency fundraising market. While overseeing all business operations, he remains a developer, writing code for ICOVO products such as DAICOVO and Gene A.I.dols. His areas of expertise are mainly blockchain, natural language processing, and game AI. He has won the top prize at various competitions for Shogi AI and held the top spot in one category from 2011 to 2015.