Have you ever tried Snapchat’s new artificial intelligence-based filter service that launched in May? The popularity of this feature, which allows users to change their gender, or create a baby-face version of themselves, has led to Snapchat becoming an internet sensation. Snapchat, which has been experiencing 0% growth for the last four consecutive quarters due to its counterparts’ imitation strategy, successfully managed to rebound, in terms of growth, after the launch of its new artificial intelligence filter service.
After the launch of this service, Snapchat’s weekly installment quadrupled, Snapchat shares increased in price by 35%, and the number of Google searches for ‘snapchat’ increased by 50%. As this case suggests, if artificial intelligence can be applied correctly, it greatly enhances the application’s appeal. Artificial intelligence used in YouTube and Amazon’s personalized recommendations, augmented realities of Pokemon GO, and camera filters that have become essential functions of social media, are expected to grow to an 89 billion dollar market by the year 2025, according to Statista.
The vision and limitations of artificial intelligence applications
Despite seemingly obvious, the development and servicing of artificial intelligence are never easy for most app developers. For several years, Global IT companies have developed an open-source machine learning framework for artificial intelligence. However, there are still limitations to overcome, including quality data and experts, and efforts to apply artificial intelligence on a computer with high performance. Even R&D, which costs anywhere from tens of thousands, up to billions of dollars, is difficult to initiate without strong confidence from the management level.
As a Service is what allows us to overcome such limitations. Machine Learning as a Service (MLaas) is currently led by cloud service providers such as Amazon and Google, and one can easily apply advanced machine learning models from the menu, and use his data on the cloud. According to MarketReportsWorld, the market for MLaaS is growing by more than 50% per annum, and will have a market size of 6.7 billion dollars by 2023.
Nonetheless, there is still room for improvements. The first issue has to do with the limited number of service providers. Cloud companies are major firms that are market leaders, and possess top quality machine learning models with tremendous amounts of data. However, there are also small to medium sized companies that also specialize in machine learning with extremely creative individual developers. Nevertheless, for a variety of reasons, consumers’ choices and participation are limited to major cloud companies.
- Machine learning service start-ups: Partnerships with large companies with sufficient funding and needs (B2B) are a viable means of survival, limiting the number of customers using their services.
- Creative individual developers: Even before they can worry about outlets for their service, deploying and developing machine learning services cost them thousands of dollars every month, limiting their active participation.
Secondly, we have the lightening of machine learning service deployment, as well as improvements in composability. For those who have difficulty understanding what this means, think of Apple computers as an analogy. They have prefabricated computers, and computers that are sold as finished products. The former has the advantage of being high-quality, and is immediately available for use, but is more expensive, and has incompatible parts. On the other hand, the latter, despite being of a lower quality and less complete, has the advantage of allowing people to configure any specification, as the parts are combined, with each part being relatively cheaper. Machine learning markets are very similar; currently, the market only offers machine learning services in the form of finished products that can be used immediately. However, if there were more participants in the market, and more services could be combined to bring more powerful services, it could lead to endless possibilities of innovation.
Microservices, and artificial intelligence utilized in unit functions
Recently in places like China that require large scale construction, modular construction methods have been applied. Part of a building is constructed in advance, and each part is brought to the construction site and assembled, which shortens the time, and eases the process of construction at the same time. Artificial intelligence applied to applications are evolving in the same way.
In the past, development was done monolithically. We then moved onto a process in which each module was developed as a microservice, and finally to a process in which modules were developed as a function (unit of logic used when coding), allowing developments to happen separately. The benefits of developing unit functions are:
- App developers: can move away from having to spend time and effort to develop their apps from A to Z, and focus on developing the app’s unique features. A core, but additional feature, such as AI filters, can be implemented by a function (using an API method) developed by a third party, and the app developer would pay only for the number of times it has been executed.
- Microservice Developers / Unit function developers: Small groups or individual developers can focus solely on development, without having to worry about deployment or monetization of artificial intelligence microservices and functions. Every time a function is called, it would generate revenue.
Until now, the only way to develop large-scale solutions such as an artificial intelligence assistant or an autonomous vehicle was being part of a large company. In the future, we will see the development of artificial intelligence and personal livelihoods through the development of micro-services on AI Network, which makes use of blockchain. At the same time, we can expect a rapid development in the artificial intelligence industry, as knowledge in the field of AI programming can be meticulously accumulated, just as Wikipedia, through collective intelligence and active engagement of individuals.
Cases of microservice AI applied to applications
aFan is a blockchain-based social media app that was developed as a function since its planning stage, with the application of artificial intelligence in mind. In the process of developing aFan, we not only need basic functions of social media apps with aFan’s exclusive features, but also smart contracts for the use of blockchain and cryptocurrencies. Generally speaking, it’s difficult for small start-ups to develop an AI that can be attached to an app, but aFan managed to do so through AI Network.
The machine learning model used is NIMA (Neural Image Assessment), an open source model that allows one to evaluate images in both technical and aesthetic terms. For technical evaluations, we use elements such as noise, blurring, and compression artifacts, whereas we focus on the emotion and beauty of the image for aesthetic assessments. Images are rated out of 10.00, and photos that receive high scores are actually very similar to those preferred by the general public.
In afan, a function that evaluates uploaded pictures on a daily basis through the NIMA model has been deployed through the AI Network testnet (alpha version). The app calls a function that initiates the NIMA model on the AI Network. Every day at 6pm. KST, it evaluates uploaded photos, and rewards the users, considering the technical and aesthetic aspects of the photo. As such, many apps will be able to apply various artificial intelligence microservices from the AI Network, as if they were shopping, with minimal developers in the team.