Vice Managing Director of Rikkei AI: To compete in the AI game, companies must own the huge data resources

AI is currently being used to accelerate progress in vital fields such as health, agriculture, finance, and transport. However, good data sources are essential for AI to further develop evenly across areas and regions. Data availability is the key for training AI systems, with products and services rapidly moving from pattern recognition and insight generation to more sophisticated forecasting techniques and, thus, better decision making.

 

Businesses need to own a huge data warehouse to join this AI competition to capture and lead the latest technology trend. Some Vietnam technology pillars such as FPT, Viettel, VinAI, and VNG have built their huge data warehouses to approach future trends. As a part of Vietnam technology leaders, Rikkeisoft is not out of the AI game with a vast audio data warehouse for the “Speech to text”  product. Besides, for readers to have a multi-view about the role of data in the development of AI; Rikkeisoft had a small interview with Mr. Nguyen Minh Tan (Vice Managing Director of Rikkei AI-  the “cradle” of almost all AI products and projects of Rikkeisoft), let’s stay tuned!

AI stands for Artificial Intelligence, which means artificial intelligence. People are always trying to think and develop algorithms to make computers smarter. Artificial intelligence was conceived in the 1950s. However, due to the limit of computing power, computers at that time could only do pre-defined coding tasks. The current era is the age of artificial intelligence. Computer processing capability continues to grow, and computer costs fall, allowing more people to access advanced technology, resulting in AI becoming more widely recognized and used in practice.

Many opinions argued data is the new source of black gold of the 21st century, not oil. An AI product is a synthesis of many different materials, and data is one of the essential components. According to statistics, up to 80% of AI product development time is spent on data-related processing. Training AI is similar to educating a child. AI can learn successfully if the correct data is used to train it. In contrast, if we teach AI with incorrect data, it will learn incorrectly. When heterogeneous data is used at an inaccurate moment, the AI becomes confused. The more data appropriately utilized for training, the smarter the AI becomes and the more precisely it can recognize.

There is no doubt that large technology companies such as Google and Microsoft and self-driving vehicle companies like Tesla, Hyundai, and Toyota have heavily invested in data to remain ahead of the competition. For example, hundreds of thousands of hours of audio data were used to train Google’s speech recognition engine. Likewise, Tesla uses millions of real photographs to teach its self-driving cars.

Companies that specialize in data labeling appear to be following the AI trend as well. In addition, data labeling technologies that are quicker and more accurate are also being developed. As a result, many of these businesses have gone on to become unicorn start-ups.

As we all know, Rikkeisoft’s Speech To Text with the core technology is a voice recognition solution deployed to the National Assembly and several ministries, provinces, agencies, and other government entities. Although Speech Recognition technology has a wide range of applications, only a few firms in Vietnam hold it. Some big names such as Viettel, FPT, VNG, and VinGroup are all significant players. So, why do so few firms take the risk of participating in this game? Because, in addition to the resources of buildings, equipment infrastructure, and technical staff capable of training AI, a large amount of speech sound data is required to create this technology. It costs a lot of money to collect and categorize this audio data. Not all firms are willing to accept financial risks, especially when the feasibility of mastering technology and product output is uncertain. We may claim that Rikkeisoft is fortunate in having the vision and a great technological team since it possesses one of the most significant technologies.

The difficulty of accurate data collecting and quality control and the expense of labeling data are the most challenging aspects of producing data wherever in the globe. Vinfast, for example, is a major artificial intelligence application company in Vietnam. Vinfast is Vietnam’s sole autonomous vehicle research and development center. Vinfast has invested a lot of time and money into putting together a fleet of cars that can go around the streets and snap photographs in a range of time and weather situations.

Rikkeisoft has been struggled to discover acoustic data sources that match reality, variety in content, geography, age, and gender when developing the Speech-To-Text product. When each person’s capacity to hear and comprehend is varied, data labeling might be challenging. People in this region, in particular, find it difficult to hear and adequately capture the sounds spoken by others. Rikkeisoft has studied and developed Rikano, a professional data labeling tool, to meet the need for internal data labeling and new service development. Rikano was created by using and referencing the features of many other labeling tools around the world. It can label the most common types of data, such as images, sounds, and text while making it simple for users to manage progress, quality of work, productivity, and the working history of each project participant. As a result, Rikano plays a vital role in Rikkeisoft’s data labeling service’s development strategy.

Businesses are becoming more interested in AI as they better understand the benefits it may provide. As a result, all data connected to the business’s activities will be restored to assist AI construction in optimizing operations.

Another AI trend is creating algorithms to prevent data labeling, a labor-intensive task that is likely to be controlled by the emotional element or the human factor. However, this is still a future issue.