Intelligent automation has grown exponentially over the last decade, with concepts that previously only existed in science fiction now becoming a reality. Automation software has been introduced in both the workplace and your personal lives, taking on rudimentary tasks and making our lives easier.
Artificial intelligence was once a phenomenon restricted to sci-fi movies, but technology has finally caught up with imagination. Now AI has become reality and amazingly, most people encounter some form of artificial intelligence in their everyday lives. Starting from health, mobilization, interaction with other humans, or robot interaction with humans.
Artificial Intelligence as a science of technology is divided into seven branches. The branches are Machine Learning (ML), Natural Language Processing (NLP), Expert System, Vision, Speech, Planning, and Robotic. But in this article we will discuss how the development of AI and Machine Learning.
There is a subtle difference between AI and Machine Learning. AI is a branch of computer science attempting to build machines capable of intelligent behavior, while machine learning can be defined as the science of getting computers to act without being explicitly programmed. In other words, AI researchers build the smart machines, while machine learning experts would make them truly intelligent. Both have an important role in the development of technology. In fact, not a few companies in various fields have now begun to implement it.
Lintang Sutawika Co-Founder of Konvergen.AI, talked about the development of the AI, especially the Machine Learning sub-field which has experienced very rapid research developments in recent years. In just 5 years, the number of submissions of scientific papers to major conferences such as NeurIPS has increased 4-fold.
In addition, a 2012 publication on machine learning methods that initiated the academic community's interest in machine learning approaches that are commonly used today has been cited more than 57 thousand times.
On the applicative side, there are many corporations that support their data teams with AI systems such as anomaly detection or recommendation and forecasting. Many machine learning-based startups are emerging and not a few are getting large funding for product development. This shows a lot of interest in implementing AI solutions in industrial and everyday problems.
From the human capital side, looking for talent engineers who understand the implementation process is quite difficult, but on the other hand the tools to apply machine learning in software products are already numerous and easier to use so that in the end startup people can build their own chatbots, or processes other machine learning processes that can be accessed in the form of APIs so that they are easily integrated into the application.
At the same time, the development of AI and Machine learning has reached a level where its limitations in solving problems are beginning to be seen. two AI experts Gary Marcus and Joshua Bengio discuss the best directions in further research in the AI ??field. In short, a field that used to be more often foar a material of Hollywood movies is now moving from the realm of research and is becoming increasingly commonly used in day-to-day operations but AI in the real sense has not yet been achieved.
Lintang considers that AI and machine learning are currently useful to reduce repetitive work but have uncertainty. As an illustration, robotic systems that can be found in the manufacturing industry generally do repetitive and definite work. To paint a car, a robotic system will be programmed to move the paint sprayer explicitly at x, y, and z coordinates. Every movement has been programmed.
"This is different from the development of AI and machine learning which not only can do repetitive work but also has uncertainty. For examples such as how Konvergen.AI products can read a receipt. Actually copying information such as nominal, date, and time from 1000 receipts is repetitive and tedious, but each receipt can have a different structure depending on the vendor that issued the receipt. An AI-based solution can study patterns and see consistency between text on the receipt, such as when a number is nominal (There is a prefix "IDR" or suffix "IDR" and is located near the end / under receipt) and be tolerant of variations such as fonts and layout, "he said.
Another example mentioned by Lintang is more like the insurance claim process. Each customer may have a different language style in reporting the same claim, or it can also be the same claim regarding damage to the vehicle where the car is dented. The thing that was reported was actually repetitive, an officer from the insurance party had to check the photo of the car bender submitted by the customer, but the dents themselves could be different such as in front of the car or behind the car. This repetitive but uncertain nature can be an opportunity for AI implementation.
Generally, data is the main problem. At present, the majority of machine learning applications in the industrial realm are supervised learning. In short, supervised learning is a learning process that connects between an input x with an output y both of which have been provided. This process is very data-intensive or requires a lot of data.
In addition, stakeholders of the solution to be developed need to understand the limitations of this AI system so that they can have more realistic expectations. In building an AI-based digital solution and the right Machine Learning must also identify the return-of-investment.
The first step is to collect data, generally there is a labeling process in this step. Second, system development which consists of making systems in general, in terms of AI there is generally a training process. Third, determine the appropriate validation metrics, which can be accuracy scores or other qualitative metrics according to business processes such as improving the performance of a team that uses the AI ??system. Fourth, deploy the temporary version as a stage of introduction to flow users, then fifth if possible, this AI system needs to be updated.
Lintang said, the challenge in developing MVP was expectations from users that could not be ignored, "At present there is a lot of hype that accompanies AI-based products but sometimes it is over-promoting and sometimes not in line with user expectations that tend to be less realistic with current technological developments," he said.
He gave an example, a predictive model for making recommendations could use sophisticated algorithms but in terms of scalability is a challenge so that the response time could be 1-2 seconds. This is clearly longer than the data query API whose response timeline can be in milliseconds of order. Trade-off speed, accuracy, are things that need to be educated so that users can make decisions according to their business processes.
To test MVP in order to compete in the market the first step is to have a validation metric. This can be the case or in terms of data-entry, the amount of time that is cut in doing repetitive work.
Konvergen.AI is a startup that focuses on developing technology based on artificial intelligence solutions to reduce manual data entry processes. Konvergen.AI is currently working with various financial companies as well as to reduce the data entry process.
Lintang assesses that a corporation generally has operations and businesses that are well-defined, targeted, and thus may experience difficulties to justify doing research in new or prospective fields or technologies. On the other hand, a startup is a business that is just starting and is generally required to offer effective solutions to a pain-point. Thus, a corporation can use services or help the startup's business process if it is in line with the corporate business process.
"For example, startups chosen to join the TINC program have a business alignment that is aligned with Telkomsel. Maybe Telkomsel as a large corporation will find it difficult to enter the business lines of these startups, but Telkomsel and the participants of the TINC program can both deliver value through cooperation that is mutually beneficial to both parties, "he explained.
Collaborating with a bona fide corporation such as Telkomsel certainly brings a good reputation. In impact, working with corporations can help introduce solutions offered by a startup.