The beginning of dt, four problems to be solved when introducing AI technology and successful solutions
2022/08/23
⏱ 4mins
-
Haley (Content Communication)
이 글은 업스테이지 이활석 CTO의 강연 자료<What is AI Pack?>을 바탕으로 작성되었습니다.
-
THOSE WHO ARE CONSIDERING ADOPTING AI TECHNOLOGY
THOSE WHO WANT TO KEEP STABLE AFTER INTRODUCING AI TECHNOLOGY
CURIOUS ABOUT HOW TO SUCCESSFULLY ADOPT AI TECHNOLOGY?
-
WE SUMMARIZED THE “REASONS WHY IT IS DIFFICULT TO INTRODUCE AI TECHNOLOGY IN PRACTICE”, WHICH UPSTAGE WAS ABLE TO MEET AND HEAR FROM OVER 100 COMPANIES THAT WANTED TO UTILIZE AI TECHNOLOGY. CHECK OUT THE INSIGHTS AND SOLUTIONS THAT UPSTAGE, AN AI SPECIALIST, HAVE DERIVED BASED ON THE TRIAL AND ERROR EXPERIENCED BY SEVERAL COMPANIES.
-
✔️The Importance of AI Transformation
✔️3 WAYS TO INTRODUCE AI
✔️THE DIFFICULTY OF INTRODUCING AI
✔️Choose the right problem definition and introduction method
✔️EFFECTIVE AI INTRODUCTION DIRECTION
Beyond DT to AI transformation
Recently, AI has emerged as a hot topic in several industries. This is because the importance of introducing AI technology used in this is emerging as more and more companies are pursuing DT (Digital Transformation) after the pandemic . In fact, global companies such as Google and Netflix are improving their service competitiveness through the powerful performance of deep learning, and their corporate value is also showing a steady rise.
According to the results of a KDI company survey conducted on 1,000 companies last year, nearly 77.8% of respondents answered that the introduction of AI technology helped their management and performance. On the other hand, to the question of whether AI technology is well applied in business, the rate of answering “no” was also high at 74%, suggesting confusion in the field of AI adoption.
WHAT FACTORS ARE NEEDED TO INTRODUCE AND APPLY AI TECHNOLOGY, SO MANY COMPANIES ARE HAVING DIFFICULTIES?
THREE WAYS TO ADOPT AI
THERE ARE THREE MAIN WAYS BUSINESSES CAN ADOPT AI.
1. Internalization
: GATHERING AI TECHNICIANS INSIDE THE COMPANY TO CREATE THE NECESSARY AI TECHNOLOGY ON THEIR OWN
2. System Integration
: COMMONLY CALLED SI, BUILDING AN AI SOLUTION INSIDE WITH THE TECHNICAL HELP OF A THIRD PARTY
3. External solutions
: When using a well-made cloud-based solution or software-type installation solution
USUALLY, COMPANIES DEFINE A PROBLEM ACCORDING TO THEIR OWN SITUATION AMONG THE ABOVE THREE AND CHOOSE AN INTRODUCTION METHOD. HOWEVER, DUE TO THE CHARACTERISTICS OF AI, EVEN IF A SPECIFIC INTRODUCTION METHOD IS SELECTED, IT WILL BE DIFFICULT TO SUSTAIN IT. WHY?
DIFFICULTY IN INTRODUCING AI
Upstage was able to meet more than 100 customers while developing AI solutions and hear vivid stories from a practical perspective on what difficulties companies have in adopting AI. There were four major difficulties that were mentioned the most.
Pain Point 1. AI model development and maintenance applied to business
The first difficulty is maintenance. AI model development largely consists of four steps : “Problem definition → Data collection → AI model development → Real server deployment” . Even if the AI model is completed through these steps, maintenance must be performed to maintain service quality.
However, it is difficult to determine where the problem occurred during the AI model development process, and even if the cause is known, trial and error is inevitable because there are not many cases that can be referred to to improve it . Therefore , advanced manpower, money, and time are absolutely necessary to perform these series of processes normally.
Pain Point 2. Development of AI technology applicable to services
The second is that it is difficult to develop a level of AI technology that can be applied to services. Companies often use externally provided AI solutions or open sources when developing AI models. In this case, it shows more than the basic performance, but since you are developing a service used by customers, you need to implement a performance that far exceeds this.
CREATING SUPERIOR PERFORMANCE BEYOND THE BASICS REQUIRES A LOT OF EFFORT AND KNOW-HOW, SO THERE ARE CASES WHERE IT IS DIFFICULT TO HAVE A PROFESSIONAL MANPOWER SYSTEM TO DEAL WITH AI.
Pain Point 3. AI maintenance is uncharted territory
Even after overcoming all of the aforementioned difficulties, the third difficulty is that in maintaining the deployed AI model, there is no identification of the cause or sharing of cases for the solution, and in the end, it is often solved by heading to the ground .
IN ORDER TO MAINTAIN AI TECHNOLOGY, IT IS NECESSARY TO PERFORM MODEL MONITORING, RETRAINING, AB TESTING, ETC. THIS PROCESS IS USUALLY DONE MANUALLY, SO IT COSTS A LOT OF MONEY, AND THE COMPANY GOES THROUGH A LOT OF TRIAL AND ERROR.
Pain Point 4. Rapid development of AI technology
The final reason AI adoption is difficult is the rapid advancement of technology . The number of AI papers published per day is about 800, so the pace of technological development is rapid. It is not easy to go through the process of looking at all these vast amounts of papers every time and implementing and verifying them with real data.
At times, you may be wondering why you should keep up with the latest AI technology, even after spending so much time and effort. The reason is This is because AI is an area where technological quantum jumps are frequent .
This is also a risk for companies that want to adopt AI technology. For example, let's say a company implements up to 60% of document handwriting recognition through OCR technology. With these figures, it was judged that it was difficult to apply the technology to the business, so it was excluded from the service, but after that, a new technology that can raise the recognition rate to 90% within one month has emerged and may meet the business requirements.
If we do not keep up with this rapidly evolving AI technology, service advancement will still remain unknown .
Choosing the right problem definition and introduction method
SO, DESPITE THESE CHALLENGES, HOW CAN ENTERPRISES SUCCESSFULLY ADOPT AI?
First, it is important to correctly define the problem to be solved with AI and select an appropriate introduction method . If you try to apply AI to solve ambiguous or non-task-specific problems, you are more likely to fail.
Therefore, it is recommended to look at each of the characteristics of the aforementioned three AI adoption methods ( internalization, system integration, external solution ) and choose the one that suits your situation . Even if it costs a lot of resources, if you want to have your own capabilities, it is better to choose an internalization method and use an external solution for specific tasks. Also, although it is a rare case in AI, if there is a problem that does not require maintenance or update after installation, using SI is also a way.
EFFECTIVE AI ADOPTION DIRECTION
As each introduction method has its own pros and cons, for the most effective introduction of AI technology, it is important to find an alternative that meets all of the following conditions or to select a partner that specializes in AI.
the right problem definition
CONTINUOUS MODEL UPDATES TO REFLECT THE LATEST AI TECHNOLOGIES
SECURING AI EXPERT TALENT
INFRASTRUCTURE GPU FACILITIES FOR LEARNING
If you have questions about how to introduce AI that is appropriate for your company's current situation, or if you have any questions you wanted to solve before applying OCR technology , find the clues in Upstage's webinar. Upstage has the know-how to develop Document AI that incorporates all the essential elements of AI technology adoption acquired through numerous cases.
ON AUGUST 31ST, WE WILL SHARE THE TRIAL AND ERROR COMMON TO THE AI ADOPTION PROCESS BY COMPANIES THAT UPSTAGE HAS BEEN ABLE TO GET TO KNOW THROUGH MEETING WITH MORE THAN 100 CUSTOMERS AND A CLEAR SOLUTION TO THIS IN THE “AI OCR ADOPTION SUCCESS EQUATION” TALK!
-
Under the mission of 'Making AI beneificial', Upstage is developing AI products such as OCR Pack and Recommendation Pack to create a better world with AI. Check the Upstage website for more information.