FPT Chief Scientist: ‘We are in the 2nd AI wave’

199

2019 will be the year of “popularization of AI” because it has exploded all over the world and gradually spread in all areas. So how do we live with “artificial intelligence”? Let’s listen to the sharing of FPT FPT Chief Scientist – Dr. Dang Hoang Vu at Smart Tech for Smart Living Conference.

2019 – the year of “popularizing AI”

AI is changing the face of the world with many applications in all areas. With nearly 10 years of experience in AI research and development, Dr. Dang Hoang Vu – Director of FPT Science Group said that 2019 will be the year of “popularizing AI“.

He emphasized that it is not until currently, that AI appears. The world chess monument was once defeated by an old generation of AI. Do not think AI is the method or technology but it is the need to replace human intelligence with machines.

Machine learning is the foundation technology of AI, which helps the machine to learn the observational data in real life. AI can be divided into two categories, one is to replace some human functions such as video processing, speech recognition, voice and to understand language through text or speech

Dr. Dang Hoang Vu – FPT Science Director shared about the topic of AI at the event.

AI will apply well in business, technology, society… Specifically, as predictive industry analysis is very popular in videos, songs … For example, in the power plant system, they can guess when the problem occurs when there are high needs for electricity to adjust machinery. Next is to reduce costs, increase production and business performance. Currently, mathematics combines computers to form an AI platform.

Automation technology has typical examples such as self-driving cars, or simpler. For example, in the past, we went to a store or company, we must make a timekeeping record. Now just scan the camera to recognize the face and know who you are, each person saves 10 seconds, the company saves quite a lot. Or when going to the store, just scan ID card to collect data into the storage center. Instead of spending a lot of time and labor to record customer information, AI can support replacement and cost savings.

“Its nature is also math but was upgraded and it can learn input data, applying to many different fields,”, said Mr. Dang Hoang Vu.

Finally, tools help the ultimate goal of making decisions, which is faster and more accurate than people. That is the most difficult problem that AI application industries are aiming for. For example, propose how much money to lend, what to sell, to whom… help eliminate human emotional factors, help reduce risks and promote business growth.

All tools serve a problem: faster, more accurate and more effective than people. Currently, AI only supports decision making of whether they should lend this money, sell each of these to customers… In the future when the accuracy increases, AI will remove people from that decision, the rotation speed of business will increase.

Where is AI in the world?

Every new technology follows a loop: when it is opened, people all show many expectations, but most of them fall, only when they can get out, can they grow. Speaking of AI, each person will have different judgments about AI development.

 “AI has a lot of different technologies, located on different roads. If there are conflicting opinions about the development path of AI, then there is the only partial description, not a comprehensive one”, he admitted.

The basic application process of AI in the enterprise: first is to take data into AI (books, video back to the store…), the second is to process (poor data type), select the model (art math, methodology), testing and putting into the application, upgrading them. About the process:

  • First: input data.
  • Second: to process data, remove valuable information.
  • Third: choose the equivalent algorithm model, method and train it.
  • Fourth: experimental phase.
  • Fifth: we put into application and keep improving.

Mr. Vu also stated that the second step is quite costly because it is necessary to handle huge amounts of data for AI to learn. Examples are as follows:

When an enterprise uses AI to serve customers better, increase sales, they use the AI model to optimize customer satisfaction index. They have a scale from 1 to 5 when applying AI, the index increases markedly but the problem is ineffective. Why?

Although the index increased, the time to use the product did not increase. Satisfaction index has little to do with business goals. That said, 1, 2 is important, what kind of collection, how to screen. AI only works according to the data you give it, but it does not know that the time using the new service determines business efficiency.

“For such a process, it is a typical process for AI technology currently in use. I have a list of current AI waves, there are many views about dividing the AI waves. In my opinion, there are three waves that we now have in which, we are in the second wave of statistics-based learning. That is, collecting data, using statistical models and learning rules from the majority”, Mr. Vu commented.

Industry experts consider the AI as smart, and if we want to push it smarter, we must help it reason better.

Behind AI is deep machine learning

Mr. Vu shared that the machine learning technology behind 99% AI is deep machine learning technology. What is the basic difference? For traditional machine learning, it is necessary for people to convert input information into understandable forms before teaching them. But deep learning machines can learn by themselves from the raw data input.

Data screening step accounts for 80% of an AI project’s time effort. If half of the effort is reduced, the money in this step will be saved a lot”, Mr. Vu said.

Currently, deep learning comes to the throne, is widely applied to vision, traffic processing, language. He gave a concrete example of a machine that learns from perfect strokes, shapes, and pictures… However, no matter how successful, mass-based statistical learning techniques still have limitations: based on the majority, unable to distinguish the abnormality, it only believes in the majority and the data it sees. This led to the case of a robot capable of conversing with humans, but after a while, it had to be turned off because it often used nonsense language to chat without itself being right and wrong, it only listens to the person who taught it, so the provider must remove it.

“Why do I name the speech as the process of popularizing AI? Previously, to write an AI program requires university professors but now only need to use tools available around the world. Of course, if you want to create a new technique more accurate than your opponent, that is, different. But if only the general level is needed, the barrier is very low, then AI has never been as easily accessible as it is now. Fortunately for us, the decision process still needs human intervention”, Mr. Vu concluded.

Source: VnExpress

Related posts: