What happened?
The introduction of ChatGPT created a new generation of AI. McKinsey has released a further study on generative AI. More than 40% of the respondents indicated that their companies continue to increase their investment in AI due to the emergence of generative AI innovations. 79% of the respondents indicated that they have been exposed to generative AI, and 22% of the respondents have been applying generative AI in their work.
- The industry sectors with the highest distribution of respondents were technology, media, and telecommunications, followed by financial services.
- According to the report, the areas where generative AI tools are most used are Marketing and Sales, Product and Service Development, and Service Operations.
Machine Learning Transforms Efficiency and Accuracy
Generative AI uses machine-learning models to mimic the patterns and relationships of human-created content and create new content on its own, whether it's text, images, data, or data analytics. The most common method for training generative AI models is supervised learning, which uses human-created content and corresponding labels. According to Google's explanation of machine learning in the context of its development of AI technology, the characteristics of machine learning are:
- Using algorithms to analyze large amounts of data, learn from in-depth information analysis, and then make informed decisions. Machine learning technology algorithms are trained to improve their performance over time as they are exposed to more data.
- A machine learning model is the output of data, or what a program learns from executing an algorithm on training data, and the more data used in the process, the more accurate the model becomes.
- AI is the broader concept of enabling a machine or system to perceive, infer, act, or adapt as a human would.
- Machine Learning is an AI application that allows a machine to extract knowledge from data and learn automatically.
Trends and Challenges of Generative AI
Generative AI can be trained with different commands to process large amounts of information and generate the in-depth analysis information and answers that users need through text, images, and simple formats. Former Google Taiwan GM Li-Feng Chien presented his observations of generative AI at the 2023 Taiwan Artificial Intelligence Conference, including the following:
- White-collar workers between the ages of 35 and 44 will have the opportunity to increase the productivity of generative AI and even more commercial applications if the use of ChatGPT increases and its stickiness improves.
- As Google and Siri begin to connect large-scale language models, more and more human-machine interfaces will emerge.
- As AI becomes more and more popular, machine learning is beginning to become commonplace.
- Goldman Sachs once predicted that AI could increase the annual growth of labor productivity in the U.S. by 1%. From this point of view, although AI may impact job opportunities, it can fill the labor shortage gap in Taiwan, a country with few children.
However, there are challenges to overcome in developing generative AI:
- Training AI requires massive computational resources, which requires enormous computing power, and the cost of cloud computing may not yet be able to support such development. Current AI modeling capabilities are limited by contextual learning and probabilistic prediction constraints and are still quite limited. For example, the accuracy of answers needs to be higher, and the content needs to be real-time, biased, and weakly connected to the real world.
- The large-scale language model is mainly in English, which may lead to the monopolization of English content. It is more challenging to achieve customized mechanisms, which will result in the same answers and can not be different from person to person. The answers obtained from other countries and different languages will be very similar.
- AI extracts much of its computational data from open Internet data, but it will grow slowly. The open data for AI training is almost used up, which will be a major bottleneck.
OneAI, a Japanese start-up company, is eyeing the potential of AI technology and has developed a conversational marketing service centered on Taiwan in 2018. Using the language training models of Google and ChatGPT, they use CVR (conversion rate) as a teacher to train AI. They have developed conversational marketing content that automatically generates a "conversion rate that can be increased" and has obtained patents in Taiwan and Japan. In an interview with OneAI, Asahi Times takes readers on a journey to explore further how generative AI can be applied to marketing and build a real-world business model.
