« AI は嘘をつく: 創造性支援と chatGPT | トップページ | 「serendipity」と「怪我の功名」 »

2023年7月10日 (月)

AI Lies: Creativity Support and chatGPT

 I asked chatGPT the following question.

"Do you prefer your life to be 'thick and short' or 'thin and long'?"

The answer is below.

The "thick and short" approach means that life should be focused on vibrant adventures and passionate experiences. In this approach, it is important to experience more in a limited amount of time and to pursue personal growth and fulfillment. The goal is to pursue life's richest experiences and enjoy the most intense moments.

On the other hand, the "lean and long" approach means focusing on a sustainable state of happiness and equanimity in one's life. In this approach, the focus is on maintaining health and well-being at a leisurely pace. It emphasizes the pursuit of sustainable happiness, such as achieving long-term goals, building stable relationships, and pursuing a stable career.

Which is better suited to your own values and goals. Some people may feel they want to live a daring life, seeking adventure and challenge. Others, on the other hand, may value stability and security and want to lead a leisurely life.

The key is to choose the life you want for yourself and move toward it. It is important to think carefully about what you value and what kind of life you want to lead, and find the approach that is right for you.

 There are three things that I think about when I look at this "Word".

 First, both in terms of content and volume (540 words), it is a very well written essay. In fact, if it were a recommendation essay for a high school entrance examination, it would surely receive a perfect score. I think it is also of a quality that would earn a passing score in a university AO exam essay.

 Secondly, although current AI technology has no human-like "reading comprehension" skills in the sense of understanding the context, I was surprised to see how far it can go in writing essays with the help of statistics, estimation theory, and big data-based reinforcement learning technology.

 Thirdly, it is said that the composition of the chatGPT sometimes contains errors, non-facts, and counter-facts, so it cannot be trusted in its entirety, but such errors may be useful as creativity. Whenever something new is created, it is always inextricably linked to mistakes, failures, and errors. The machines that have assisted humans up to now have been machines that execute and realize what has been calculated and planned without missing a beat. In this way, they have amplified the welfare of humankind. However, as a matter of course, machines do only what they are told to do. It is accurate, but creativity is not to be expected, and that is fine. However, current AI, such as chatGPT, is excellent and accurate to a certain degree, but occasionally makes mistakes that we do not expect. This is why I believe that AI is effective only as a machine or engine to support human creativity.

※ AI has an advantage in "writing" rather than "reading comprehension. This point was made half a century ago by Satoshi Watanabe (1910-1993), one of the most creative scientists in modern Japanese scientific history and one of the important early pioneers of computer science (AI). However, it was inevitable that even the genius Watanabe could not have envisioned 50 years ago that the exponential improvement in computer processing power would allow us to solve the problems of machine "reading comprehension" and "writing" through statistical processing of big data (of words) in an engineering, rather than mathematical, pseudo-solution. I cannot help but feel sorry for him.

   However, the structural analysis method has many weaknesses. There is generally no exception-free and successful way to determine whether a given figure is grammatical or ungrammatical. Chomsky's grammar is primarily a set of rules for producing grammatical sentences, and conversely it is weak in the task of determining whether a sentence is grammatical or not.
 In the case of actual writing, the distinction between grammatical and ungrammatical sentences is crucial, and it is determined regardless of how the grammar is written. In the case of figures, the distinction is determined by a self-created grammar, which is very different from the grammaticality of ordinary sentences, which is recognized independently of grammar.
 There are no satisfactory answers to the fundamental questions of what constitutes an element, how to discover a grammar, and so on. And the most difficult element identification still relies on ordinary pattern recognition. Therefore, I do not think that this method has much future beyond its intellectual interest.
Satoshi Watanabe, "Recognition and Patterns," Iwanami Shinsho 1978, pp. 154-5, Chapter 4: Pattern Recognition by Computer, Section 6: Structural Analysis Method

 My main impression from this trial is that if AI can do far more trial-and-error in a shorter time than individual human beings can do, it may become a powerful assistant, secretary, and collaborator to support creativity at the individual level.


« AI は嘘をつく: 創造性支援と chatGPT | トップページ | 「serendipity」と「怪我の功名」 »

科学哲学/科学史(philosophy of science)」カテゴリの記事



Watanabe, Satoshi (渡辺慧)」カテゴリの記事


AI ( artificial intelligence)」カテゴリの記事




« AI は嘘をつく: 創造性支援と chatGPT | トップページ | 「serendipity」と「怪我の功名」 »