Column by the President of Hitachi Research Institute, Mizoguchi
The Industrial Revolution, which began in England in the late 18th century, radically changed the world. As the number of farmers decreased and the number of factory workers increased, the influx of people into cities accelerated, and the cities expanded rapidly. Developments in science and technology and changes in the economic structure drove the development of industries such as iron and steel, machinery, shipbuilding, and military equipment. Productivity improved dramatically, and it became possible to mass-produce a variety of products and make them widely available to consumers. To deliver those products far and wide quickly, means of transportation were developed, and logistics networks utilizing motor vehicles, railways, and marine transport were constructed. Profits of investors increased, money became more fluid, productivity further improved, capitalists and workers became separated, and “capitalism” was established. Competition between nations driven by advances in technology and industry intensified, and technological and economic strength determined the rise and fall of nations. Not only economies but also societies and their politics were transformed by the Industrial Revolution—which, incidentally, began with the technical innovation of the cotton loom.
The ongoing generative-artificial intelligence (AI) revolution, in conjunction with the spread of digital infrastructure, will also bring about major social changes comparable to those brought about by the Industrial Revolution of the 18th century. Much intellectual work can be replaced by AI, value in cyberspace will continue to be diluted, new industries will be born, and some industries will decline. AI will also affect the productivity of companies, force changes in military strategy, encourage the accumulation of capital in one’s own region or industry, change the rules of the game of international competition, and transform social hierarchies. As a result, the world will rapidly take on a different appearance.
Why could generative AI have such an impact? Is it because AI functions through a mechanism based on words? Only with words have humans been able to think, and through that thinking we have built civilizations. Generative AI shares the source of its power of value creation with humans.
If the basic mechanism of generative AI is simply to process subsequent words by statistical prediction, I don’t think it will be able to accomplish a great deal. In fact, because of the way it works, generative AI can plausibly spout nonsense and even encourage criminal behavior in its responses. Nevertheless, the usefulness of generative AI is rapidly increasing because it incorporates improvement steps into large language models (LLMs). LLMs can suddenly solve problems that are difficult to solve when the number of parameters is in the hundreds of millions as the number of parameters increases to hundreds of billions, trillions, and more, and emergent abilities are manifested. As an analogy, when we are learning a foreign language and are having trouble improving, at some point, we suddenly become able to speak it. Similarly, when the amount of information that generative AI learns exceeds a certain threshold, it will improve dramatically. And that improvement is repeated.
Generative AI will become infinitely superior through repeated improvement. In time, therefore, it will surpass human intelligence and achieve the so-called “singularity.” However, if the fundamental principle of generative AI is statistical processing of words, “hallucination” cannot be eradicated.
Wittgenstein held that the meaning of a word is determined only in relation to the language as a whole. According to “The Philosophy of Language Begins” by Shigeki Noya, the reality of language usage is that the whole language is not made up of parts; instead, the parts have meaning in their relationship with the whole, the parts and the whole are closely connected, and the relationship between the parts and the whole is cyclical. LLMs do not assume such understanding of the context of the whole language. Generative AI can reduce the degree of error through reinforcement learning and steadily evolve, but it does not use the words or think in relation to the words as a whole, as humans do. In other words, generative AI has not mastered “symbolic grounding.” However, if an AI system responds in an infinitely human-like manner, feels infinitely human-like, and gives answers that seem to be better than human answers in a quicker response time, it is infinitely closer to the truth to say that it is more than human.
The problem with generative AI, however, is that while its risk may become infinitesimally small, it will never be reduced to zero. As AI becomes more superior through repeated improvement, the tail risk of any eventuality becomes correspondingly huge, and AI will become a threat to humanity. However, a way to counter that treat may also be found. In fact, James P. Hogan’s “The Two Faces of Tomorrow,” published in 1979, already suggests a way based on words and awareness. As a master of hard science fiction, he foresaw the risks of AI and proposed a way to counter them 45 years ago. However, I can’t reveal what it is—doing so would be a spoiler.
Search
Global