In the midst of the summer holiday, it feels that the flow of ‘the fever of generative AI’ has started to cool down. Suppose we trace this AI furore back to its origins. In that case, we can see that it was only possible thanks to the development of the Perceptron, a product of the efforts of pioneers who were quietly but steadily conducting research in Japan at a time when no one paid any attention to artificial intelligence. It cannot be said with certainty that it was imported. However, the typical mentality of Japanese businesspeople (after the collapse of the Heisei bubble), means that if you say, ‘Something amazing has arrived from overseas!’ it is easy to get funding and you don’t have to take responsibility if something goes wrong, has been triggered, and the movement itself has become unstoppable. This ‘Technology Blog’ is written by the author educated to be a pro of AI social implementation as a ‘planner’ who plans from the social implementation of ‘what should AI be used for?’ and its realization, and not as an ‘engineer’ who creates codes only to improve numbers on metrics without aiming to its social implementation. This ‘Technology Blog’ is written by the author educated to be a pro of AI social implementation as a ‘planner’ who plans from the social implementation of ‘what should AI be used for?’ and its realisation, and not as an ‘engineer’ who creates codes only to improve numbers on metrics without aiming to its social implementation. From this point of view, ‘the fever of generative AI’ above is not such ridiculous. Why?

‘The key to text generation AI is a large-scale language model (LLM). That’s why we must develop an LLM, especially for Japanese, on a large scale.’ As usual, people who get excited about ‘Hinomaru Japan’ are creating Japanese LLMs using this kind of logic and announcing them one after another. I reckon those are good quality since these LLMs are specialised in Japanese which OpenAI is not good at (I guess). However, the very existence of such a trend is, to an AI planner whose role is to connect AI and the real world, completely unacceptable.

There is only one fundamental principle of business. That is,’ creating added values by using materials for free of charge, and sell these with quite a bit of price.‘ Although it is a simple fact, the textbooks of MBA or even the books by well-known consultants do not write about it. Those books usually omit the first phrase in the statement ‘using materials for free of charge’ and start to discuss from the next phrase ‘creating added values’. And then, the part about Marketing Sales that ‘sell with quite a bit price’ will be talked about. However. As mentioned earlier, the fundamental principle of business is ‘using materials free of charge’. If this is not followed conscientiously, in the end, the products or services which is the outcome, need to keep pricing high, and that is why the business goal to make money cannot be accomplished.

We need to be aware that this can be said the same thing in ‘generative AI’. Regardless of LLM-generating texts or the model-generating images, the base mechanism is set up in IT tools that have been used for more than 20 years. We operate the tools there (for example webmail systems (such as Gmail) or social media) on your computer. Or OS often gets restarted, and every time you notice something upgraded? but you leave it alone. The trial and error that has gone into using these over the last 20 years have ultimately been essential to perfecting the algorithms that can generate a fairly accurate answer when you give them a specific prompt about text or an image. But I want you to think about it, have you got any pecuniary rewards from joining ‘the greatest project in human history’ without knowing it. … the answer is NO (!). Billions of humans have worked for free to realise this ‘greatest project in human history’.

In the first place, Japan was the top country in the world of computing until the 1980s. Although this is a surprising fact, it is true. The reason why this status was lost was because the hard reason was that acceleration of computation speed was not enough, there was not enough computation data that should be there and also it needed to be ‘input manually’. (This is the failure in the second AI boom.) Even though the Japanese government spent a substantial amount of money on AI development, it failed as a result. However, even if they had succeeded on small technical things, it would have greatly questioned whether this could be a ‘business’. It is because they invest enormous amounts of money to collect and input data that is essential for AI. In other words, as this is ‘not free of charge’, it does not succeed as a business.

The trending ‘generative AI’ is ingenious at this point. At first, how they collected data in the last 20 years was not touched at all due to the ‘business secret’. LLM or GPT4 used around ChatGPT was published as a technical report to understand its summaries, but the data or learning methods used there were not open. No wonder, the origin of the values ChatGPT provides, which we pay to use, is the enormous accumulated data and its fixed information like stated above, and they do not want to reveal the fact that it is the outcome of our unpaid work. It is a business model following the basics of business, ‘having business using things for free’ and it cannot be avoided to say it is ingenious. On the contrary, Japanese companies shell out a lot of money like ‘LLM is the next’ to belatedly develop LLMs. Looking at it as an AI planner, this is truly the height of stupidity and it is obvious from the beginning that it cannot reach in business.

‘So, what about Prometheus, the AI ​​analysis tool for financial index volatility developed by your research institute? Does it stand on the basis of the business model mentioned here?’

I can almost hear such voices. And that is true. I, as a person in charge of coding, want you as users to evaluate this point as ‘high’. As previously mentioned, Prometheus extracts the closing prices of various financial indicators on a daily basis, recognises the difference as volatility, performs preprocessing, and then executes calculations. Therefore, saying this from this context, it is the biggest problem in a business model how we can keep obtaining closing prices of various financial indices precisely ‘for free of charge’. Moreover, it is better to take data on closing prices of various financial idiocies from websites without spending time and effort on codes. As a means of so-called scraping from a website, you can use beautiful soup, but in any case, the site being scraped needs to be sufficiently prepared for it. (In other words, it is preferable that the coding is not complicated when viewed in HTML.) Furthermore, we would like the website that is the target of this scraping to contain a comprehensive list of various financial indicators.

For such reasons above, Prometheus uses a Polish website called Stooq.com. I will omit the codes (as it is a company’s secret), because this, enables me to do the last computing and plot in one go with one click. In addition, using Stooq.com does not cost. We add value by coding in Python, and ultimately, by illustrating the data and providing it on a daily basis (on a business day basis), we are able to offer it at a reasonable price. By that meaning, Prometheus our Institute offers follows the business basics of basics. We released it in July, last year (2022) and through releasing the second edition, it has become about 160 million yen worth of business as a yearly turnover. Considering the current situation in Japan, where the majority of AI startups look good but don’t even reach 100 million yen in annual sales, I consider this to be a success.

Finally, a summary. AI business is different from ‘AI research’ and you need to look back at the business basics. In this case, the key ultimately lies in ‘how cheaply you can obtain materials (preferably for free)’ and ‘how highly you can sell them.’ It is an AI planner’s work to plan these things, not an AI engineer. Building on the immediate success of Prometheus I and II, IISIA is now making a major shift towards the next AI business model.

18^th August, 2023, Kyoto

Written by CEO/Global AI Strategist Takeo Harada