I would like to start writing a new blog from this time onwards. It is named ‘IISIA Technology Blog’.

Our Institute has entered the AI enterprise since last year (2022). While designing to expand this enterprise, I had one job interview and was asked this question.

‘I understand that you have entered AI sector and your business development is stable. Do you publish anything like technological blog?’

Now that you mentioned it, it is true. This may be natural in manufacturing industry. Since we are the independent thinktank researching closer to ‘social sciences’, we have developed columns showing some parts of our research analysis on this official blog, however, we have not implemented strategies on AI enterprise other than product’s PR.

However, it would be the primary problem for those who will be joining to our Institute that ‘what you can acquire there (take away)’, so I have again thought to explicitly write about it. As such, I have named this column series’ title as ‘Technology Blog’, and I will draw the context of AI development in our Institute while blending my (Takeo Harada) humble opinion as a global AI strategist with a light touch.

Though, I am not so-called a GEEK. As such, I am not intending to develop this discussion like an AI engineer often saying ‘C language is…’. It was the ‘Planner Course’ I was taking to get a master’s degree at Rikkyo University, which has established the first independent graduate school for artificial intelligence and science and I was the second-generation student there. In other words, its role is to stand between AI and real world to translate each other, and create the flow towards next generation. Therefore, AI enterprise (at this moment) in our Institute has been developed from this ‘Planner’s view’ and this ‘Technology blog’ will be written through those points of view, which I appreciate your understanding.

Such ‘social implementation of AI’ is in fact the real problem in Japan. The government sighs over ‘the lack of AI talent’. However, there are plenty of AI startups. On the other hand, it is the fact that they do not fly=grow their business at all. It is because most of the young founders established their companies without having a connection with real society, so they are not able to ‘sell’ though they have a good technology skill. So you might think that large companies with enough money to reap the rewards of the AI ​​that these AI startups produce could just acquire them through M&A, but this is also a very time-consuming process. And before you know it, the technology or product itself becomes outdated, and the startup self-destructs and disappears into thin air. What a shame.

In contrast, I would like to start this Technology Blog by highlighting that our AI enterprise at IISIA is completely different. The membership-based service “Harada Takeo Gemeinschaft” has a strong customer base of approximately 1,500 members (as of August 2023).The company releases AI products that address the needs/wants of these members, and as a result, it has already generated annual revenue of more than 150 million yen (from the AI ​​business alone). Needless to say that those ‘successful experience’ from the beginning is very important for AI businesses.

Having written this, I would normally like to move on to explaining the AI ​​product, Prometheus I/II, but since this is the first post, I will stop here for now. However, in order to prove that this is the ‘Technology Blog’, I would like to show the structure of algorithm (deep learning) used in Prometheus from the result of running python code.

As indicated above, Prometheus is a combination of convolutional neural network (CNN) and LSTM. You may think ‘Why using such an old-fashioned algorithm in the world of ChatGPT?’. Of course, there is a reason for this. A typical technical blog would carry out numerical verification by listing a number of metrics one after another, and then discuss solely the pros and cons of each. But at that moment, AI startups fall into a trap. In short, they are stuck on the worst possible path: creating an AI product that doesn’t sell.

Why is that? …In the next article (part 2), I would like to write about the above points, touching on the transformer that is the origin of ChatGPT, and writing about some of the research and development on the issues surrounding “overlearning”, “generalisation”, and the peculiar quirks of “time series analysis”.

August 6, 2023, Marunouchi, Tokyo

CEO/Global AI Strategist

Written by Takeo Harada