We help you to adopt AI
You can start developing AI without being fully prepared.
Datumo works together to analyze existing datasets and review business plans.
We analyze the existing dataset held by the company using the Feature Space Tool. Anyone can easily understand the characteristics of the dataset through a 2-dimensional distribution graph.
We conduct planning for collecting pre-processed data based on deep learning, which allows for early cost savings in data construction.
Introduces AI services that can be developed with data sets owned by companies. Estimate the cost of curating data early to review the feasibility of the project.
We propose an appropriate AI model that can implement the planned service. We support quick business decision-making for companies with simple prototype modeling.
Many places are actively adopting artificial intelligence so that it can reduce costs and automate tasks. However, there are many reasons that they can't implement it even though they know its importance.
We invited two CEOs of Datumo, who have worked on 312 data projects and accumulated approximately 130 million pieces of data since its establishment in 2018. Now, let's talk about how artificial intelligence is being used in various industries.
Julien: AI is already a big part of our daily lives. Off the top of my head, I would say Facebook ad exposure algorithms and YouTube video recommendation algorithms.
David: Apart from those, TTS(text-to-speech), which converts text to a voice reading the text, is appearing more and more. Recently AI Sonny*, which was a TTS service using Sonny’s voice. Datumo’s data were used for the project.
*Sonny is a nickname for Son Heung-min, a Korean football player who plays for Tottenham Hotspur, Premier League.
David: There are unusual examples because our clients are doing so many kinds of AI. Here are some unusual examples. Generally, if you look at a power pole, there are various parts. In order to maintain those parts, a person goes up and checks whether the actual parts are now rusty and whether there is a problem and then fixes them. To automate them, there was a company that made artificial intelligence that can recognize power pole components.
In the case of livestock farming, whether pigs or cows are sick or healthy is managed visually or by watching CCTV. There were some companies that managed it through AI.
On the national defense, there was also a case where various information from satellite images was quickly captured so that it is used through artificial intelligence. I know that technology was applied to help even during the recent war in Ukraine, which was heartbreaking.
I want people to think AI in simple terms. AI is just a tool. “There's something I'd like to automate.” then, I think it would be nice if you could think that AI can be applied.
Ryan: I have a question Identifying a power pole’s components is an expert's know-how but is it OK to know that AI can also be changed with technology that can identify?
David: Yes, that's right. In the end, when making a judgment based on what kind of perception they have, it is because they have recognized a specific pattern. What artificial intelligence is good at is learning by looking at various data and extracting patterns. Therefore, if a person makes a judgment based on some basis or pattern, it suggests that artificial intelligence can replace it.
Ryan: Wait, I think the stock market has a pattern. Can AI learn the pattern?
David: If there were patterns in the stock market, I would not be crying every night..
David: I can tell how frustrated you must felt from “not being able to communicate”. First, we need to look at AI as the black box mentioned above. When setting up an organization and work, if we look at the top to see what kind of functions the organization has, and support and put in what the organization needs, It's like looking at what kind of output can be produced. AI is the same. In order to actually plan services, it is good to know how AI learns. In the end, there's certain input, and AI produces the output. It's about how to further improve the quality of the output. We create a correct answer sheet called learning data, and when we give it to AI, the AI sees the correct answer sheet and solves the problem again. By repeating that, AI increases performance.
Also, since a service is not planned and finished once, it is necessary to develop the service, but in order to do so, “What kind of data is needed to make this AI work?” I think it's important to consider these things when planning a service.
Ryan: So maybe miners working currently can give them some insight into the AI industry in the future.
David: Nowadays, AI technology has advanced so much that it seems that the time when AI can be introduced even if the performer or engineer is not inside is much closer now. Of course, to provide that AI, we can use solution companies or things provided by places like Google or Amazon in the form of APIs. After all, what do we need to adopt AI in our company? Then, you can think of it as data. Recently, AI leaders always say that it is important to clearly understand what problems I want to automate in the end, as there are people who plan data rather than AI engineers among the main members to introduce AI. If there's a problem I want to automate, just simply digitize it. If you think of AI more simply, you can think of it as a black box. It's just a device that produces the desired output when putting in the intended input.
For example, if I were to automate the task of classifying something, there would be an image (input) to classify it, and there would be a result (output) of the classification. If you only prepare data for that, I think I could tell you that it is important to build up data internally well because there are actually many companies that can learn and provide that data to AI.
Ryan: I think the name of the data is ambiguous. It's hard to recognize what exactly the data is.
David: The part that I want to emphasize repeatedly is what kind of things I want to automate. The voice recognition function in a mobile phone is to put in (input) voice (data) and then change the output to text (data). Data is just a title of calling for such things, what do I put in, what do I stick out... What behavior or function do I want to teach AI? Please think about these things.
David: One of our clients worked on a project that received similar personal taste surveys to recommend wine and traditional alcohol. So, if AI finds a pattern, it can tell you. What we need to do is, in this case, how can we find patterns associated with individual coffee tastes? We might be able to do a survey about their tastes. Among them, I think we can obtain meaningful data by deducing surveys that correlate with coffee preferences and repeating these tasks. In this case, I think the difficult case is correct. What is the input data that can check the coffee tastes? Since it's something we need to find, I think I can explain that it's possible when we find a pattern.
Ryan: The comic “Droplets of God” comes to my mind. If so, may I ask what exactly unstructured data mean?
David: I think you're describing unstructured data here as unorganized data that doesn't fit in well. In engineering terms, unstructured data is called data that is not visible as an Excel or table, such as images, text, voice, or video.