AI software development process | How to develop a company’s own AI?

With the recent development of AI, there are increasing opportunities for AI-equipped software to be developed. Each company is developing its own AI to compete with its competitors, and even companies that develop traditional software are increasingly being required to develop AI. This time, let’s explain the flow and points when developing software using AI.

AI can be broadly divided into two types

From a functional perspective, AI can be broadly divided into two types. Let’s take a closer look at their features to take advantage of the benefits of each.

Specialized AI

Specialized AI is AI that is strong in a specific field. What is generally called AI falls under this specialized AI. Roles and processing contents are limited, which has the advantage of increasing work efficiency.

Widely used examples include weather prediction systems, autonomous driving systems, image recognition systems, voice recognition systems, and chatbots.

General-purpose AI

General-purpose AI is an AI that has a thinking ability that is closer to that of humans than a specialized type, and is equipped with the ability to learn what actions to take on its own. The AI ​​that appears in science fiction movies falls under this general-purpose AI. It is hoped that it will be put into practical use, and research is progressing in countries around the world, but it has not yet come to fruition.

Business scenes where AI-powered software is utilized

Software that utilizes AI is used in many business situations and helps improve operational efficiency. Here is an example of how it can be used.

Demand forecast

Software that can forecast demand is used in industries that have inventory of goods and products, including retail, wholesale, and manufacturing industries. Demand forecasting using AI can calculate predicted demand values ​​and order quantities with high accuracy to prevent excess inventory and waste loss. This makes it possible to improve the efficiency of ordering operations and optimize inventory.

Inventory control

One of the issues raised in inventory management operations is that it is not possible to completely prevent human errors during work. There are many cases where the intuition and experience of veteran employees is relied upon, and there are many cases where predictions deviate significantly. By introducing AI, it becomes possible to perform accurate inventory management and demand forecasting, thereby reducing labor costs.

Create shift

An increasing number of companies are using tools that incorporate AI technology when creating employee shifts. This not only reduces the workload of creating shifts, but also allows staffing to be done based on the Labor Standards Act and employment regulations, as well as optimizing personnel costs and staffing. On the other hand, it may be difficult to consider employee turnover or shift creation based on individual circumstances.

sales forecast

Instead of relying solely on the experience of the person in charge, AI instantly makes a grounded sales forecast based on the data accumulated over the years. By making sales forecasts that closely reflect the actual situation on the ground, there is the advantage of preventing excess inventory and minimizing the impact on management.

Examples of AI supporting software development

AI is not only used in cases of incorporating AI into software, but also in the software development process.

Bookmark, an IT start-up company in Toronto, Canada, is building a platform that uses AI to analyze customers’ business models and automatically outputs designs for websites and e-commerce sites within two minutes. It is already being used at 40,000 locations, mainly local small and medium-sized businesses and individual business owners.

DEEPCODE, a startup originating from the Swiss Federal Institute of Technology in Zurich, Switzerland, provides a platform that automatically learns source code published on Github and other sites and suggests code defects and rewrites.

In this way, there are increasing cases of automating the software development process by automatically learning from big data.

Automatic code generation is expected to progress gradually.

Automatic generation of the code that forms software is expected to be gradually enabled by AI. “Every set of software in the future is going to be the combination of some procedural code and some (AI) models,” Sid Sijbrandij, CEO of the web-based Git repository manager GitLab, wrote on the company’s blog. The future of software development will be a combination of procedural code and AI models.

In the same blog, GitLab Senior Developer Evangelist Brendan O’Leary wrote, “But this isn’t going to replace humans – it’s going to make the human role more critical to understand what’s important.” (This will make the role of humans more important in thinking about what is important, rather than what is important.) He suggests that AI will become a second set of eyes that check the details of the code that humans cannot find.

Software development using AI is different from the traditional flow

The process of developing software using AI is different from that of general software development.

In general software development, the finished product is first envisioned, and then a development process is established to create the finished product. In contrast, software development using AI aims to create highly accurate models with few errors by adjusting data and parameters.

The important thing when using AI is to build a model with fewer errors compared to previous data. However, it is very difficult to reduce the error to 0% using AI, so it is important to repeat trial and error to get the error as close to 0% as possible.

The quality of software improves by repeating learning to find patterns with fewer errors. You need to adjust it to suit your learning content, and you need to be aware that it will take time.

Specific flow of AI-equipped software development and operation

In order to develop and operate software that utilizes AI, it is necessary to create a flow that is different from traditional software. Let’s take a closer look at how we should proceed.

Conceptualize the outline of the software to be developed

The first thing you need to do is figure out what problem you want to solve using AI software. We must avoid the idea of ​​“just adding AI”.

Once we find a problem that can be solved with AI, we will gather the necessary human resources and set up a system. If it is difficult to recruit human resources in-house, one option is to collaborate with an external partner.

Verify whether the concept is feasible

Once the system is in place, we create a temporary AI model (mockup) and verify whether the purpose of solving the problem can be achieved without any problems. There are three things I would like to verify at this time:

① Machine learning model and data: We verify whether the quality and quantity of data is sufficient and whether the system and performance set at the structure stage can be demonstrated.

②Operation: Verify whether it is possible to handle cases where the model is output by mistake, and whether there are any problems with the speed of learning and inference.

③ Confirmation of investment results and schedule: Verify whether the return on investment and schedule that were envisioned at the stage of developing the concept are achievable.

actually develop the software

Once verification is complete, we will develop software that can be used in practice from the mockup model. The general flow is the same as general system development.

Among the reasons for failure not only in AI systems but also in system development, poor requirement definition is the most cited reason. It is important to develop a system that can secure data capture capacity and processing speed.

Operate and verify effectiveness

We will verify whether there are any problems with the AI ​​system that has started operation and confirm whether the initial goals can be achieved. If there are any problems, we will review the algorithm and take steps to improve it as we continue with tuning. Since there are few precedents for software using AI, post-launch verification is extremely important.

How to develop a company’s own AI

There are two ways to develop a company’s own AI: developing it through programming and using AI development tools.

The former method reduces development costs, but requires programming skills and development engineers. The latter allows you to utilize AI algorithms without any programming. The system is designed to be easy to use, even for first-time users, so you can install it with confidence.

summary

In order to promote greater operational efficiency than ever before, the development of software that utilizes AI is expected to continue to advance.