Introducing Artificial Intelligence (AI) in Companies

There is a lot of interest in introducing AI in companies, and many types of language models also are created by many organizations.
AI has the size indicated by the parameters of a language model, and AI has purposes in a commensurate manner with its size.
This article describes how to introduce AI in companies.

• Large Language Model (LLM) and Small Language Model (SLM)

AI comes in different sizes and aims. LLM is aiming for AGI (Artificial General Intelligence). And I think SLM aims to solve specific tasks with AI orchestration. Therefore Multimodal AI, RAG (Retrieval-Augmented Generation), and so on are essential for SLM.

What I would like you to note here is that Apple is not included in the figure above.
Apple’s Edge AI does not add to this article about the size and purpose of Edge AI, because it’s used in a different use-case than Cloud AI.
It is possible to introduce Edge AI into a company, but at this stage, I think it is difficult to introduce Edge AI into a company unless cloud AI has already been introduced.
This article will focus on cloud AI, rather than Edge AI, as an introduction to AI in companies.

• Utilization of accumulated data in the AI era

Size is one of the indicators for AI classification, and another indicator for AI classification is the “confidentiality of the information source.”　More precisely, it is the right to access information.

Many companies find it useful to orchestrate using a small language model without the need for a large language model. At this time, it is possible to use RAG to generate a solution that incorporates information from outside the company, or it is also possible to generate a solution using confidential internal data. What is important is that companies utilize accumulated data in the same way that they have done in the past, and this will not change in the AI era.

Oracle, one of the companies that specializes in data utilization, provides the foundation for corporate data utilization extending with AI (below left). This website describes an architecture for introducing AI with a product called VizSeek (below right) that provides secure access to data.

• If the organization already introduced Microsoft 365

Accumulated data with Microsoft 365 are already protected with Microsoft 365. In responding to implementing organization maturity of security approach(below right), there is secure access that is under compliance policy, and paradigm with the governance of the organization.

Thus for introducing AI to organizations, it is not good that move their data to other storage bother, such as S3 of AWS or Google Drive, because it should re-struct IAM roles of AWS Control Tower of AWS Organizations IAM, or Google Cloud Project.

Utilizing accumulated data of the company with AI is the same as leveraging SharePoint data or OneDrive data in Microsoft 365. The data can not be accessed in SharePoint is also can not be accessed using AI. On the other hand, when generating content with AI, it will be grounded with SharePoint data OneDrive data, and so on that can be accessed. Copilot for Microsoft 365 is the service, and Microsoft Copilot Studio (below left) is used to customize its service for specific companies.

• AI in Companies without Microsoft 365 data

Accounting data or human resource data is divided from business data generally, so there are cases in which several workforces have an accounting system’s ID or human resource system’s ID. If the identity federation is not complete, Its permissions are held in these systems. In addition, there might be device management without Intune, asset management, and access to IoT data as unmanaged.

These data should be stored in individual storage match access permission, then divided schemes of several AI on orchestration foundation. These cases can be implemented with Azure AI Studio or semantic kernel programming.

>>Next Model local launch(under construction)