The advent of OpenAI is causing enterprises to scramble to build a more coherent data strategy

As a reader, you might be thinking, “Not another ChatGPT article!” Let me assure you, that this isn’t about the technology but rather it is about connecting the dots for executives on how generative AI can and should fit into a Data and AI transformation within their business.

To begin, we need to understand that prior to generative AI we started with artificial “narrow intelligence.” Now, the traditional approach of AI model development aligned to narrow intelligence involves models being developed and trained separately or in silos. So, for example, if you want to train an entity recognition model, it requires specifically tagged entity data and, similarly, sentiment analysis model development requires specifically trained sentiment data, which can be a bottleneck in training machine learning models. As a result, not only is a lot of tag data required, but the data remains segmented and used for one specific “narrow” task, leading to just higher development costs and slower deployments.

Today, the models are flexible, re-usable and that can be applied to just about any domain or industry. Task and foundation models are really at the heart of many of the latest breakthroughs. Foundation models are really just a machine learning model that is trained on a large amount of unlabeled data, and it allows the models to adapt to a wide variety of tasks, moving away from the narrow use-case to a more general one. These models capture the general patterns and structure of data, and the aim is to replace these task specific models by unifying different tasks and mortalities into one. Basically, it eliminates the need to train individual models and integrate several models together. This is a massive leap forward and we all must appreciate the immense significance this has for businesses.

Models like GPT— or Generative Pre-trained Transformers — are called “transformers” because they are based on self-attention, allowing the model to focus on positional importance of the input sequence when processing it. It is pre-trained because it is already trained on the corpus of the internet. Meaning, it does not need to use your specific data to provide answers to your queries – though that data can be added to the training set to make it more relevant to your business. Importantly, this data and service can exist within your cloud environment.

However, transformer technology has been around for some time. So why are businesses so excited about the potential of this technology only now? For the first time ever:

1. These models exist as a first-party service making it extremely easy for businesses to call via an API and it is both cost and time efficient to get to business value.
2. These models come with enterprise-grade security and governance within Microsoft’s cloud environment (e.g. active directory, role-based access control, and customer-managed keys)
3. Usage of these models in business processes comes pre-built with responsible AI monitoring and moderation to detect and mitigate harmful use.
4. Integration of these services into the business applications is seamless and comes with the reliability, consistency, and scalability of Azure.

What all this means is that for the first time businesses can use this cutting-edge AI technology in a safe, governed, and responsible way immediately without spending months to customize the deployment for a specific use case.

This is the watershed moment in the cycle of AI and particularly for its adoption into business.

In addition to building your own application with OpenAI models, Microsoft is already pre-integrating advanced AI capabilities into its services such as Teams, Power Apps, GitHub Copilot, Dynamics Co-pilot. There are a myriad of use cases, ranging from content generation, summarization, code generation, and semantic search, across multiple functions in almost all industry sectors.

Given OpenAI models are at the bleeding edge of technology, they can make mistakes and lead to questionable outcomes. This is why majority of the uses are limited to low-risk business processes like call-center automation, marketing and sales – for now.

What is most interesting is that this exposure to what generative AI can do, has generated interest in other AI capabilities such as pre-built services, like OCR and machine vision. For example, Microsoft also recently released Florence, which combines language and vision foundational models — a new frontier in Artificial Super Intelligence.

Executives and boards are demanding that more be done to leverage the power of AI and enterprises are now re-imagining how their business could look, transformed with AI. While many of these visions existed on paper years ago, the advancements in AI today are making them a reality.

However, many enterprises are realizing that their data stack is not ready for this new wave of AI. With GPT or other forms of embedded intelligence, comes a tsunami of unstructured data, and hence there needs to be a way to capture, govern, store and use this data responsibly and at scale across the enterprise.

Accordingly, a coherent data strategy is a must, starting with having a clear understanding of the business value that AI can bring to the business even before talking about the technology.

In the next part of this article, we’ll discuss how to plan and execute an effective data strategy that leverages the AI advances discussed above and best practices for doing so. So, keep an eye out for Part 2.