AI's Reality Check: Why Readiness Isn't Just About Algorithms

Are you on the AI hype train? It’s hard not to be. Even if you are a data scientist who feels you’ve seen this train before, you’re under pressure to adopt AI. Not just any AI, but generative AI and neural networks promise industry transformation.

Beneath the glossy marketing and buzzwords, the reality is harsh. Companies are scrambling to keep up, often diving headfirst into AI without genuinely understanding the challenges that lie ahead. From disastrous data management to underestimating the role of storage, this article unveils the hidden pitfalls that derail AI projects and what to do about them.

Shifting sands: The appeal of RAG and pre-trained models

In the not-so-distant past, companies often embarked on their AI journeys by attempting to build foundation models from scratch.

This approach required substantial resources, including massive datasets, specialized infrastructure, and highly skilled personnel. It’s like a black hole regarding budgets and adding resources.

A notable shift has occurred in recent times. As Yifeng Jiang, APJ principal solutions architect for data science at Pure Storage, summarized it, "More organizations started to think about inference, especially RAG (Retrieval-Augmented Generation), as the entry point to their AI journey."

This shift is driven by practical considerations and the availability of powerful pre-trained models. RAG, in particular, offers a more accessible path to AI adoption, enabling enterprises to leverage existing models and tailor them to their specific needs.

What they never tell you about data quality and quantity

RAG may be accessible, but data remains the cornerstone of AI success. While data quality and quantity are crucial (with the constant mantra about “garbage in, garbage out), their relative importance can vary depending on the stage of the AI journey.

This is a challenge for companies who may be new to the AI journey. As Jiang explains, “If you are to say, I want to train a foundation model from scratch, then you need to heal your master data. Well, if you just want to do RAG, you can start with some relatively smaller piece of data, but it still has to be high quality.”

Getting quality is where companies underestimate the challenges of data preparation. It’s a data science problem that AI just makes it worse.

"AI just makes it even more difficult because you're not just using a single source of data," Jiang explains. "You're not just using your structured data. You also need your unstructured data."

Gathering, cleaning, and organizing data from diverse sources can be time-consuming and complex, but unfortunately, it is necessary.

Underprovisioning space

Another area that companies overlook amidst the focus on data, algorithms, and computational power is the role of storage in AI readiness.

However, as Jiang emphasizes, "Storage is also important, not just for training, but also very important for inference and RAG." The storage requirements for AI workloads can be immense, particularly when storing vector embeddings for RAG.

"We did some experiments, for example. We realized that storing your data in this so-called embedding format expands the size of your data up to 10x," Jiang reveals.

Inadequate storage can bottleneck performance, hindering the efficiency of AI models, especially in inference scenarios where real-time responsiveness is crucial. “After all, you don't want your customer to wait for 10 seconds before the response comes back,” Jiang observes.

Furthermore, the costs associated with underprovisioning can escalate rapidly. As data volumes grow and AI models become more sophisticated, the need for scalable and high-performance storage becomes even more critical.

Organizational readiness: It’s not just about technology

While technological infrastructure is essential, AI readiness extends beyond hardware and software. Organizational factors play a pivotal role in determining the success or failure of AI initiatives — a significant issue where companies often backtrack after starting an AI project.

One crucial factor is executive buy-in. As Jiang notes, "Every successful AI project, I would say, should have executive sponsorship." Leaders need to understand AI's potential, set realistic expectations, and champion its adoption throughout the organization.

Moreover, companies must be prepared for the transformative impact of AI. “With AI, if done correctly, you're not just replacing a process," Jiang observes. "You might be changing the process and even impacting organizational structure.”

Embracing this change and fostering a culture of innovation is critical to unlocking the full potential of AI. But with anything to do with organizational change, it’s easier said than done.

Starting an AI Project: Key Considerations

Embarking on an AI project requires careful planning and preparation. Here are some key factors that Jiang asks companies to consider:

• Clear Use Case: Define your AI initiative's specific and measurable objective.
• Realistic Expectations: Set achievable goals and avoid over-promising.
• Data Readiness: Assess the quality, quantity, and accessibility of your data.
• Infrastructure: Ensure your infrastructure, including storage, can support your AI workloads.
• Executive Buy-In: Secure leadership support and foster a culture of innovation.

So, how do you know whether your company's AI is ready?

Assessing AI readiness involves evaluating both technological and organizational factors. Here are some questions to ask:

• Do you have a clear understanding of your AI goals and use cases?
• Is your data organized, accessible, and of sufficient quality?
• Does your infrastructure have the capacity and performance to handle AI workloads?
• Do you have the necessary skills and expertise in-house or access to external resources?
• Is there executive buy-in and a willingness to embrace change?

By addressing these questions and taking a proactive approach to AI readiness, companies can position themselves for success in the age of AI. 

This article is part of our recent eGuide on AI Readiness. Get the full guide by downloading here. 

Image credit: iStockphoto/BitsAndSplits