Generative AI, such as ChatGPT, has reached millions of users because it promises impressive benefits. An entirely different AI-based solution is Predictive AI, which drives data analytics engines to find and predict patterns in complex data lakes. Let's understand use cases and challenges from a Performance Engineering perspective.
What is Predictive AI?
As we can guess from its name, predictive AI uses historical data to predict future events. We find predictive AI in many observability tools, such as Dynatrace, to create capacity forecasts or identify the root cause of complex reliability problems.
What is Generative AI?
When discussing generative AI, we refer to creating new content from existing content. ChatGPT is one of the famous GenAI solutions that provide capabilities to understand so-called "prompts," which refer to our questions and generate answers.
Use Cases for Predictive and Generative AI in Performance Engineering
Many AI projects fail because they choose the wrong approach. For instance, using GenAI to predict system utilization would result in colossal waste and project failure.
One of GenAI's best use cases involves high-quality training data. When we start the GenAI training phase, the system will understand the training data's structures, underlying patterns, and relationships. Once the training is complete, our GenAI solution can create new content based on its understanding of the training data.
Predictive AI is powerful when we have high-quality historical data. Applied to performance engineering, we see use cases such as identifying the root cause of performance problems. Compared to GenAI, we have a much shorter learning phase for Predictive AI.
What are the Challenges?
A drawback for GenAI is the trustworthiness of its created content. We can easily fall into a trap and take all its output for granted, but it's a fact that GenAI systems hallucinate.
Both AI approaches concern data quality. You can only expect correct outcomes if you predict the future from correct historical events. Similarly, GenAI will create incorrect content from low-quality training data.
From a performance engineering perspective, we see massive infrastructure requirements for GenAI and the need to optimize the training phase, its throughput, and latency.
What are the Advantages?
We are sitting on a gold mine of data these days. By using GenAI and Predictive AI to make better conclusions and predictions or create new ideas, we have powerful tools.
Data analysis was the domain of humans in the past. Thanks to the invention of Predictive AI this time consuming work is automated and we can focus more on optimization and innovation.
Generative AI is powerful when we use it to create new ideas. In some cases, we can also learn new content faster because it answers our questions instead of providing a long list of links to websites, and we would have to read them and find our answers by ourselves.
In our next blog post, we will share ideas on how to implement AI for Performance Engineering more effectively.
Happy Performance Engineering!
Comments