Is Generative AI Overshadowing The Proven Workhorses Of Modern Tech?

Is Generative AI Overshadowing The Proven Workhorses Of Modern Tech?

Generative AI has emerged as the next wave of innovation amidst the ongoing evolution of the technological landscape, attracting the attention of both researchers and investors. However, the increased focus on generative AI has inadvertently cast shadows over several other technologies, slowing down investments and shifting focus away from them. These technologies, while still critical to various sectors, are seeing diminished focus and investment in favor of the advancements and potential offered by generative AI.

This article explores five such technologies that are being impacted by the stellar rise of generative AI.

1. Traditional Machine Learning and Deep Learning

Machine learning and deep learning have been the cornerstones of artificial intelligence, driving advancements in various sectors. However, the advent of generative AI, with its ability to create content and generate new data instances, is sidelining traditional ML models that are more focused on predictive analytics and pattern recognition. While generative AI builds on the principles of machine learning, its flashy capabilities and broad applications have attracted a lion’s share of funding, leaving conventional ML models grappling for attention and resources.

Generative AI, despite its revolutionary capabilities and potential, cannot entirely replace models based on traditional machine learning (ML) and deep learning for several reasons. Firstly, generative AI, particularly those models that produce new content or data, rely heavily on the foundational principles and techniques developed through traditional ML and deep learning. These underlying models are crucial for tasks such as pattern recognition, predictive analytics and classification, serving purposes that generative AI is not primarily designed for. Furthermore, generative AI models, especially the more advanced ones, require substantial computational resources, including processing power and memory, which can be prohibitive for many organizations.


The dependency on compute resources becomes significant when deploying these models at scale or in real-time applications, where the computational and energy costs can be substantial. Additionally, the training of generative AI models demands vast datasets, which can introduce challenges related to data privacy, availability and bias. In contrast, some traditional ML and deep learning models can be more efficient in terms of resource utilization and can be trained on smaller, more specific datasets. Hence, while generative AI opens new avenues for innovation and application, it complements rather than replaces the broad spectrum of existing ML and deep learning models, each serving distinct roles within the technology ecosystem.

2. Edge Computing and Edge AI

Edge computing, which is intended to bring computation and data storage closer to where they are needed in order to improve response times and save bandwidth, is shifting its focus.

The spotlight on cloud-based generative AI models, which require significant computational power and are often centralized in data centers, is diverting attention and investment from edge computing initiatives. This shift could slow the development of edge technologies that are crucial for real-time applications in IoT, autonomous vehicles and smart cities.

Edge computing faces significant challenges in fully embracing generative AI due to its inherent resource constraints. Generative AI models, particularly the more advanced and capable ones, require substantial computational power, memory and energy resources, which are often beyond the capacity of current edge devices. These devices are typically designed to be low-power and have limited processing capabilities to ensure efficiency and practicality in remote or distributed environments. Consequently, edge computing continues to rely on traditional ML models to bring intelligence to the edge. Traditional ML models are generally more lightweight, require less computational power and can be optimized to run efficiently on the limited resources available at the edge. They are capable of performing a wide range of tasks, from predictive maintenance and anomaly detection to image recognition, without the need for constant connectivity to centralized cloud resources. This makes traditional ML an indispensable tool for enabling smart, autonomous decision-making in edge computing scenarios, where real-time processing and low latency are critical.

Generative AI’s dependency on powerful GPUs for processing reflects a significant challenge for edge computing, as most edge devices lack the requisite computational power, rendering them not yet ready to fully support the demands of this evolving technology.

As edge computing evolves, there may be advancements that allow for more sophisticated AI models to operate at the edge, but for now, traditional ML remains the backbone of intelligence in edge computing architectures.

3. Natural Language Processing (Non-Generative Focus)

The field of NLP has been bifurcated by the rise of generative AI. While generative models are a part of NLP, they are now commanding a disproportionate amount of research and funding. This imbalance is at the expense of non-generative NLP tasks such as sentiment analysis, classification and entity recognition. These essential aspects of NLP, crucial for understanding human language, are being overshadowed, potentially slowing their advancement and application.

Running task-specific Natural Language Processing (NLP) models rather than relying on large-scale foundation models for language-related tasks presents significant economic and efficiency advantages. Task-specific models are typically smaller, more focused and can be fine-tuned to address specific language tasks—such as sentiment analysis, named entity recognition, or language translation—with greater precision and less computational overhead. This specialization allows for faster processing times, reduced memory requirements and lower energy consumption, making them more suitable for applications with limited resources or those requiring real-time responses.

On the other hand, foundation models, despite their versatility and broad capabilities, require substantial computational power to train and run, leading to higher costs and energy use. Moreover, the one-size-fits-all approach of foundation models may not be necessary for many applications where a bespoke, task-specific model can achieve better performance with a fraction of the resources. By choosing to deploy task-specific NLP models, organizations can achieve more efficient and cost-effective solutions that are tailored to their unique needs without the overhead associated with large, general-purpose AI models. This approach not only conserves resources but also allows for more scalable and sustainable AI implementations across a wide range of linguistic tasks.

4. Computer Vision

Computer vision technology, pivotal in enabling machines to interpret and understand the visual world, is facing competition from generative AI models that can generate realistic images and videos. These generative models, capable of creating visual content from textual descriptions, are overshadowing advancements in computer vision aimed at understanding and analyzing existing images and videos. The dazzle of content creation is sidelining the critical need for content interpretation technologies.

Foundation models based on vision and multimodal Generative AI, while offering extensive capabilities across a broad spectrum of applications, can represent an overkill for specific computer vision-based tasks. These large-scale models, designed to handle diverse inputs and generate or interpret complex multimodal data, often come with substantial computational and resource demands.

For applications requiring focused visual processing tasks, such as face recognition, custom-trained convolutional neural networks offer a more streamlined and efficient solution. CNNs can be finely tuned to the intricacies of facial features, enabling them to perform with high accuracy and speed while consuming significantly less computational resources compared to their generative counterparts. This optimization is crucial in real-world scenarios where rapid and reliable facial recognition is needed, such as security systems or identity verification processes.

Developers can achieve superior performance for targeted computer vision tasks by utilizing task-specific models like CNNs without the needless overhead that foundation models introduce. This approach not only ensures resource efficiency but also maintains the focus on the precision and reliability essential for applications like face recognition, where the stakes can be high and the margin for error is minimal.

5. Data Warehousing and ETL Technologies

Data warehousing and ETL (Extract, Transform, Load) technologies, essential for organizing, storing and analyzing data, are facing a new challenge. Generative AI’s ability to synthesize and analyze data is making these traditional data processing tools seem less critical. As more companies invest in AI that can automatically generate insights from raw data, the role of manual data preparation and analysis might diminish, impacting investments in these foundational technologies.

Even as vector databases and Retrieval-Augmented Generation models become mainstream, offering innovative ways to handle and process data, traditional ETL processes retain their importance in the data management ecosystem. Traditional ETL is fundamental for preparing and structuring data from diverse sources into a coherent, standardized format, making it accessible and usable for various applications. This structured data is crucial for maintaining the accuracy and reliability of information within vector databases, which excel at handling similarity searches and complex queries by converting data into vector space.

Similarly, RAG models, which leverage vast databases to augment content generation with relevant information retrieval, depend on well-organized, high-quality data to enhance their output’s relevance and accuracy. By ensuring data is accurately extracted, cleaned and loaded into databases, traditional ETL processes complement the capabilities of vector databases and RAG models, providing a solid foundation of quality data that enhances their performance and utility. This symbiotic relationship underscores the continuing value of traditional ETL in the age of AI-driven data management, ensuring that advancements in data processing technologies are grounded in reliable and well-structured data sources.


The rise of generative AI has indeed shifted the technological focus, overshadowing some of the core technologies that have been instrumental in our digital progress.

However, recognizing the unique value and irreplaceable roles of these foundational technologies is crucial. They serve specific purposes that generative AI cannot fully replicate, especially in scenarios requiring efficiency, precision and resource sensitivity.

Investing in and advancing a broad spectrum of technologies will ensure a more resilient, balanced and versatile digital future.

Source link


Be the first to comment

Leave a Reply

Your email address will not be published.