Don’t Forget About All the Boring AI

The AI space has undergone significant transformations over the past few decades. The pivotal breakthrough came with advancements in Deep Learning in 2012, which quickly led to the development of Large Language Models (LLMs) by 2018. Nowadays, when someone talks about AI, they usually refer to LLMs. Millions of people are now using ChatGPT, and the initial awe inspired by its capabilities is starting to wear off. The impact it has had on various sectors — from engineering to education to marketing — is immense. However, this hype cycle has also had some unintended negative consequences.

Businesses have faced the daunting task of adapting to rapidly accelerating technological advancements over the past few decades. This evolution has occurred through various hype cycles, such as “Big Data,” “Chatbots,” and “Deep Learning.” These buzzwords have been used to create startups, develop new features, and initiate projects — many of which have failed.

The pace of change is so rapid that companies are often stumbling ahead, missing critical steps. Some businesses are diving into generative AI initiatives without even having a robust document search setup. Others are jumping onto LLMs, neglecting simpler solutions like a Naive Bayes classifier, topic tagging pipelines, or recommender systems that could add significant value. They are overlooking the “boring” AI that existed before LLMs became mainstream.

Here’s what constituted AI before LLMs came onto the scene:

Classification

Classification involves categorizing data into predefined classes. For instance, a customer support system could automatically classify incoming tickets into categories like “Billing,” “Technical Support,” or “General Inquiry,” streamlining the support process.

Regression

Regression is used to predict a continuous value based on input variables. This technique can be employed to forecast monthly revenue based on historical sales data, user engagement metrics, and marketing expenditures, aiding in financial planning.

Clustering

Clustering is the task of grouping a set of objects so that objects in the same group are more similar to each other than to those in other groups. This method can be used to segment users based on their behavior, enabling more targeted marketing campaigns.

Optimization

Optimization involves finding the best solution from all feasible solutions. It can be used to allocate resources efficiently, such as in server load balancing, ensuring optimal performance and cost-effectiveness.

Search

Search algorithms are designed to retrieve information stored within some data structure. Advanced search features can help users quickly find relevant documents or records, enhancing user experience and productivity.

Ranking

Ranking algorithms are used to order items in a list based on certain criteria. This can be applied to prioritize tasks, tickets, or leads based on their importance or urgency, improving workflow efficiency.

Topic Extraction

Topic extraction identifies the main topics or themes within a collection of texts. This can be utilized to analyze customer reviews or feedback, helping businesses understand common issues or popular features.

Recommendations

Recommendation systems suggest items to users based on various factors. They can recommend relevant features, tutorials, or add-ons to users based on their usage patterns, enhancing user engagement and satisfaction.

Fraud Detection

Fraud detection involves identifying fraudulent activities within a dataset. Algorithms can monitor transactions and flag suspicious activities, helping to prevent fraud and ensure security.

Time Series Prediction

Time series prediction involves forecasting future values based on previously observed values. This is useful for predicting server load, aiding in capacity planning and ensuring a smooth user experience.

Why are we not doing all this?

Implementing these is not as easy as it sounds. In fact, deploying AI was quite challenging even before that huge buzzing LLM bee entered the room. Here are some of the main hurdles you would face:

Data Awareness

Every company generates and collects some kind of data stream. The first challenge is understanding what data you have, where it is stored, how often it is updated, and who has access to it. Without this foundational awareness, it’s impossible to make informed decisions about how to leverage your data.

Data Quality

Even if you know what data you have, the quality of that data can be a significant hurdle. Incomplete, inconsistent, or dirty data can lead to inaccurate models and unreliable insights. Ensuring data quality often requires substantial effort in cleaning and preprocessing.

Data Integration

Data is often siloed in different systems and formats across an organization. Integrating these disparate data sources into a unified, accessible format is a complex but necessary step for effective AI implementation.

Talent and Expertise

Building and deploying ML models requires specialized skills that may not be readily available within your organization. Data scientists, machine learning engineers, and domain experts are essential for the successful implementation of AI initiatives.

Infrastructure

Running ML models, especially at scale, requires robust computational infrastructure. This includes not only powerful servers and GPUs but also data storage solutions that can handle large volumes of data efficiently.

Model Interpretability

Understanding how a model makes its predictions is crucial for trust and accountability. Many traditional ML models can be complex and difficult to interpret, making it challenging to explain their decisions to stakeholders.

Scalability

Once a model is built and validated, scaling it to handle real-world data volumes and usage patterns can be another significant hurdle. This involves optimizing the model for performance and ensuring it can operate reliably in a production environment.

Regulatory Compliance

Depending on the industry, there may be various regulations governing the use of data and AI. Ensuring compliance with these regulations while implementing AI solutions can add another layer of complexity.

Change Management

Introducing AI into an organization often requires changes in workflows, processes, and even organizational culture. Managing this change effectively is crucial for the successful adoption of AI technologies.

Can LLMs Help Me with Boring AI?

LLMs are excellent companions for the ideation, research, and prototyping phases of a Data Science project. They can noticeably speed up these parts of the project. However, when it comes to preparing the final product, you will need experienced engineers and data scientists to pull it off.