Credit: Artem Peretiatko / iStock

Machine learning and generative AI: What are they good for in 2025?

Jun 2, 2025

Why It Matters

While generative AI is widely accessible and useful, businesses need to know when to use other AI tools, like traditional machine learning.

Less than five years ago, machine learning was one of the predominant ways businesses were using artificial intelligence. In April 2021, we called machine learning a “pervasive and powerful form of AI … changing every industry.”

But after ChatGPT-3.5 was released in 2022, many organizations shifted focus to a subfield of AI, generative AI, which can be used to create new content. Generative AI is machine learning (more on that below).

Traditional machine learning is now an established technology in many organizations, and today leading firms are focusing on use cases for generative AI. In a 2024 survey of senior data leaders, 64% of respondents said that generative AI has the potential to be the most transformative technology in a generation.

While generative AI is widely accessible and has many novel applications, you still need to know when it’s best to turn to other forms of AI, like traditional machine learning.

We talked with two MIT Sloan AI experts — associate professor and professor of the practice — about where generative AI is replacing predictive machine learning, when machine learning is still the most effective tool, and how businesses are using the technologies together.

What is machine learning?

Machine learning is a type of artificial intelligence that enables computers to learn without explicitly being programmed. Where traditional computing requires people to create programs that give machines detailed instructions on what steps to take to complete a task, machine learning programs can learn from examples.

Machine learning is used for many purposes, from predicting customer behavior to assessing potential fraud in bank transactions to creating tailored search results on shopping sites.

The data used to fuel machine learning — including generative AI tools — can be numbers in a spreadsheet, text, images, audio, or video. The more data a machine learning model is trained on, the more accurate the model will be. For machine learning to work, there must be patterns within the data that the application can identify and analyze.

“The basic idea of machine learning is, it’s a lot easier to collect data than to collect understanding,” Ramakrishnan said. For example, it’s easier to provide a machine learning program with thousands of photos of animals and tell it which ones depict cats and which depict dogs, as opposed to trying to teach a program all the complicated ways a cat can be distinguished from a dog. Feeding the program labeled data helps it learn how to tell the difference between the two on its own.

Machine learning “makes decisions that generalize patterns that we would not have found otherwise,” Gupta said. “It’s as good as the data and the models that we have.”

Thus, machine learning is best suited for situations with lots of data — thousands or millions of examples, like recordings from conversations with customers, sensor logs from machines, or ATM transactions.

What is generative AI?

Generative AI is a newer type of machine learning that can create new content — including text, images, or videos — based on large datasets. Large language models — AI programs that can process and generate text — are a prominent type of generative AI. A hallmark iteration of generative AI, ChatGPT, was released by OpenAI in 2022 and quickly took off because of how well it was able to respond to user prompts written in plain language and then quickly generate new content. Other commonly used chatbots or LLMs include Anthropic’s Claude, Google’s Gemini, Microsoft’s Copilot, and Meta’s Llama, which have all been updated in the past year to provide more accurate results and be more responsive.

“Machine learning captures complex correlations and patterns in the data we have. Generative AI goes further,” Gupta said. Fine-tuned, specific generative AI models can identify relationships within traditional datasets that machine learning cannot. “That’s where the edge lies,” Gupta said.

Instead of making a prediction or identifying a pattern, generative AI creates new content — it can answer questions, compose emails, or brainstorm ideas, for example. “There are so many use cases for GPT models these days,” Gupta said. “You see a lot of companies trying to find a way in which they can use them within their own frameworks, be it to transcribe calls in a call center, navigate policy documents, or help new employees learn the company’s existing software code.”

However, Gupta cautioned that companies developing or using generative AI or machine learning should be aware of potential issues, including inaccuracies and bias.

Best use cases for generative AI

In addition to its main function, which is generating new content, generative AI is taking over tasks that traditional machine learning has historically performed. These situations include:

When you’re dealing with everyday language or common images. LLMs have been trained on a large amount of text or images and can be used “off the shelf” to classify and detect things. For example, a company might want to analyze online product reviews to identify user reports of product defects. This once meant building a machine learning model trained to identify such reviews — a process that takes effort, time, and money. Today a company can input product reviews into a LLM and ask it whether the dataset contains any product improvement insights, Ramakrishnan said.

GPT-4 and similar models “can be more accurate than a custom-built machine learning model, and you can get an application up and running much sooner,” he said.

Generative AI models are also becoming more affordable, Ramakrishnan noted, so over time, fewer companies will be priced out of using them.

When you want a more accessible option. Using generative AI models is something many software engineers can do without a large amount of extra training, whereas building machine learning models requires technical expertise. Generative AI “is a democratizing force in that sense. It makes it way more accessible,” Ramakrishnan said.

If a problem or opportunity is based on using everyday information, “try generative AI first,” he advised. “Don’t reflexively go back to machine learning like you used to.”

When traditional machine learning is the better option

In some cases, though, machine learning is still the best option. Those situations could include:

When you have privacy concerns. You must exercise caution when feeding proprietary, sensitive, or confidential information into LLMs, because there is the potential for data leaks. And while it is possible to build your own private models, that requires specialist technical skills that may not be easily available in your organization. In these situations, you might want to stick to “the old-fashioned way,” Ramakrishnan said.

When you’re using highly specific domain knowledge. LLMs are trained on widely available data and suited to deal with everyday information. But they may not be as accurate for highly technical or niche tasks, like medical diagnoses based on MRI images. “If you are working on a domain-specific problem in which a lot of technical knowledge is required, a lot of jargon is involved, and the particular problem you’re working on is very particular to your company or your organization … you probably want to go the traditional [machine learning] route,” Ramakrishnan said — though he noted that generative AI models are improving rapidly, so that might change over time.

When you already have a machine learning model. Organizations have put a lot of effort into building machine learning programs for specific applications, like identifying potential fraud in credit card transactions. “There is probably not a huge urgency to rip them out and try to replace them with a generative AI system,” Ramakrishnan said. “The question is, what are the new use cases, the new things? That’s really where the decision point is going to come up.”

When to use machine learning and generative AI together

In several situations, machine learning and generative AI can be used together for better outcomes. These scenarios include the following:

When you want to augment a machine learning model. “Algorithms don’t have twenty-twenty vision of the world, and they’re as good as the models that we provide them. So if we can provide them [with] more context about the world using generative AI, then that only makes them better,” Gupta said.

She offered the example of a dataset containing people’s names, their heart rates, and the speed at which they have run. “A machine learning model will be able to predict information like the cardiac fitness of each person, cluster them into groups, or do a performance benchmarking,” Gupta said. “Generative AI-augmented machine learning might be able to squeeze a little bit more from the name of the person — inferring age and other demographics, using context external to this data.”

When you want to easily design a machine learning model. If you want to build a machine learning model, you can feed the data and instructions about desired function and techniques into a generative AI tool and ask it to build models, evaluate them on other datasets, and report on the accuracy of the models.

When humans and AI work best together

AI uses lots of data center energy — but there are solutions

How generative AI can boost highly skilled workers’ productivity

Generative AI “is changing the life and workflow of machine learning people,” Ramakrishnan said, noting that the output of the models needs to be continually analyzed and critiqued to ensure that hallucination and errors aren’t compounded.

When you want to generate data for a machine learning model. In cases where you don’t have enough data to properly train a traditional machine learning model, generative AI can be used to create synthetic data, which has the same statistical properties as a real-world dataset.

When you want to prepare structured data for a machine learning model. Tabular data in situations like industrial settings often contain errors, such as missing values, that need to be addressed before the data can be used to train a model. Rather than having to be cleaned up manually, the data can be uploaded to an LLM with a prompt to look for anomalies or mistakes.

“Generative AI makes the traditional machine learning workflow more efficient, all the way from data procurement to data cleaning to actually modeling,” Ramakrishnan said. “Every step of the process, you can use generative AI as sort of a turbocharger. But this is not a free lunch. The price you pay is the need to for constant vigilance to ensure that LLM-generated outputs are accurate.”

Given the variety of AI tools out there, deciding when to use what tool is becoming another skill AI practitioners need to have.

Ramakrishnan’s main takeaway: “If you want to generate stuff, use generative AI. If you want to predict things, but with everyday stuff, try generative AI first. If you want to predict things on domain-specific stuff, do predictive stuff, [use] traditional [machine learning]. It’s as simple as that.”