Artificial intelligence (AI) is playing an increasingly important role in drug discovery, especially in early-stage research where computational tools promise faster data analysis and more informed decision-making. However, translating these advances into reliable scientific workflows remains a major challenge.
At SLAS Boston 2026, Drug target review 20/15 We spoke to Dr. Raminderpal Singh, Global Head of AI and GenAI Practice at Visioneers, about how AI is providing value in early detection and why expectations for the technology often outstrip actual implementation.
Origins of AI in modern drug discovery
According to Singh, the current discussion around AI in drug discovery began more than a decade ago with major advances in biological data generation and computing infrastructure.
“10 to 15 years ago, machine learning became very popular because large amounts of data became very cheap and easily available,” Singh explained.
While the expansion of genomics and other omics technologies is generating vast datasets, cloud computing has made storing and processing them much more affordable.
“Genomics has become a big topic with the availability of sequence data,” he said, adding that cloud computing has also made a big contribution and helped drive interest in AI.
However, Singh argued that the technology being adopted during that period was not necessarily what people understood as AI.
“It wasn’t really an AI wave. Nobody even knew what AI meant. It was more like being able to do more advanced math with a lot of data,” he said.
These developments have created strong expectations that AI will rapidly transform pharmaceutical research. Many in the industry believed that a data-driven approach would significantly accelerate target identification and drug development.
10-15 years ago, machine learning became very popular because large amounts of data became very cheap and easily available.
“Drug discovery is going to be the big winner,” Singh recalled people saying at the time.
But years later, according to Singh, the reality was less impressive.
“Five to seven years into the late 2010s, people are scratching their heads and thinking, ‘This AI is crap,'” he says.
Much of that frustration arose when early optimism about AI ran into the practical challenges of drug discovery. While machine learning can analyze large data sets and generate predictions, translating those insights into validated targets and clinically actionable drugs has proven to be much slower and more complex than many expected.
Generative AI changes the conversation
There was a resurgence of interest in AI around 2022 with the advent of generative AI (GenAI) tools. Large-scale language models (LLMs) such as ChatGPT have demonstrated the ability to analyze and generate natural language, allowing researchers to extract knowledge from scientific literature.
“Large language models can do natural language. They can also do knowledge extraction,” Singh said.
This feature allows researchers to work not only with numerical datasets, but also with written information such as research papers, figures, and experiment descriptions. These developments are creating new enthusiasm across the industry. But Singh believes the excitement is once again outpacing the actual reality of implementation.
Growing expectations for generative AI
According to Singh, generative AI is sparking renewed interest across the technology and pharmaceutical industries.
“In 2025, the word ‘agent’ will appear and everyone will be a building agent,” he said. “Everyone’s building an agent for this and an agent for that.”
Part of the excitement comes from the experience of interacting with conversational AI systems like ChatGPT that can provide instant answers to questions.
“We’re in a TikTok world right now,” Singh said. “With ChatGPT, we can ask questions and expect good answers right away.”
However, scientific research requires a much more structured system than simply asking questions.
Because of ChatGPT, we expect to ask a question and get a good answer right away.
“These tools are processing engines, similar to machine learning models. They work as part of a workflow. Someone has to design those workflows.”
According to Singh, the potential of AI technology can be easily overstated if systems are not designed carefully. In drug discovery, these systems must operate within structured workflows with carefully selected data, validation steps, and clear guardrails. Without this framework, even powerful models may produce outputs that appear convincing but are difficult to reproduce or translate into real experimental decisions.
The hidden costs of generative AI
Another challenge that emerges with generative AI is the cost of using large language models at scale.
“When you’re using a large language model, you’re paying an access fee,” Singh says. “Every time the model infers something or performs a task on my behalf, I’m paying a token.”
This token-based pricing model means costs can quickly add up for researchers who rely heavily on the tool.
“We hear people say, ‘This is really great, but we’re spending two or three thousand dollars a month,'” Singh explained. At the scale of large pharmaceutical companies, which employ thousands of scientists, this can create new budgeting challenges.
“When you have thousands of scientists moving around, you can’t have each of them spending $2,000 or $3,000 a month on tokens,” Singh said.
Workflow remains the central challenge
Even though AI technology is advancing rapidly, Singh believes the biggest barrier to adoption lies in how organizations design their research workflows.
“The barrier goes back to the discipline in creating the workflow,” he said.
Building an effective system requires careful planning for data acquisition, data engineering, processing, and interpretation.
“Building complex workflows takes time,” Singh explains.
Large-scale language models are also probabilistic systems, meaning they do not always produce the same answers to the same questions.
“If you ask the same question twice, you won’t get the same answer,” Singh said.
Therefore, organizations must invest time in designing systems that guide how the model is used.
Looking beyond large language models
While large-scale language models are increasingly integrated into research workflows, Singh believes the next big step in AI development may come from systems known as world models.
“The next generation beyond large-scale language models is the world model,” he said.
World models aim to simulate complex systems by integrating different computational approaches to represent biological processes.
The next generation beyond large-scale language models is the world model.
“World models are the use of large language models and other types of models to create large-scale simulations of systems,” Singh explained.
Such simulations could eventually allow researchers to test hypotheses on computers before conducting laboratory experiments.
“Early drug discovery will be a very different experience once the global model becomes a reality,” Singh said.
For researchers feeling overwhelmed by the rapid pace of AI development, Singh recommends a simple starting point: start using existing tools.
“The first step for scientists is to actively use large-scale language modeling apps,” he advised.
These tools are useful for tasks such as literature analysis, knowledge extraction, and report generation.
“For about $20 a month, you can get a tremendous amount of work done. It’s like having a colleague, an intelligent colleague, in the room.”
Different models have different strengths, but Singh encourages scientists to experiment and find the tools that work best for them.
You can do a huge amount with about $20 a month. It’s like having a colleague, an intellectual colleague, in the room.
“For me it’s Claude,” he said. “But when I do a lot of searching on the web, I use Perplexity because it’s really good at scraping and collating information.”
Ultimately, Singh believes the best way for researchers to understand the potential of AI is to simply start using it.
“Choose your tools, find one you like and use it,” he concluded.
As AI technology continues to evolve, many organizations are exploring how best to integrate it into their existing scientific processes. Tools such as large-scale language models are already helping researchers navigate complex datasets and scientific literature, but their long-term impact on drug discovery will depend on how effectively they are integrated into research workflows and experimental decision-making.
