How AI is Changing Scientific Research and the Role of the Scientist

Scientists have always worked with incomplete information, large volumes of literature and limited time. AI is beginning to change how much of that information can be found, synthesized and evaluated, but the implications go beyond making literature review faster.

As AI becomes embedded in scientific workflows, researchers can spend less time assembling evidence and more time interrogating it, testing hypotheses and deciding what should happen next.

Janis Tam, a Science Liaison at Causaly, has seen that shift first-hand. Trained as a molecular biologist, she works directly with scientists across pharmaceutical R&D to understand their research questions and how AI can support them.

In this Q&A, Janis discusses what scientists need from AI before they can trust it, how agentic AI is changing the way scientific evidence is explored and evaluated, and why scientific expertise becomes more important as more of the research process is supported by AI.

Having worked with scientists across R&D, where do you see AI making the biggest difference to how they work?

One of the biggest is time.

Scientific research often requires scientists to search multiple databases, review large numbers of publications and bring information from different sources together before they can begin interpreting what it means.

AI can reduce some of that work by bringing relevant evidence together and synthesizing it around a specific scientific question.

But the value is not simply getting an answer faster. Scientists need to understand where that answer came from.

When evidence from different sources can be brought into one research environment, scientists spend less time moving between databases and documents and more time evaluating what the evidence means for their question.

There is also an organizational knowledge problem. Scientists joining an established program may need to understand years of previous work, including decisions made by colleagues who may have moved teams or left the organization.

Being able to interrogate both external scientific evidence and an organization’s existing knowledge could make that prior context much easier to surface.

When scientists are using AI to inform their research, what do they need to see before they can trust the answer?

Transparency is critical because scientists need to be able to inspect the evidence behind an AI-generated conclusion.

Scientists are trained to question evidence. They look for assumptions, limitations, conflicting findings and gaps in what is known. An answer without a clear connection to its underlying evidence therefore has limited value in scientific decision-making.

This is something we see directly when scientists begin working with AI.

Even when they receive a synthesized response, many will still open citations and inspect the underlying publications themselves. They want to understand whether the evidence supports the statement being made and whether anything important has been missed.

That is why features such as citations, links to the underlying source and evidence highlights matter. They allow the scientist to move from the generated answer back to the evidence supporting it.

For scientific AI, the question cannot only be “What is the answer?” It also has to be “Why should I trust this answer?”

You have mentioned that scientists are trained to question the evidence in front of them. How does that shape the way they approach AI?

Scientific skepticism is part of scientific training.

Researchers are expected to critically evaluate evidence rather than accept conclusions at face value. When AI produces a scientific answer, that same standard applies.

There is also understandable caution because AI has developed extremely quickly. Scientists want to know what information has been considered, whether important evidence has been missed and how a conclusion was reached.

I do not think that skepticism is something AI platforms should try to remove. It is something they should address.

The goal should be to give scientists enough visibility into the underlying evidence so that they can apply their own expertise and judgment to the output.

What makes scientists trust an AI-generated scientific answer?

Trust comes from being able to verify the answer against the evidence.

One of the strongest moments we see when introducing scientists to Causaly is when they can inspect the evidence behind a statement and see exactly where the support comes from.

For me personally, I experienced something similar when I used AI to investigate research questions I had previously worked on myself. I could see it retrieving many of the references I would expect to find and reaching logical, well-supported conclusions.

That kind of experience matters because scientists can compare what the AI produces with their own domain knowledge.

The AI is not asking them to replace their expertise with an answer. It is giving them a body of evidence that they can interrogate using that expertise.

Are you seeing scientists move beyond using AI to find evidence and start using it to help prioritize where to focus?

Yes, AI can support prioritization by gathering relevant evidence and evaluating potential options against defined criteria.

The purpose is not for AI to independently decide which target or disease a pharmaceutical company should pursue. Scientists can define the criteria that matter for the decision, then use AI to help assemble and assess the evidence against them.

The resulting prioritization table or scoring framework gives scientists something structured to interrogate.

They can examine why one option scored above another, review the supporting evidence and apply scientific and organizational context that may not be contained in the literature.

This is where AI becomes more useful as part of a decision-making process rather than simply as a search interface.

How will the role of the scientist change as AI becomes more capable?

Scientists could spend less time searching for and assembling information and more time thinking about what that information means.

Historically, researchers might spend substantial amounts of time searching databases, monitoring new literature, reviewing papers and assembling findings into reports.

If AI can support more of that work, the scientist can focus more heavily on scientific interpretation.

Does the evidence make sense biologically? What are the limitations? What is missing? Is a hypothesis worth testing? What experiment should happen next?

Those questions require scientific training and context.

The value of scientific expertise therefore does not disappear as AI becomes more capable. It becomes concentrated on the parts of research where judgement matters most.

Having seen how quickly AI has evolved in science over the past few years, where do you think it takes scientific research next?

The most meaningful future for AI in research is one where scientists can investigate more questions, make evidence-backed decisions faster and direct more of their expertise towards scientific reasoning.

We are already moving from AI that retrieves information towards systems that can synthesize, evaluate and help structure scientific decisions. Scientists are also looking for capabilities that go further, including calculations, statistical analyses and more specialized workflows.

Ultimately, my hope is to see AI enabling safer and more effective treatments developed faster, including in underserved areas where patients may have been waiting a long time for new options.

If AI can enable researchers to make more confident, evidence-based decisions, it could help turn scientific knowledge into meaningful discoveries and better patient outcomes.

‍