Elsevier recently published its Researcher of the Future: a Confidence in Research report, which surveyed more than 3,200 academic and corporate researchers globally to understand how they are adapting to rapid technological, cultural, and institutional change. The report points to rapid AI adoption – with 58 percent of researchers now using AI for work – but also highlights continuing concerns around trust, training, and the reliability of AI-generated outputs.
Here, Mirit Eldor, Managing Director, Life Sciences Solutions at Elsevier, discusses the report’s main findings, where AI is already being used in drug development, and what will be needed to build confidence in AI-generated insights.
Based on the report, what are the most important trends shaping how researchers are using AI today?
Adoption has surged, with 58 percent of researchers using AI for work, up from 37 percent in early 2024. Most are using it for practical tasks such as finding and summarizing research and performing literature reviews, with many saying AI already saves them time. At the same time, 61 percent believe it will be "the creative force driving new knowledge" within two to three years.
But confidence is still patchy. Only 22 percent describe AI as trustworthy, and nearly half feel undertrained on how to use AI. For confidence to increase, researchers believe that tools should cite sources, be trained on current peer-reviewed literature, and demonstrate factual accuracy. For drug discovery R&D, the opportunity for AI is significant, but adoption will depend on whether tools are transparent, evidence-based and reliable enough to support high-stakes decisions.
In practice, where is AI having the most meaningful impact across drug development?
Data synthesis is one of the standout areas. Drug discovery researchers spend many hours reviewing huge volumes of data, and don’t want to miss any potential hidden connections. AI has huge value when it comes to pulling together literature, compound data, reaction dynamics and clinical results. These historically lived in separate places and took weeks to consolidate.
One example of this in practice is RARE Hope, which used AI-powered analysis to examine nearly 2,000 existing drugs, using curated knowledge graphs and biological relationship data to score and prioritise drug candidates based on their interaction with disease-related targets. This has helped narrow a long list of possibilities into candidates for expert review and validation.
Target identification and lead optimization are also areas where AI is accelerating drug discovery workflows. In target ID, AI helps researchers interrogate disease biology, connect evidence across datasets and prioritise targets for validation. In hit identification and lead optimization, it can narrow thousands of potential compounds to a workable shortlist. Beyond the time saving, it makes running multiple design-make-test-analyze cycles feasible in a way that wasn't practical before.
One pattern I've heard consistently from researchers is that while AI is indeed useful for hypothesis generation, validating those hypotheses is a different challenge. Solving that challenge requires scientific robustness that isn't yet embedded into workflows and available AI tools. This is where sound scientific context is required to support this important step.
There is increasing discussion around “agentic AI.” How do you see the role of AI evolving from assistive tool to active research partner?
The most capable agents are starting to go beyond information retrieval to replicate expert reasoning, working through complex, multi-step problems with something approaching scientific logic.
When an agent can be trusted to handle the evidence-gathering, the dataset searches and the target ranking, then AI becomes a genuine research assistant. That means researchers can have more time for creative ideation, for experimental de novo design and for the original thinking that moves science forward.
Beyond data workflows, autonomous agents are beginning to appear in the physical lab too. This is in the form of robotics platforms that can run experiments, adjust parameters based on results and feed findings back into the research pipeline without waiting for human instruction. The design-make-test-analyse cycle is an obvious candidate for this kind of end-to-end automation.
The risk that comes with handing agents greater autonomy over both data workflows and physical lab processes is the "black box" problem. If researchers can't follow how an agent reached a conclusion, then trust and reproducibility suffer. The answer is guided autonomy: predefined workflows and transparent, documented steps that keep humans in control of the logic while AI does the legwork. At Elsevier we see AI as a co-pilot, not an autopilot. It should augment rather than replace humans, helping researchers find what they need faster and make well-informed decisions earlier.
What factors matter most in building confidence in AI-generated insights?
Trusted tools and trusted data. Neither works without the other. In drug discovery specifically, a misidentified target has serious financial, reputational and patient consequences, so the bar for evidence-based decision making is high.
Our research showed that to increase use of AI, nearly six in ten researchers want AI tools that automatically cite sources. A similar proportion want AI tools trained on the most up-to-date literature (55 percent), have high factual accuracy (55 percent), and use peer-reviewed content (55 percent), and 49 percent want outputs to undergo regular expert review. While half of all respondents want input data kept confidential (50 percent) – this finding was even higher for corporate researchers (63 percent). This is understandable given drug discovery scientists are often working with proprietary data, unpublished trial data and commercially sensitive findings.
That's a demanding but reasonable checklist, and it maps to what good scientific practice looks like.
What needs to change to make scientific data truly “AI-ready?”
To make scientific data AI-ready at scale, organizations need to treat data preparation as a long-term strategy. Most scientific data is still built for human consumption. Today, it must also be managed and connected in ways machines can understand.
That starts with FAIR principles: making data Findable, Accessible, Interoperable and Reusable. It also requires data are enriched via a semantic layer that maps raw data to standard concepts, terms and relationships, preserving meaning and context while making it retrievable.
Ontologies are an essential element of a semantic layer because they allow AI to recognise that terms such as “Glucagon-like peptide-1” and “GLP-1” refer to the same thing, even when different datasets use different language. Combined with semantic enrichment and knowledge graphs, ontologies create a foundation where data is searchable, linked and reusable.
As AI becomes more embedded in R&D, how might the role of the researcher itself evolve?
More time doing science, less time doing administration. Our report finds that 58 percent of researchers say AI already saves them time, with 69 percent expecting more gains within a few years. As AI takes on routine tasks across data synthesis, target identification and lead optimisation, it will evolve into the role of augmenting human creativity. This will include supporting de novo discovery, experimental design and the cross-disciplinary thinking that generates genuinely novel hypotheses.
The researcher of the future will be collaborative and AI-empowered. And for drug development, that should result in shorter timelines.
What's the biggest opportunity – and risk – with AI in drug discovery?
The biggest opportunity is a fundamental change in the economics of drug discovery. The current model – billions of dollars, over a decade of work, and an 80-90 percent failure rate – means that huge areas of unmet need, particularly rare diseases, struggle to attract investment. AI could change that because if you can identify non-viable candidates sooner and run more cycles in less time, or connect biology and chemistry data to better understand rare diseases, previously uncommercial research becomes viable.
The biggest risk is overpromising. Anyone using AI daily knows it makes mistakes – confidently and in ways that are hard to spot. In drug discovery, a misidentified target could cost years. That's why tools that cite sources and are grounded in peer-reviewed data must become the expected standard.
Newsletters
Receive the latest pharmaceutical news, personalities, education, and career development – weekly to your inbox.
