Artificial intelligence holds great promise for transforming the pharmaceutical industry, yet it also faces limitations that temper expectations. As Julien Willard, an expert on AI applications in pharma, explains, effectively harnessing AI requires focusing on highest-value opportunities while establishing proper governance to ensure models remain accurate and explainable over time.
Cautious Optimism Around AI
Julien expresses cautious optimism about AI’s potential. “I’m very excited about artificial intelligence in the pharmaceutical industry,” he shares. “I believe it is both hyped but also genuinely promising.”
Many pharma executives Julien works with agree AI brings huge potential but struggle to demonstrate solid ROI so far. “It is fair to say that in the near future, many players in pharma will start seeing diminishing returns from generative AI,” Julien notes.
Current AI models generalize well within their training datasets but hit walls extending beyond them. Julien worries the rush for AI proof-of-concepts distracts from focusing on business value and acknowledging tech limitations.
AI’s Severe Limitations
Julien highlighted several severe limitations of today’s AI systems:
- Brittleness: LLMs are not the magic solution some believe them to be. They have severe limitations. They are brittle.
- Fabrication: LLMs often can fabulate because they can’t tell truth from falsehood.
- Lack of reasoning: They lack reliable reasoning.
- No fact checking: They cannot fact check their output.
Because of these flaws, LLMs generate false and true statements through the same faulty process. “Everything LLMs generate, whether true or false, comes from the same process, leading to inevitable errors even with perfect data,” Julien said.
The Data Bottleneck
Recent research reveals data limitations more severely constrain AI than initially thought. “AI model performance improves linearly with exponentially more data,” Julien notes.
This means even tiny AI performance gains require massive data growth. So, models fail to generalize outside their training data. “However, if we look at the pharmaceutical business, there are several areas with significant potential for AI,” Julien continues.
Drug Discovery Breakthroughs
One high-potential area is drug discovery. Previously, running high-precision experiments across multiple targets was extremely difficult. Experiments were limited to single targets in silico.
Now, thanks to AI, it’s possible to run simultaneous experiments on multiple targets and collect higher resolution data. “This is a huge breakthrough for target discovery,” Julien emphasizes.
While biological knowledge and data remain limited, rapid advances on both fronts converge to enable major AI impact.
Optimizing Clinical Trials
Another valuable application is optimizing product development and clinical trial design. This includes selecting optimal trial sites, stratifying patient populations, and boosting enrollment. AI can also assist with regulatory documentation and transforming medical affairs to boost productivity.
Implementing AI Governance
Julien advises clients to remain focused on high-value AI applications and ensure proper governance from the start.
He warns model performance inevitably degrades over time. “We know models drift, meaning prediction accuracy worsens compared to performance during training,” Julien notes. Sustaining benefits requires ongoing processes to keep models up-to-date, explainable, and well-governed. AI is not a one-and-done exercise.
Moving Forward With Caution
AI holds tremendous potential to accelerate pharmaceutical advances, but Julien cautions against getting distracted by hype. Focusing on proven high-value applications and robust governance will ensure pharma companies harness AI’s power responsibly and effectively.
With a measured, value-driven approach, AI can enable breakthroughs from drug discovery to clinical trials. But unsupported promises of magic solutions must be viewed realistically. By acknowledging limitations and mitigating risks, pharma leaders can maximize real-world benefits.
“My advice is to focus on highest business value opportunities, avoid distractions, and ensure proper governance from day one,” Julien concludes. With the right precautions and priorities, AI’s pharmaceutical promise can become reality.
To learn more about Julien Willard, visit his LinkedIn profile here or his website.
