In a recent ION Influencers Fireside Chat, Romain Clement, Founder of Arca Science, explored the hidden trends in life sciences. The conversation moved beyond typical biotech talk, delving into the raw, personal motivation behind using artificial intelligence to fix a broken system: the staggering 95% failure rate of drugs in clinical trials.

Clement’s journey is not one of a detached technologist but of a patient. Diagnosed with a brain tumor, he leveraged his access to top pharmaceutical minds to save his own life—and in the process, identified the critical inefficiencies plaguing the industry.

Here are the key insights from the discussion on how Arca Science is using “exact AI” to bridge the gap between the 250,000 drug candidates studied each year and the mere 200 that get approved.

1. The “Rabbit Hole”: Unstructured Data and the Cost of Knowledge

Clement’s journey began with a shocking discovery: in 2014, 80% of data in the pharmaceutical industry was unstructured. Even global heads of R&D at giants like Sanofi struggled to browse their own internal resources.

This “cost of knowledge” means that critical insights are often lost, buried in silos. The result? A massive disconnect between the potential of a drug and the patients who need it. Clement’s mission became clear: create a meeting ground where the interests of the industry (efficient drug approval) and the interests of patients (access to the right treatments) could finally align.

2. Decoding the 95% Failure Rate

Why do drugs fail? Clement broke down the brutal statistics of clinical development:

• 97.5% of candidates are discarded before the preclinical phase, often due to resource allocation and “bets” on specific compounds.

• Of those that make it to human trials, 93-95% fail.

• 50-60% fail due to lack of efficacy.

• 15-20% fail due to unmitigated safety issues (toxicity).

The core problem, according to Clement, is the inability to prospectively model the “benefit-risk” of a candidate for a specific patient group.

3. The AI Revolution: Why “Generative” Isn’t the Answer for Clinical Trials

While the world embraced ChatGPT, Clement warned of its limitations in high-stakes science. Generative AI (LLMs) relies on predicting the next plausible word or state, which inherently degrades reality. In clinical development, where a patient’s reaction cannot be simulated, hallucinations are lethal to accuracy. He noted that when asked for sources, LLMs provide only 3% of relevant sources, and 50-80% of those are often invented or inexact.

Arca Science’s Solution: “Exact AI”
To solve this, Arca Science built a different kind of system. Instead of one large language model, they deployed 24 specialized AI models, each with a hyper-narrow area of expertise. This system extracts, qualifies, and stratifies data, allowing users to audit every single step. The result is a platform with zero hallucinations—a non-negotiable feature for regulatory compliance and patient safety.

4. The “IBM Watson” Trauma and the ChatGPT Opportunity

Before the generative AI boom, the industry was burned by the “IBM Watson big disappointment,” making them skeptical of NLP (Natural Language Processing). When ChatGPT launched in 2022, it didn’t threaten Arca Science; it revitalized the conversation. It gave AI a positive image, allowing Clement to reopen doors at pharma giants and explain why his “exact” approach was critical where others failed.

5. The Moat: Not Technology, but Implementation

Can a startup survive if Google or OpenAI decide to enter this space? Clement argues that technology alone is not a durable moat. The true competitive advantage lies in deep implementation: understanding the client, the specific disease states (from multiple sclerosis to glioblastoma), and most importantly, the regulatory environment.

By building a system that is regulatory-compliant, auditable, comprehensive, and precise, Arca Science is positioning itself as the new standard. Their ultimate goal? To be validated and showcased by regulators like the EMA and FDA, effectively becoming the benchmark for benefit-risk evaluation.

6. The Future: World Models and the New Standard

Looking ahead to 2030, Clement is excited about new AI architectures like “World Models” (championed by Yann LeCun). He believes this is the next frontier for healthcare.

But the immediate milestone is clear: having Arca Science recognized as “the new standard for benefit-risk evaluation.” He envisions a world where regulators and pharma companies use his platform not just for speed, but for certainty—ensuring that no critical data is left out when deciding if a drug is safe and effective for a specific population.

Final Thought:
When asked what could go wrong, Clement remained optimistic. For him, success isn’t just about Arca Science. If a new drug reaches the market using his technology—even for a small patient population—it proves the model. “It can get exponential from that moment,” he said, highlighting that for patients with no other options, precision in drug development isn’t just a metric; it’s a lifeline.

Key timestamps:

00:07 Introduction
03:49 Mission of Arca Science
06:00 Candidate Selection in Clinical Trials
08:15 Ethical Dilemmas in Clinical Development
10:42 Human Intervention in AI Systems
13:17 Impact of Generative AI on Clinical Development
16:25 Partnerships with Major Pharmaceutical Companies
22:11 Future Milestones for Arca Science
27:01 Investor Expectations and Growth Strategy
28:16 Conclusion and Closing Remarks