AI in Pharmaceutical Development: A Lifesaver or an Expensive Toy? A Deep Dive into the Current Reality
Hello everyone, I'm your financial journalist. Today, we're going to discuss a topic that sounds quite advanced, but it's closely related to everyone's health: Can AI really help us develop better drugs faster and more cheaply?
Recently, at the 2026 Zhangjiang Pharmaceutical Valley Conference, something significant happened: Rentosertib, the world's first drug entirely developed by AI and advanced to the final stage of clinical trials (phase three), welcomed its first patient. It sounds like a scene from a science fiction movie coming to life, but the story behind it is filled with anxiety, debate, and careful practical considerations.
In simple terms, AI has indeed made the process of finding new drugs faster and cheaper, but testing the drugs remains an incredibly challenging task. Below, I'll break down this complex industry news into five key points to help you understand the excitement and the realities of AI in pharmaceutical development.
---
1. The Current Reality: AI in Drug Development is Entering the “Deep Water” – Even Nobel Laureates Are Nervous
Previously, we thought AI in drug development was still in the conceptual stage, but now it's actually at the critical point of implementation.
Key Fact:
A drug developed by Insilico Medicine for the treatment of pulmonary fibrosis has entered phase three clinical trials. This is the first drug in the world to be developed entirely using generative AI.
Why Are People Nervous?
Even Michael Levitt, the Nobel laureate in Chemistry in 2013, admitted, “Everyone in the company is a bit nervous.” Why? Because phase three clinical trials represent the ultimate test of drug development.
- The First Two Steps (Discovering the Drug): AI performs exceptionally well, acting like an extremely efficient filter.
- The Final Step (Testing the Drug): This is the stage where real money is invested, and it's a matter of life and death. If phase three fails, the savings made earlier might not even cover the losses.
Plain Language Explanation:
Think of AI as a genius chef who can quickly design a perfect recipe (molecular structure) and even prepare the ingredients neatly (pre-clinical testing). However, to determine whether the dish is tasty and safe for consumption, we need 100 real diners (patients) to try it for a year or longer. Now, as this “dish” is being served to the first diner, everyone is holding their breath.
---
2. Advantages: AI Makes the Drug Discovery Process Much Faster and Cheaper (90% Savings)
Although the final step is difficult, AI shows remarkable efficiency in the earlier stages of drug discovery.
Data Comparison (Traditional vs AI):
- Identifying Targets (Determining What to Treat): Traditional methods take 1 year and cost $94 million; AI takes 2 months and $2 million.
- Finding Potential Compounds (Initial Screening): Traditional methods take 1.5 years and $166 million; AI takes 4 months and $4 million.
- Optimizing Lead Compounds (Refining the Formula): Traditional methods take 2 years and $414 million; AI takes 11 months and $2 million.
Conclusion:
In the early stages of drug discovery, AI reduces costs by more than 90% and significantly shortens the time required.
Plain Language Explanation:
Traditional drug development is like searching for a needle in a haystack, both time-consuming and costly. AI, on the other hand, acts like a system with advanced sonar and navigation, directly locating the needle and even pulling it out. In this step, AI is clearly the “king of efficiency.”
---
3. The Challenge: The Huge Gap Between “Atoms” and “Humans”
If AI is so powerful, why can't it predict the results of phase three trials and eliminate the need for expensive clinical trials? This is the biggest technical hurdle at the moment.
Core Problem: The Gap in Scale
Nobel laureate Levitt pointed out that we have excellent models for simulating atoms, electrons, and proteins, but we don’t have good models for simulating humans.
- Molecules in computers are standard and pure.
- Real patients are complex and unique, with differences in genes, physical conditions, and lifestyles.
Industry Debate: Is AI Useful in Clinical Settings?
- Pessimists: Some say AI is useless in clinical settings because humans are too complex.
- Rationalists (Huang Xiaolu): AI isn’t useless; it’s just not yet effectively applied. The problem is that AI communicates in a “machine language” (vectors and scores), while doctors use clinical language (decisions about patient selection, treatment outcomes, etc.). There’s a significant semantic gap between the two.
Plain Language Explanation:
AI is like an engineer skilled in physics who can calculate how much a bridge can support theoretically. However, when real people with different personalities, weights, and walking styles walk on the bridge, it’s hard for AI to accurately predict who might fall. Currently, AI lacks the ability to perfectly simulate the behavior of a real human population.
---
4. The Risk: Phase Two is a “Life-and-Death” Moment – AI Needs to Help Companies Make Cost-Saving Decisions
There’s a critical phase in drug development called the transition from phase two to phase three, which is both the most expensive and the most risky.
Challenging Statistics:
- The success rate of transitioning from phase two to phase three is only 28.9%.
- The cost of a single phase three trial can range from $19 million to $127 million.
- If phase three fails, all previous investments are likely lost.
**AI’s New Role: From “Discoverer” to “Decision Advisor”
Huang Xiaolu from Wuxi Natural Constant Technology suggests that AI’s role should be to help companies make decisions at the end of phase two: should the drug proceed to phase three? If AI can predict with high accuracy that the drug will fail, companies can avoid billions in unnecessary expenses. This is known as “In Silico Second Opinion” (a computational approach to support decision-making).
Plain Language Explanation:
It’s like starting a business. AI helps you quickly create a prototype (phase two), and now you need to decide whether to invest tens of millions in large-scale trials (phase three) or give up. AI’s role is to analyze the data and tell you, “Based on historical data and models, the project has a 70% chance of failure; it’s recommended to stop.” At this stage, companies are most willing to spend money on this kind of “ certainty.”
---
5. The Future: Breaking Down Data Barriers – Whoever Controls the Data Wins
Finally, we need to consider the deeper challenges the industry faces: homogenization and data barriers.
Two Major Issues:
1. Model Homogenization: Many companies use the same open-source AI models, resulting in similar drugs with little differentiation.
2. Data Barriers: This is the most critical issue. AI needs data to learn, but companies are reluctant to share their “failure data” (the reasons why drugs didn’t succeed).
- Success data is shared willingly.
- Failure data is closely guarded, as it contains valuable business insights.
Who Will Win in the Future?
Tian Zhenglong from Shanghai Pharmaceutical suggests that the company that breaks down these data barriers and creates a private database containing both successful and failed cases, allowing AI to form a closed-loop feedback system, will gain an irreplaceable advantage.
Plain Language Explanation:
Currently, AI in drug development is like using a standard cooking app. While the app is useful, the dishes made with it all look similar. The future competition won’t be about which app is better, but about who owns the exclusive recipes and data. If you can collect data on failed projects that others are hesitant to share and use it to train AI, your AI will be better at identifying issues and developing more effective drugs.
---
Summary
AI is reshaping pharmaceutical development, but it’s not magic; it’s a powerful tool.
- It excels at: Quickly and cost-effectively discovering potential drugs (the first two stages).
- It struggles with: Accurately predicting human responses (phase three trials) and making complex clinical decisions.
- Key Breakthroughs: Overcoming the gap between “machine language” and “clinical language” and breaking down data barriers.
For the general public, this means more new drugs may become available at lower prices in the future. For investors and industry professionals, don’t blindly adore AI; focus on who can solve the problem of creating a complete data loop from scratch. The “first half” of AI in drug development (cost reduction and efficiency improvement) is already underway, and now we’re entering the “second half” (increasing success rates and decision-making accuracy). That’s where the real challenges lie.