NOTE
🇬🇧 English (AI-translated). Original in Thai: 🇹🇭 อ่านภาษาไทย
Speaker: Dr. Piyarit Ittichaiwong
Physician, Siriraj Data Innovation Center (SiData+), Faculty of Medicine Siriraj Hospital, Mahidol University
Event: Techsauce Healthspan Festival 2026
AI is Approaching Human-Level and AGI
- AI currently possesses human-like capabilities in digital roles, particularly in white-collar jobs which could decrease by 10% in 2-3 years.
- AGI (Artificial General Intelligence) may arrive this year or next, according to predictions by Dario (CEO of Anthropic) and Chinese experts.
- Frontier AI models are evolving from “tools” to “co-workers” that can actively assist in work.
AI Capabilities in Medicine
- GPT-4o scored 86% in MedQA (an exam similar to USMLE), placing in the 98th-99th percentile.
- However, on a “real-world” benchmark like HelmetChart, the older GPT-4o model scored only 0%, whereas o3 scored 31.6% and GPT-5 scored 46.2% (physicians average around 40%).
- GPT-5.2-5.4 can outperform specialists (radiologists, specialists) in several tasks, indicating that these models perform at least as well as general practitioners today.
Model Development and Costs
- Training models with data from books and public records is reaching its limit, necessitating higher-quality data.
- AI prices have dropped significantly, from $1 to 1 Baht (about a 30-fold reduction), making the use of multi-agent systems feasible.
- Chinese models (Zhimi, Jiaoyue) offer top-tier quality at a much lower price.
Multi-Agent AI Systems
- Current AI is shifting from single chatbots to multi-agent systems that collaborate like a medical team.
- An orchestrator AI (analogous to a general practitioner) coordinates with various specialist agents.
- They must be explainable and transparent to allow for verification and auditing.
Achievements of Thai Teams
- A team from Siriraj, Bangkok Hospital, and Carely (a subsidiary of PTTEP and Primes) competed against 300+ teams and won 3 awards in a competition organized by Harvard and other leading universities, defeating teams from Carnegie Mellon and others.
- The Thai team ranked 4th globally in a competition organized by ACL (a premier NLP conference) for building an AI that reads X-ray images and generates reports, losing only to Meta due to having many times fewer GPUs.
- AI applications developed by the Thai team are being used in at least 30 hospitals and are currently expanding.
AI Applications in Medicine
- 81% of physicians use AI, such as large language models and search tools, and 88% show responsibility in its usage.
- 70% believe AI will help reduce administrative paperwork, which doctors are not trained to do.
- AI assists in image diagnosis (such as X-rays) and generates reports, helping general practitioners in rural provinces.
- Medical AI (such as Claude AI) achieves 90-93% accuracy, comparable to specialists.
- Google DeepMind and Insilico Medicine use AI for drug design, reducing the timeline from several years to just a few months.
Challenges and Limitations
- AI still suffers from under-triage (under-triaging cases by 50-60%) because it does not ask follow-up questions like a human would.
- Even though AI performs well in demonstrating empathy, it remains uncertain whether it is genuine empathy.
- Regulatory and legal issues still need to be addressed.
Strengths and Opportunities for Thailand
- Thailand has a massive volume of data (Thai physicians examine 60+ patients daily compared to 10-15 in the UK/US), providing high-quality data for training models.
- Thai teams possess world-class capabilities but lack GPUs and resources.
- The future will bring multi-modal and multi-lingual AI, where Thai language data will be a key strength.
Future Directions
- Healthcare and wellness will be a major theme, as once everything else is secured, what people value most is life and health.
- Thailand has the potential to become a leader in health and wellness, with government support.
- The medical market will become more flexible and tech-integrated, no longer operating as separate silos.
- Collaboration between humans and AI will yield the best outcomes.
- Inviting students and Thai people worldwide to return and collaborate to help Thailand become a leader in health and wellness.