At Rise of AI 2025 in Berlin, Ramin Assadollahi, founder and CEO of ExB Group, argues that AI is about to transform longevity and preventive medicine. He frames the goal as reaching a personal 'longevity escape velocity', where each medical advance extends lifespan long enough to reach the next one, referencing Aubrey de Grey's idea that the first 1,000-year-old may be only ten years younger than the first 150-year-old. He organizes longevity research into four systems: genes, the microbiome, the central nervous system and the immune system, which he calls a post-processor for the other three. He explains that conventional drug development costs up to 2.5 billion dollars and 10-15 years per drug, with under 11% of candidates reaching phase one trials, and proposes that AI-driven digital twins and in-silico simulation, an approach he compares to 'vibe coding', could make drug development interactive and personalized. He surveys wearables such as implantable EEGs, glucose-and-hydration sensors and sweat monitors as new sources of fast personal data, and closes by calling for a meta-framework linking simulations across biological levels and for changed legislation and data access to support personalized medicine.
Key takeaways
- Assadollahi defines longevity escape velocity as each medical advance extending lifespan long enough to reach the next advance.
- He organizes aging biology into four systems: genes, microbiome, central nervous system and immune system, with the immune system as the key post-processor.
- Conventional drug development costs up to 2.5 billion dollars and 10-15 years, with under 11% of candidates reaching phase one trials.
- Only about 16 digital twins of human organs exist today, including a Siemens heart model needing 12 hours to simulate one heartbeat.
- He argues drug development needs an interactive, 'vibe coding'-style loop combining LLM-generated hypotheses with in-silico simulation of organs and cells.
Detailed notes
Longevity escape velocity
Assadollahi opens by reframing longevity not as avoiding death but as dying when one chooses, quoting Aubrey de Grey's claim that the first 1,000-year-old is likely only about 10 years younger than the first 150-year-old. He explains longevity escape velocity: every medical improvement should extend a person's remaining lifespan far enough to reach the next improvement. He lists the common longevity approaches: sleep, physical activity and nutrition (SPAN); supplements, citing Ray Kurzweil's reported 120 pills a day; body modification such as Brian Johnson's blood exchange with his son; and, without discussing it further, uploading personality and knowledge into an LLM copy of oneself.
Four systems that drive aging
He organizes the biology of aging into four systems: genes, which shape organ morphology and disease risk (for example the ApoE gene and Alzheimer's risk, identified through research on centenarians); the microbiome, roughly 100 trillion organisms split into neutral, beneficial and competitive groups, where competitive organisms increasingly dominate after about age 80 and weaken the immune system; the central nervous system, which shrinks and loses energy production with age, a process he connects to his own experience watching his father's seven-year decline with Alzheimer's; and the immune system, which he calls the most important of the four because it acts as a post-processor reacting to problems across the body, tied to the concept of 'inflammaging'.
Why drug development is broken, and digital twins as a fix
Assadollahi describes conventional drug development as costing 400-500 million dollars in the best case and around 2.5 billion dollars once failures are included, taking 10-15 years, with fewer than 11% of candidates reaching phase one because trials cannot easily target the right patient subgroups. He argues AI tools like 'the AI Scientist', an open-source system that generates and tests research ideas, are not yet used for drug discovery because biology cannot be fully simulated. He points to digital twins as the fix in progress: about 16 exist today, including a Siemens-built heart model that takes 12 hours of computation to simulate a single heartbeat at the cellular level, and argues hardware companies such as Dell, HP and Nvidia should invest in this space. He adds that personal genome sequencing now costs about 400 dollars and returns roughly 150 disease-predisposition reports, and highlights new fast-data wearables: Brainscape, an implantable 24/7 EEG he reviewed through a Swiss angel club; a Dexcom-Oura integration linking glucose and ring data; and sweat sensors reading metabolite data, alongside live AI monitoring of clinical ECG data that already alerts nurses in real time to save lives.
Vibe coding for drugs
Drawing an analogy to AI coding tools such as Bolt, Cursor and Replit, which mediate between a person's ideas and rigid, syntactically defined code, Assadollahi says drug development needs the same interactive loop: hypotheses about molecules or pathways tested in-silico against simulated organs and cellular twins, rather than waiting 10-15 years. He cites Google DeepMind's Nobel Prize-winning protein structure work, noting it covers structured molecular data but not the next level, systems biology, such as how rapamycin interacts with the mTOR pathway. He counts about 40 AI drug-discovery startups with 8 billion dollars in funding since 2010 and expects this to grow sharply following the LLM boom of the last three years, alongside a rising number of open healthcare LLMs (such as BioMistral and Meditron) and MCP servers now used for protein structure and clinical decision support.
What needs to change next
Assadollahi calls for four next steps: a meta-framework to link simulations across gene, cell and organ levels; more venture capital to fill gaps in the in-silico tooling landscape; broader access to healthcare insurers' existing patient data; and a legal shift so aging itself, not only named diseases, can be a target for drug registration and personalized medicine, given that current trials and approval processes assume drugs for large, uniform patient populations.
Q&A highlights
Asked whether digital twins predicting lifespan could damage people's joy in living, Assadollahi says there are two open questions: whether the concept of death gives life meaning, noting his own parents chose to let go peacefully, and how social security systems would cope if lifespans broadly doubled. Asked about Germany's electronic patient record, he says new health-data-usage-law initiatives (the exact German term is unintelligible in the transcript) and permission for clinics to use cloud infrastructure are starting to make research access more workable. On ultra-processed food, he explains it is molecularly smaller than whole food, which changes how it moves through the intestines and disrupts the microbiome, and says Europe's food-labeling debates on sugar and fat have been weakened by industry lobbying compared to countries like Singapore. Asked how a digital twin can be built when so little of the brain is understood, he says the Siemens-style twins do not model cognition but reproduce known structural, genetically defined patterns such as neuronal strand organization for pharmacological use, not a full simulation of a person.
Quotes
Ideally I want to die when I want to die.
Ramin Assadollahi
I think we need vibe coding for drugs.
Ramin Assadollahi
This revolution is just ahead of us, it's very very close from my perspective.
Ramin Assadollahi
Questions answered in this talk
What is longevity escape velocity?
According to Ramin Assadollahi at Rise of AI 2025, it means each medical improvement should extend a person's remaining lifespan far enough to reach the next improvement, citing Aubrey de Grey's claim that the first 1,000-year-old may be only about 10 years younger than the first 150-year-old.
Why is drug development so slow and expensive?
Assadollahi states that a new drug costs 400-500 million dollars in the best case and around 2.5 billion dollars once failures are counted, takes 10-15 years, and that fewer than 11% of candidates even reach phase one clinical trials.
What are digital twins in longevity and drug research?
He describes them as computer simulations of human biological systems, such as a Siemens-built cellular-level heart model that needs 12 hours to simulate one heartbeat, and notes only around 16 such digital twins exist as of the talk.