# A.I. Scientists Are Here. But Is Progress Accelerating? | EP 170 Source: https://www.youtube.com/watch?v=Dn4OWufAggk Recap page: https://rapidrecap.app/video/Dn4OWufAggk Generated: 2025-12-26T12:33:31.051+00:00 --- ## Quick Overview AI progress in drug discovery and core science is accelerating significantly, as demonstrated by the 2024 generation of AI models like those from Google DeepMind and Anthropic, which are already producing results previously expected years later, although the hype surrounding some claims still needs to be tempered with scientific rigor and long-term validation. **Key Points:** - Sam Rodrigues, co-founder and CEO of Future House & Edison Scientific, is the guest expert discussing AI's impact on scientific discovery, particularly in biology. - The current generation of AI models (like those from Google DeepMind and Anthropic) is producing results much faster than previously anticipated, with some achievements that were expected years out now occurring in 2024. - A key breakthrough discussed is AI's ability to generate novel scientific hypotheses, such as designing a new antibiotic that inhibits a virus or bacterium that previously had no known inhibitor. - The cost of running these complex AI experiments is high, with one model run costing $200 per prompt, necessitating careful prioritization by scientists. - Rodrigues notes that while AI excels at tasks like predicting protein structure (e.g., AlphaFold), the bottleneck is often the subsequent need for slow, expensive wet-lab validation, which requires patient scientists and significant funding. - The expectation is that AI will continue to accelerate scientific breakthroughs, potentially solving major diseases within the next decade or two, but this relies on continued progress in areas like BCI (Brain-Computer Interfaces) and automated lab work. ![Screenshot at 00:04: The two hosts, Kevin Roose \(left\) and Casey Newton \(right\), introduce the show 'Hard Fork' as Sam Rodrigues prepares to join the discussion on AI's impact on science.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-04.jpg) **Context:** This episode of the Hard Fork podcast features Kevin Roose and Casey Newton interviewing Sam Rodrigues, the co-founder and CEO of Future House and Edison Scientific. The discussion centers on the current state and future trajectory of AI, specifically focusing on how rapidly AI is advancing scientific discovery, particularly in biology and drug development, and whether the public hype matches the actual, verifiable scientific progress being made. ## Detailed Analysis The discussion begins with Kevin Roose expressing his obsession with separating AI hype from reality, particularly regarding claims that AI will solve major scientific problems like curing diseases or fixing climate change within the next decade. Sam Rodrigues, an expert with a PhD in physics from MIT who co-founded Future House and Edison Scientific, joins the conversation to provide a grounded perspective. Rodrigues points out that AI is already proving transformative in science, such as generating novel hypotheses for drug discovery, exemplified by an experiment where their AI model, Cosmos, designed an antibody that inhibited a virus that previously had no known inhibitor. However, he stresses that running these experiments is costly ($200 per prompt) and that the biggest bottleneck remains the slow, expensive process of wet-lab validation, which requires patient human scientists. Rodrigues notes that while large language models are advancing quickly, models like those from Google and Anthropic still require careful, systematic testing rather than simply trusting the AI's output. He contrasts this with the progress in areas like protein structure prediction (like AlphaFold) where demonstrable success exists. He suggests that the next major leap will involve AI that can not only generate hypotheses but also plan and execute the necessary physical experiments, leading to a faster scientific iteration cycle. Rodrigues concludes that while the progress is real and transformative, the timeline for solving major problems like aging is likely further out than some hyper-optimistic claims suggest, noting that the scientific community remains appropriately conservative with validation. ### Introduction of Guest and Topic - Sam Rodrigues, co-founder/CEO of Future House & Edison Scientific, joins to discuss separating AI hype from reality in scientific progress - Kevin Roose sets the stage by mentioning the hype around AI solving cancer and climate change. ### Cosmos AI Breakthrough - Rodrigues details how their AI, Cosmos, generated a novel antibody design that inhibited a virus in a petri dish, a significant scientific finding - This required running experiments on a large dataset of genetic variants and analyzing the results, which took significant computational effort. ### The Cost and Bottleneck of Validation - Rodrigues highlights that running these AI experiments is expensive ($200 per prompt) and that the main slowdown is the necessary wet-lab validation required by scientists before trusting the results - He notes that many scientists are still conservative about trusting AI predictions without rigorous testing. ### AI in Drug Discovery vs. General LLMs - The discussion contrasts specialized AI tools (like those used for protein folding or drug discovery) with general LLMs like GPT-7, suggesting specialized tools are currently more impactful in science - Rodrigues points out that simulating entire virtual cells is a major, distant goal, whereas current AI excels at specific tasks. ### Future Outlook and Hype Check - Rodrigues believes the acceleration of AI in science is real and will lead to rapid progress, but he tempers the hype by suggesting that massive, fundamental breakthroughs in areas like aging might still be decades away, despite the current excitement. ![Screenshot at 0:04: Kevin Roose introducing the podcast and identifying himself as a New York Times Tech Columnist.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-04.jpg) ![Screenshot at 0:05: Kevin Roose and Casey Newton sitting at the round table with laptops and microphones, setting the scene for the podcast interview.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-05.jpg) ![Screenshot at 0:12: Visual graphics overlaying the speakers, representing abstract data flow or computation related to the AI discussion.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-12.jpg) ![Screenshot at 0:20: Casey Newton smiles as Kevin Roose explains his current focus on separating AI hype from reality in scientific discovery.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-20.jpg) ![Screenshot at 0:47: Kevin Roose gestures widely while explaining how AI tools are being built to tackle complex scientific problems like climate change and disease.](https://ss.rapidrecap.app/screens/Dn4OWufAggk/00-00-47.jpg)