Steve Hsu, Professor of Theoretical Physics, discusses polygenic embryo selection, phenotyping through AI, IVF, eugenic public policy, addressing concerns over reproductive technology, eliminating hallucination in LLMs, introspection for LLM error correction, and surviving cancellation attempt at MSU.
Genomic Prediction offers genetic testing of embryos, helping parents make informed decisions about embryo selection based on comprehensive information about genetic risks associated with various health conditions.
Genomic Prediction aims to improve the accuracy of genetic predictions by accessing more diverse and abundant data sets beyond European populations, enabling better predictors for traits like cognitive ability.
Genomic Prediction's offerings extend beyond reproductive medicine, with potential applications in estimating disease risks for the general population and supporting personalized medicine.
Deep dives
Genomic Prediction: Advancements in Genetic Testing of Embryos
Genomic Prediction is a tech startup specializing in genetic testing of embryos. As a standard practice in in-vitro fertilization (IVF), about 70% of embryos produced in the US undergo some form of genetic screening. The company aims to provide parents with comprehensive information about the genetic risks associated with their embryos. By analyzing the genome, they can assess the risk of various health conditions, including diabetes, breast cancer, heart disease, and schizophrenia. This information helps parents make informed decisions about embryo selection, contributing to a healthier outcome for their future child.
The Role of Data Availability in Genomic Prediction
Genomic Prediction emphasizes that the availability of data is the primary determinant in improving the accuracy of their genetic testing. While they have made significant progress in solving key genetic traits such as height for European populations, they acknowledge the need for more diverse and abundant data sets to further develop predictors for traits like cognitive ability. The company believes that, with access to more comprehensive data on populations beyond Europeans, they can improve the accuracy of genetic predictions for a wider range of traits, benefiting a broader population.
Expanding Applications of Genomic Prediction
Currently, Genomic Prediction's offerings predominantly focus on assessing the genetic risks associated with various health conditions. By analyzing an individual's genome, they generate an overall health index that predicts life expectancy. This data aids in making informed choices during embryo selection, looking to maximize health outcomes. Additionally, there is potential for genomic prediction to expand its scope beyond reproductive medicine. The technology could be utilized widely, such as in estimating disease risks for the general population or supporting personalized medicine. Despite concerns over eugenics and the potential for classism, this technology has immense value in providing individuals with comprehensive genetic information and enabling better decision-making regarding their health and future.
The Ethics of Genetic Manipulation and its Potential to Exacerbate Class Differences
The podcast episode delves into the ethical implications of genetic manipulation and its potential to exacerbate class differences. The speaker argues that advancements in reproductive technologies, such as IVF and embryo selection, may be considered as milder forms of eugenics. However, he defends their implementation on moral grounds and suggests that these technologies have faced initial opposition in the past but eventually become normalized. On the topic of class differences, the speaker acknowledges that initially, these technologies may be accessible only to the wealthy. However, as with other technologies like early computers and mobile phones, over time, their costs are likely to decrease, making them more widely available to everyone. The speaker also expresses a willingness to support a system where egg freezing and IVF are incorporated into the national healthcare system, as it would prevent the creation of inequality by making these technologies more accessible to all.
Challenges of Collecting Behavioral and Cognitive Phenotype Data for Genetic Research
In the podcast episode, the speaker highlights the challenges of collecting behavioral and cognitive phenotype data for use in genetic research. They explain that while there is abundant data available for studying physical health conditions, the same is not true for cognitive traits or behavioral traits like the big five personality scores. The speaker attributes this lack of data collection to a combination of factors, including privacy laws and a reluctance among scientists to collect such data due to potential criticism and controversy. They even suggest that in the coming years, East Asian countries may make more progress in collecting large-scale cognitive phenotype cohorts. Overall, the episode emphasizes the limited availability of comprehensive population-wide health data for cognitive and behavioral traits, hindering advancements in the field of genetic research.
Steve Hsu is Professor of Theoretical Physics at Michigan State University, as well as the founder of Genomic Prediction and SuperFocus AI; he also hosts the Manifold podcast, and the Information Processing blog.
Steve and I speak about polygenic risk scoring and embryo selection, using AI to predict phenotype from genotype, in-vitro fertilization (IVF), egg freezing, eugenic public policy, addressing Christians' and right-wing traditionalists' concerns over reproductive technology, SuperFocus AI's plan to eliminate hallucination in large language models (LLMs) by separating memory from inference, introspection for LLM error correction, and surviving the failed cancellation attempt at MSU.
