Google’s artificial intelligence energy use and DNA access for police

Google released per-query estimates for the energy, water and carbon cost of prompts sent to its Gemini artificial intelligence apps. The newsletter also examines the ethics of allowing police to search private DNA databases and other technology headlines.

Google published a report that for the first time provides per-query estimates of the resource costs of prompts to its Gemini artificial intelligence apps. The company said the median prompt consumes 0.24 watt-hours of electricity, roughly the energy used by a standard microwave for about one second. Google also provided average estimates for water consumption, about five drops per query, and for associated carbon emissions. The report includes detailed information on how the company calculated its figures and represents the most transparent per-prompt disclosure yet from a major tech firm, answering questions researchers raised after earlier reporting by MIT Technology Review found no such per-query data from leading companies.

Separately, the newsletter includes a first-person account from a writer who uploaded a DNA profile to the private genealogical database FamilyTreeDNA and consented to police searches of the data. The article recounts how forensic investigative genetic genealogy, or FIGG, helped identify the Golden State Killer in 2018 by matching relatives and building family trees, and notes that the technique has since helped solve hundreds of murders and sexual assaults. The author says the personal motive for sharing DNA was to challenge privacy concerns about genetic data and to provoke debate about how sacred or private DNA should be, and the piece originally appeared in the Checkup biotech newsletter.

The edition also highlights the emergence of new formats and debates in research and sport driven by artificial intelligence. An October conference, Agents4Science, will feature work researched, written and reviewed primarily by artificial intelligence and presented via text-to-speech, a model that some scientists applaud for discovery potential and others criticize for errors and limiting opportunities. The newsletter’s curated must-reads cover topics from high-level industry moves to policy and geopolitics, including reported talks between Elon Musk and Mark Zuckerberg about OpenAI, AI companies seeking real-world data, a state-backed Russian messenger app being preinstalled on phones, debates over funding for HIV programs, and the introduction of artificial intelligence judging at the 2023 gymnastics world championships, with advocates pointing to fairness and critics warning about loss of narrative judgment.

72

Impact Score

How biased training could deepen global divisions in artificial intelligence

Competing ideologies are shaping Artificial Intelligence models, from over-cautious debiasing in the West to religiously aligned systems in the Middle East and state-controlled outputs in China. The article urges transparency and verifiable data, potentially via blockchain, alongside convergent global regulation.

AMD ROCm software for artificial intelligence

AMD’s open ROCm stack targets artificial intelligence workloads on AMD GPUs with upstream framework support, extensive libraries, and scale-out tooling. The page aggregates models, partner case studies, and developer resources including containers and cloud access.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.