Scientists from Stanford University in the US filled an impressive 386 pages with their latest AI annual report. Nine of the top ten AI research institutions are from China, followed by the US Massachusetts Institute of Technology in tenth place. For the ranking, the authors evaluated the number of AI-relevant publications between 2010 and 2021. However, universities are becoming less important in development: According to the AI Index Report, 35 major models for machine learning were published in 2022 – 32 by companies and only 3 by research institutions.
According to the AI Index Report, nine Chinese universities are among the top 10 AI research institutions.
(Bild: Stanford Institute for Human-Centered Artificial Intelligence (HAI))
In 2022, private investment in AI fell for the first time since 2013. Globally, they amounted to 91.1 billion US dollars, a decrease of 26.7 percent compared to 2021. A look at the past ten years shows a clear upward trend: in 2022 private investments in AI were 18 times as high like in 2013.
Politicians deal loudly AI Index Report more frequently with the topic, according to an analysis of the minutes of parliamentary debates in 81 countries. In it, the number of mentions of AI in global legislative processes increased by a factor of 6.5 between 2016 and 2022.
How much the population welcomes AI depends heavily on the country in which you ask. The report cites an IPSOS survey from 2022, according to which 78 percent of respondents in China agreed that AI products and services have more advantages than disadvantages. In none of the countries examined was approval higher. The researchers found the lowest rating in the USA: only 35 percent of the Americans surveyed agreed with the statement.
The abuse of AI is increasing rapidly. The nonprofit AIAAIC database documents such cases and served as a data basis for the researchers at Stanford. According to the report, the number of AI incidents and controversies has increased 26-fold since 2012.
On the impact of AI on the environment, the report reports on the work of Canadian Sasha Luccioni, who showed that the training run of the largest open AI language model, “Bloom”, causes 25 times as much carbon emissions as a single traveler on a plane trip from New York to San Francisco.
Due to the great progress that AI models have made recently, previous benchmarks will be useless in the future to map further increases. The report therefore also deals with new, more comprehensive benchmarking suites such as Google’s “BIG-bench” and the “Holistic Evaluation of Language Models” (HELM) developed at Stanford.
For the sixth time, the Stanford scientists have presented their AI Index Report. According to their own statement, they are pursuing the claim of being “the world’s most credible and comprehensive source for data on and insights into AI”. Both the complete report and the data used are available for download free of charge.
(Bild: c’t 10/2023 )
The language models learn to surf! AI search engines scour the web for you and link your sources. c’t tests seven search services with AI support and shows new security risks that arise as a result. The upcoming end of support for Windows 10 threatens to turn working computers into electronic waste, because many computers do not meet the requirements for Windows 11. We ask what politics and business actually have to say about this. We also tested mini PCs, explain how to bring back deleted files under Linux and remember the c’t “Hommingberger Gepardenforelle” campaign. You can read that and more in c’t 10/2023!
(dwi)
