https://hai.stanford.edu/news/2023-state-ai-14-charts
Image
The 2023 AI Index is out, covering the world of artificial intelligence from technical performance achievements, ethics advances, education and policy trends to economic impact, R&D, and the hiring and jobs scene.
The AI Index is an independent initiative at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), led by the AI Index Steering Committee, an interdisciplinary group of experts from across academia and industry. It tracks, collates, distills, and visualizes data relating to artificial intelligence, enabling decision-makers to take meaningful action to advance AI responsibly and ethically with humans in mind.
TL;DR? Here, learn about the state of AI in 14 charts.
Large language models keep scaling in size and expense. GPT-2, released in 2019 and considered the first large language model, had 1.5 billion parameters and cost an estimated $50,000 to train. Just three years later, PaLM launched with 540 billion parameters and cost an estimated $8 million. It’s not just PaLM: Across the board, large language and multimodal models are becoming larger and pricier. (And since these are estimates, we've qualified them as mid, high, or low: mid where the estimate is thought to be a mid-level estimate, high where it is thought to be an overestimate, and low where it is thought to be an underestimate.)
On the technical side, current AI tools keep meeting or beating benchmarks. While we saw benchmark saturation last year, this year the trend is much more pronounced. This shows us AI systems have become increasingly capable on older benchmarks and will require more difficult tests to be fully challenged. (Learn more about benchmark saturation from AI Index steering committee member Vanessa Parli.)
Big models emit big carbon emissions numbers – through large numbers of parameters in the models, power usage effectiveness of data centers, and even grid efficiency. The heaviest carbon emitter by far was GPT-3, but even the relatively more efficient BLOOM took 433 MWh of power to train, which would be enough to power the average American home for 41 years.
According to the AI, Algorithmic, and Automation Incidents and Controversies repository, reported issues are 26 times greater in 2021 than in 2012. Chalk that up to both an increase in AI use and a growing awareness of its misuse. Some of those reported issues included a deepfake of Ukrainian President Volodymyr Zelenskyy surrendering, face recognition technology to try to track gang members and rate their risk, and surveillance technology to scan and determine emotional states of students in a classroom.