The Biggest Artificial Intelligence Trends, According to Stanford

If you break down artificial intelligence by numbers and charts, it’s not that scary. This week, the Stanford Institute for Human-Centered Artificial Intelligence (HAI) released its 2023 AI Index, which offers a “snapshot of what happened this past year in AI R&D, education, policy, hiring and more” from a set of academic, private and non-profit organizations.

Some of the results aren’t surprising: the current language models are surpassing the current set of technical performance benchmarks, the bigger models emit large carbon emissions, China is all in on industrial robots, etc. However, a few surprising numbers appeared.

Italy Temporarily Bans ChatGPT (and Other Countries May Follow)

Concerns over personal data collection were cited for the block

The big takeaways:

It’s getting more expensive to train large language models: GPT-2 cost $50,000 to train three years ago, but last year PaLM cost $8 million.
There’s been an uptick in AI-related controversies. The examples Stanford noted include a deepfake of Ukrainian President Volodymyr Zelenskyy surrendering, face recognition technology used to try to track gang members (and rate their risk) and surveillance technology that attempts to determine the emotional states of students in a classroom.
There’s been a huge increase in job postings seeking AI skills across several sectors, with “Information” as the highest category (5.3% of all job postings in that category).
Rather shockingly, corporate investment (mergers/acquisitions, minority stakes, private investment and public offerings) in AI dipped in 2022 from 2021 highs, though investments did increase 13-fold in the last decade.
The United States passed the most laws (nine) regarding AI in the last year, though it really doesn’t feel like it.

The best news is that AI has shown some decidedly positive effects in the field of health and science. “AI models are starting to rapidly accelerate scientific progress and in 2022 were used to aid hydrogen fusion, improve the efficiency of matrix manipulation, and generate new antibodies,” as the paper notes.

The most interesting statistic, however, is a possible domestic concern and proof that our worries about AI persist. Americans simply aren’t trusting artificial intelligence (and least not yet) — only 35% of those sampled in this country agreed that “products and services using AI had more benefits than drawbacks.” The support was more than twice as much in China, Saudi Arabia and India.