THE AI INDEX REPORT

The 2022 Artificial Intelligence Index from Stanford University is now available. The fifth edition contains 230 highly informative pages organized into five chapters (Research and Development, Technical Performance, Technical AI Ethics, Economy and Education, and AI Policy and Governance). Worth reading or scanning through the report. Most of the data is also available on a GoogleDrive for own analyses. Here are my three key takeaways:

Massive increase in private investment

Let’s talk about money first. The total private investment in AI reached $93.5 billion in 2021 and more than doubled compared to 2020. The total number of newly funded AI companies continues to drop (746 companies in 2021 compared to almost 1200 during the peak in 2018. The number of small investments (below $50 million) remained virtually constant, but significant investments strongly increased. This leads to a concentration of investment. 40% of the newly funded companies are located in the US, where most investments take place. Despite the fact, the US only contributed less than 14% to AI Journal publications, followed by the EU (19%) and China (31%).

Training gets faster and cheaper

The training time for image classification dropped 27x since the introduction by MLPerf in 2018. Even more impressive is the decrease in training cost to reach 93% accuracy (ImageNet). Costs of over $1000 in 2017 dropped to $4.59 in 2021. This is a factor of more than 220 in only four years.

Image generation is further improving

The Frechet Inception Distance based on the STL-10 dataset of artificially generated images drops from over 35 in 2018 to 7.7 in 2021. For CIFAR-10 images, the FID is down to 2.1 from close to 30 in 2017. A score of 0 would mean that a fake image is identical to a real one. Luckily the deepfake detection of Face2Face or DeepFake also increased and surpassed 99% in 2021.

There is many, many more valuable information and data in this report. Grab your favorite beverage and read through it :)