A Selection of the Most Representative Charts——Artificial Intelligence Index Report

Every year, Stanford University's human-centric Artificial Intelligence Institute (HAI) releases its AI Index. The 2022 AI Index is as impressive as ever, with 190 pages covering R&D, technical performance, ethics, policy, education, and economics.

It is worth noting that many of the trends reported in last year's 2021 Index are still present. For example, we are still living in a golden age of AI, with a growing number of publications. the AI job market remains global, and there remains a troubling gap between corporate awareness of AI risks and attempts to mitigate said risks.

Below is a selection of 12 of the most representative charts.

1. Decoupling of investments

The amount of money pouring into AI continues to be incredible, with the most noteworthy portion coming from private investors worldwide. This number soars from $46 billion in 2020 to $93.5 billion in 2021. The growth comes from an increase in the number of large funding rounds, with four rounds exceeding $500 million in 2020 and 15 in 2021. The report also notes that all of this money is going to fewer companies, as the number of newly funded startups has been declining since 2018. It's a great time to join an AI startup, but maybe not to find one yourself.

2. Complex U.S.-China relations

There are a lot of talks these days about the artificial intelligence race between the United States and China. "When you see all the news about geopolitical tensions, you'd think the amount of cooperation between these two countries would be decreasing," said Daniel Zhang, a policy fellow at Stanford University's HAI and editor-in-chief of this year's AI Index. Instead, he told IEEE Spectrum, "U.S.-China collaboration has been trending upward for the past 10 years." In terms of cross-country collaborations on publications, China and the U.S. produce more than twice as much output as China-UK collaborations.

3. Applying for a patent and obtaining a patent are two different things

China leads the world in patent applications; the report states that China accounts for 52% of global patent applications in 2021. But the U.S. dominates the number of patents granted, accounting for 40 percent of the global total. Zhang notes that patents are granted to prove that your patent is actually credible and useful, and says the situation is somewhat similar to that of publications and citations. While China leads in the number of publications, citations to publications, and conference publications, the U.S. still leads in citations to conference publications, suggesting that prominent papers by U.S. researchers continue to have a significant impact.

4. The plateau of computer vision?

The field of computer vision is evolving so rapidly that it's hard to keep up with the latest news. AI indices show that computer vision systems are very good at tasks involving still images. Object classification and facial recognition, for example, are also getting better at video tasks (such as activity classification).

But a relatively new benchmark shows the limits of what computer vision systems can do: they're good at recognizing things, but not at reasoning about what they see. The Visual Common Sense Reasoning Challenge, launched in 2018, asks AI systems to answer questions about images and explain their reasoning. For example, a picture shows people sitting at a table while a waiter approaches with a plate; the test asks why one of the seated people is pointing to the person across the table. The report notes that performance improvements have become increasingly insignificant in recent years, "suggesting that new techniques may need to be invented to significantly improve performance."

5. Artificial intelligence is not ready for law school

The field of natural language processing (NLP) began to flourish a few years later than computer vision, but it is in a somewhat similar position to computer vision. Benchmarks for tasks such as text summarization and basic reading comprehension show impressive results, with AI systems often outperforming humans. But when NLP systems must reason about what they read, they run into trouble.

This chart shows the performance of a benchmark test consisting of logical reasoning questions from the LSAT used as a law school entrance exam. While the NLP system performed well on the easier set of questions on the benchmark, the model that performed best on the harder set of questions was only 69% accurate. The researchers obtained similar results from a benchmark test that required NLP systems to draw conclusions from incomplete information. Inference remains at the forefront of artificial intelligence.

6. Tremendous interest in AI ethics

There is good news in the report: judging by the attendance at conferences such as the ACM Fairness, Accountability and Transparency Conference (FAccT) and the ethics-related workshops at NeurIPS, there is now a tremendous interest in AI ethics. For those who haven't heard of FAccT, the report notes that it was one of the first major conferences to focus on the analysis of algorithmic social technologies. This chart shows the growing industry participation in FAccT, which Chang sees as further good news. "The field has been dominated by academic researchers, but now we're seeing more private sector involvement. It's hard to guess what this participation means for how AI systems are designed and deployed within the industry, but it's a positive sign.", he said.

7. Detox： Ethical Issues

One of the big ethical issues in AI involves large language models, such as OpenAI's GPT-3. It has a very bad habit of generating text that is filled with every bias learned from its training data (the Internet). Several research groups (including OpenAI itself) are working on this toxic-language problem with new benchmarks to measure bias and detoxification schemes. But the figure above shows the results of running the language model GPT-2 through three different detox methods. All three methods impair the model's performance on a metric called perplexity (lower scores are better), with the worst impact on performance on texts involving African-American-aligned English and references to minority groups. As experts say, more research is needed.

8. Universities are crawling with CS students

The artificial intelligence pipeline has never been fuller. The latest figures from the Computing Research Association's annual survey, which collects data from more than 200 universities in North America, show that more than 31,000 undergraduates completed computer science degrees in 2020. That's an 11.6 percent increase over the 2019 figure.

9. Artificial intelligence needs women

The same survey looked at new PhDs in AI, and the results were disheartening. The percentage of new female AI and CS PhDs has only increased by a few percentage points over the last decade, at least in North America. This is actually a repeat of last year's coverage of the 2021 report, but everyone should keep talking about it until things change.

10. Artificial intelligence requires people of all racial backgrounds

Ditto on this point. the AI Index shows data for AI and CS PhDs on different charts, but they tell the same story. The AI field needs to get better at diversity long before people get their PhDs.

11. Lawmakers are looking at artificial intelligence

In 2021, there are more bills related to artificial intelligence than ever before. Of the 25 countries the AI Index has been following, Spain, the United Kingdom, and the United States lead the way, each passing three bills last year. The report also notes that in the United States, the three bills passed were among a whopping 130 bills introduced. It's not clear from the report whether most of these bills promote AI through public funding or create regulations to manage the risks that AI can pose. Zhang said it's a mixed bag and said HAI will release a more detailed analysis of global legislation in the coming year.

Related News

1、MediaTek, Qualcomm announce joining Russia sanctions

2、Automotive chips rose across the board!

3、Apple M1 Ultra -- The Technology Behind the Chip Interconnection

4、Foxconn Announces Investment of $9 Billion to Build A Chip Factory in Saudi Arabia

5、Japanese Companies Increase Investment in Power Semiconductors

报告下载地址：

https://aiindex.stanford.edu/wp-content/uploads/2022/03/2022-AI-Index-Report_Master.pdf