2021 edition of 'Artificial Intelligence Index' that summarizes a huge amount of research data on AI is released



The Stanford Human-Centered AI Institute (HAI), which was founded by Lee Fei-Fei , a world authority on artificial intelligence (AI) research, has compiled a huge amount of research data on AI, ' Artificial Intelligence Index '. 2021 edition has been announced. The impact of the new coronavirus infection (COVID-19), which raged in 2020, on AI development has been analyzed from multiple perspectives.

The AI Index Report – Artificial Intelligence Index
https://aiindex.stanford.edu/report/



Annual index finds AI is'industrializing' but needs better metrics and testing | VentureBeat
https://venturebeat.com/2021/03/03/annual-index-finds-ai-is-industrializing-but-needs-better-metrics-and-testing/

The Artificial Intelligence Index is edited by a steering committee of 11 members of Stanford University's HAI and is a compilation of texts by contributors from organizations such as Harvard University, the Organization for Economic Co-operation and Development , Partnership on AI , and SRI International.

The 2021 edition of the Artificial Intelligence Index lists the following nine highlights of AI development in 2020.

1: Significant increase in investment in the AI field in drug development
The use of AI in fields such as drug development, cancer treatment, and new drug discovery is receiving more attention than ever, and in 2020, total AI-related private investment, M & A, and global investment increased by nearly 40% year-on-year. It turns out that. The total investment in private-sector AI-based drug development in 2020 was over 13.8 billion yen (about 1.5 trillion yen), which is 4.5 times that of 2019.

In fact, using AI to find an existing drug that is effective against another disease ...

AI will be developed to find out from existing medicines that are effective for other diseases --GIGAZINE



Efforts are being made to use AI to predict the development of breast cancer and to distinguish between those who respond to cancer immunotherapy and those who do not.

It is possible to distinguish 'people who work with cancer immunotherapy and those who do not' using AI --GIGAZINE


by DC_Studio

Meanwhile, funding for AI-related startups has declined for the third consecutive year.



2: Industry changes continue
In 2019, 65% of PhD graduates in North America entered the AI industry, an increase from 44.4% in 2010.



3: AI produces everything
AI-based systems are capable of automatically generating text, audio, and images at a sufficiently high level. Therefore, human beings have a hard time recognizing what AI has generated as 'AI-generated'.

For example, if you use 'This person does not exist' that can generate a fictitious human photo using AI, you can realize the high accuracy of the photo generated by AI.

'This person does not exist' that allows you to easily generate an image of a person who does not exist in this world with a single touch --GIGAZINE



In addition, there is a website that distinguishes between 'real human photos' and 'AI-generated fictitious human face photos' in a quiz format, and it is difficult to easily distinguish whether they are genuine or not using AI. You should be able to realize that you can generate a photo of.

'Which Face is Real?' Review to test whether you can see through real human and AI face images --GIGAZINE



However, it has also been pointed out that AI that can generate fictitious portraits is being abused.

AI that creates faces of non-existent cats and women is being abused by Twitter bots --GIGAZINE



Efforts in the audio field using AI are also being made, and web apps such as 'DrumBot ' that automatically plays drum parts to music have appeared, and rapper Travis Scott's music has been learned. AI 'Travis Bott' is releasing new songs.

AI who machine-learned the songs of professional rappers has announced a new song, and a PV that looks like an AI rapper is also being released --GIGAZINE



In the text field, the successor model ' GPT-3 ' of the text generation AI model 'GPT2 ', which was regarded as 'too dangerous' by the development team because it easily creates text with too high precision, was released in June 2020. It was published on Reddit, and it became a hot topic that I was interacting with humans on Reddit for a week without being noticed by AI.

It turns out that the sentence generation AI 'GPT-3' was talking with humans on Reddit for a week without being noticed by anyone --GIGAZINE



4: AI has diversity as an issue
On the other hand, among the doctoral degree graduates living in the United States in 2019, 45.6% of the races newly hired in the AI industry are Caucasian, 22.4% are Asian, 2.4% are African American, and 3.2% are Hispanic. , Non-Hispanic multi-racial was 1.6% and unknown was 24.8%.



In addition, a survey by the

Association for Computing Machinery revealed that only 18% of AI-related PhD graduates were women.



5: China overtakes America with citation of AI treatise
China's scientific research has undergone remarkable evolution, and it is reported that in December 2019, China overtook the United States in terms of the number of chemistry papers published.

China surpasses the United States in the ranking of contributions to the publication of chemical treatises, ranking first in the world, and Japan defending fourth place --GIGAZINE


by geralt

In addition, China has more citations than the United States. However, in the last 10 years, the United States has published considerably more AI-related papers than China, and it seems that the number of citations is by far the largest.

6: The majority of AI-related doctoral students in the United States are from overseas and stay in the United States after graduation.
The percentage of international students who have completed AI-related doctoral programs in North America continued to increase in 2019, reaching 64.3%. This figure is an increase of 4.3% from 2018. In addition, 81.8% of them stay in the United States after graduation, and only 8.6% are working outside the United States.



7: Surveillance technology is becoming more and more ubiquitous, fast and cheap
The technology required for large-scale surveillance is rapidly maturing. Technologies such as image classification, face recognition, video analysis, and voice recognition used for surveillance have all made great progress in 2020.

Face recognition technology, which is a representative technology related to surveillance, has evolved to a level that can be used not only for human faces but also for individual identification of wild animals and livestock.

AI that identifies the faces of bears and cows has appeared, promising in wildlife protection and livestock tracking-GIGAZINE


by Princess Lodges

Efforts to protect endangered primates with a high-precision face recognition system --GIGAZINE



On the other hand, there are big privacy concerns such as IBM's announcement of withdrawal from the face recognition market and the ban on the introduction of facial recognition systems in big cities in the United States.

Big cities in the United States ban face recognition systems one after another --GIGAZINE


by

Mike MacKenzie

8: AI ethics lack benchmarks and consensus
Many research groups have produced qualitative or normative outcomes in the area of AI ethics, but in this area they generally 'measure the relationship between broader social debate about technology development and the development of the technology itself. Or there is no benchmark to evaluate, 'the Artificial Intelligence Index said. In fact, there is a lot of debate about whether the development of unethical AI should be stopped, and after analyzing 500,000 blogs and 60,000 news articles, posts about AI ethics are related to AI. It turns out to be the most popular of the articles in.

In addition, as an article about AI ethics that made a noise in the industry in 2020, 'Roman's call on AI ethics' published jointly with Microsoft and IBM by Pope Francis of the Roman Catholic Church ...

What is the 'Call for AI Ethics' announced by the Pope in collaboration with Microsoft and IBM? --GIGAZINE


by Long Thiên

There are cases where Dr. Timnit Gebru, who was the technical leader of the ethical AI team at Google, was dismissed.

Google's dismissal of researchers on the ethical AI team has been blamed for 'unprecedented censorship,' with more than 1,200 Google employees signing protests-GIGAZINE


by Thomas Hawk

9: AI is in the spotlight of the US Congress
The 116th Parliament, which was held from January 2019 to January 2021, had 3 statements about AI compared to the 115th Parliament, which was held between January 2017 and January 2019. It has doubled.

Below is a graph summarizing changes in the number of AI and machine learning (ML) related references in Congress.



In addition, from the data provided by LinkedIn, the highest level of AI technology was adopted in Brazil, India, Canada, Singapore and South Africa from 2016 to 2020, and the number of stars in the GitHub repository makes it the most popular AI-related software library. The one with the highest value is 'TensorFlow', the number of AI-related papers published on arXiv increased from about 5,500 in 2015 to about 35,000 in 2020, Queer in AI 2020 According to a member survey of the library, about half of the respondents experienced harassment and sexual discrimination.

in Note, Posted by logu_ii