The 2019 AI Index Report Gives Us a Glimpse Into AI Progress
Wed, April 14, 2021

The 2019 AI Index Report Gives Us a Glimpse Into AI Progress

Annual reports about artificial intelligence keep us updated from all of the innovations and improvements it has in a particular year. These reports allow us to see how it expanded and continued to drive success across all industries / Photo by: taa22 via 123RF

 

Annual reports about artificial intelligence keep us updated from all of the innovations and improvements it has in a particular year. These reports allow us to see how it expanded and continued to drive success across all industries. It’s important for people, especially those who are investing in AI, to decide if AI will still worth be the shot in the next coming years.

At the same time, these reports keep us on track, helping experts determine AI still provides the same benefits as before or it only confuses us. A great example of this is the AI Index, an annual report conceived within the One Hundred Year Study on AI (AI100) and now hosted at Stanford’s Human-Centered AI Institute (HAI). The report aims to ground the conversation about AI and its progress in data. 

In Stanford University’s HAI’s official website, it stated that the report provides unbiased, rigorous, and comprehensive data to develop a deeper understanding of the complex field of AI. It consists of nine chapters: research and development, conferences, technical performances, economy, education, autonomous systems, public perception, societal considerations, and National Strategies and Global AI Vibrancy.

The AI Index uses the Global AI Vibrancy Tool which compares 28 countries’ global activities across 34 indicators. Aside from that, the report has an arXiv Monitor that helps people conduct their own research into current technological progress in AI. This year’s report has volumes of relevant data on the economic, technical, and societal progress of AI. Here are some highlights:

Interest in AI Continues to Increase

The interest in AI is driven by its advancements, which continue to provide us with benefits. The AI Index reported that the attendance at AI conferences continues to increase. For instance, NeurIPS, a machine learning and computational neuroscience conference held every December, expected over 13,500 last 2019 – a 41% increase since 2018 and 800% since 2012. Other AI conferences like AAAI and CVRP have also seen around 30% increases in attendance.

The time it takes to train AI models has also significantly decreased due to the advancements of technology. For instance, an image classification system that usually takes three hours to train in October 2017 took about 88 seconds to train in July 2019. The cost of training a similar system also decreased in that same period. The demand for AI jobs significantly increased, particularly in the tech and manufacturing sectors. In 2012, AI jobs only made up 0.3% of the industry. Last year, it made up 0.8% of the industry. 

The AI Index reported that the interest in AI research has skyrocketed. The volume of peer-reviewed papers between 1998 and 2018 increased by about 300%. The share of published AI papers by 2018 increased three-fold in the last 20 years. The subject currently accounts for about 9% of all conference papers and 3% of peer-reviewed publications. 

The interest in AI is driven by its advancements, which continue to provide us with benefits. The AI Index reported that the attendance at AI conferences continues to increase / Photo by: Wu Bingfeng via 123RF

 

According to Analytics Insight, an online site that monitors developments, recognition, and achievements made by AI, big data and analytics companies across the globe, China publishes more AI papers than any other nation. In 2018, Chinese government institutions produced almost three times more AI papers than Chinese corporations. 

At the same time, it has seen a 300-fold increase in government-affiliated AI papers since 1998. Meanwhile, corporate AI papers increased by 66-fold in the same period. However, it was found that the US a greater impact on those papers compared to other countries. The report showed that US authors were cited 40% more than the global average.

Additionally, investments in AI startups are increasing. In 2010, $1.3 billion was raised for AI startups in 2010, and in 2018, $40.4 billion was raised. The researchers stated that the average annual growth rate during those years is more than 48%. It was also reported that the US has invested more in private AI than any other country, with nearly $12 billion. The country is followed by China with $6.8 billion. 

AI is Outpacing Moore’s Law

While the rapid advancements of AI come as no surprise, this is not normal, according to Moore’s Law. Moore's Law refers to the perception that technology’s speed and capability increase every couple of years, which means that application developers expect a doubling in application performance for the same hardware cost. However, the AI Index showed that AI computational power is accelerating faster than traditional processor development. 

“Prior to 2012, AI results closely tracked Moore’s Law, with compute doubling every two years. Post-2012, compute has been doubling every 3.4 months,” the report said. 

According to ComputerWeekly.com, a digital magazine and website for IT professionals in the UK, the researchers explored how far computer vision had progressed. They looked looking at innovative algorithms that push the limits of automatic activity understanding. Using the ActivityNet Challenge, the team recognized human actions and activities from videos. It showed that algorithms have massively improved as it can accurately recognize hundreds of complex human activities in real-time. However, more work is needed. 

“After organizing the International Activity Recognition Challenge (ActivityNet) for the last four years, we observe that more research is needed to develop methods that can reliably discriminate activities, which involve fine-grained motions and/or subtle patterns in motion cues, objects and human-object interactions,” Bernard Ghanem, associate professor of electrical engineering at King Abdullah University of Science and Technology, said.