Earlier this year, we predicted that the number of AI related papers on arXiv would continue exponential growth to the impressive level of 47000 papers for this year.
Now, with the data for January through April complete, we can confirm that the AI field is on track to reach that number, as measured by the output of research papers on arXiv, despite lockdowns, economic slowdowns, and all crisis related challenges.
With almost 12500 unique new AI-related papers in 2020 on arXiv*, we are seeing an average increase in the number of papers of 41% over the same months of 2019. The graph below shows the numbers by month for 2020 and the two previous years.
With April’s numbers up ‘only’ 29% over last year, we still have more papers than in any month of last year. Could it still be a signal of a small slow down? Let’s wait and see how the normally highly productive months of May and June will look.
This phenomenal growth does make it ever harder to keep up with the AI literature. In their keynote Q&A at ICLR 2020, Joshua Bengio and Yann LeCun actually admitted they gave up on being up to date: “There is just too much!” An opportunity for AI based search, recommendation and knowledge navigation engines for the field of AI? At Zeta Alpha, we’re working on it!
*) We counted Machine Learning, Artificial Intelligence, Computer Vision and Pattern Recognition, Information Retrieval, Neural and Evolutionary Computing, and Computation and Language categories from arXiv. Keep in mind that when submitting a paper to arXiv the author can select multiple domains, here we counted each paper only once.