There as been discussion in mid-2018 as to whether AI research may be entering a new "winter," as argued in this article. We shall define an index as follows:

• +1 if on Dec. 1, 2021 5-year Google trends search on "AI winter" shows the May 27-June 2 2018 peak (which is 100 at launch) to be at < 25, indicating that a value four times as large has occurred between those two dates; -1 otherwise.

• +1 if two or more keynote talks at NIPS or IJCAI contain the phrase "AI winter" in any of those two meetings taking place in 2020 and 2021. -1 otherwise.

• +1 if the fraction of listed YC funded startups that mention "AI" or "machine learning" in their description is lower in the first half of 2021 than in the first half of 2018. -1 otherwise.

• +1 if search of the CS section of the arXiv for abstracts containing "artificial intelligence" OR "machine learning" for jan-may 2021 returns less than 3000 items, about 50% more than the corresponding search for 2018. -1 otherwise. (For reference, 2018 returns 1922; 2017 returns 756; 2016 returns 420.)

• +1 if the ratio of results returns in the above archive search limited to "machine learning" divided by those limited to "artificial intelligence" is higher in 2021 than the 2018 value of 1752/233 = 7.52; -1 otherwise.

Any of the above can resolve ambiguously if there is not a good consensus resolution based on reliable data. If the total number of items that resolve unambiguously is N, the index is given by the total score divided by N.

What will be the late-2021 value of the AI winter index?