Statistical AI systems have pretty decisively won out over rule-based systems.

The claim is hedged ('pretty decisively'). It says that by July 2017, statistical or machine-learning approaches had clearly overtaken rule-based or symbolic approaches in AI. TRUE: as of mid-2017, state-of-the-art results and deployed systems in major AI areas (vision, speech, machine translation, game playing) were dominated by statistical or neural methods, and the expert consensus recognised this. MOSTLY TRUE: dominance was clear in most major areas, with some notable domains (for example certain planning, verification or expert-system applications) still relying on rule-based methods. MIXED: statistical methods led in some major areas, while rule-based methods remained competitive or dominant in others of comparable importance. MOSTLY FALSE: statistical methods were rising but not yet dominant in most areas. FALSE: rule-based approaches still dominated in 2017.

Set before the research began.

1. E1 The JHU Machine Translation Systems for WMT 2017 Johns Hopkins University (Ding, Khayrallah, Koehn, Post, Kumar, Duh); Proceedings of the Conference on Machine Translation (WMT), Volume 2 · September 2017

   “Ourphrase-basedsystemswereclearlyoutperformedbyNMTsystemsforalllanguagepairs,byadifferenceof3.2to8.3BLEUpoints.”

2. E2 The University of Edinburgh's Neural MT Systems for WMT17 arXiv (ar5iv rendering), University of Edinburgh authors

   “Among the constrained submissions to the news task, our submissions are ranked tied first for 11 out of the 12 translation directions in which we participated.”

3. E3 Learning Transferable Architectures for Scalable Image Recognition arXiv (Zoph, Vasudevan, Shlens, Le; Google Brain) · July 2017

   “achieves stateof-the-art accuracy of 82.3% top-1 and 96.0% top-5 on ImageNet, which is 0.8%”

The evidence shows neural systems beating or leading phrase-based systems in two teams' 2017 machine translation results [E1][E2], plus one state-of-the-art image-recognition result [E3]. It says nothing about speech, game playing, areas that still used rule-based methods, or expert consensus, so it cannot show whether statistical methods dominated most major AI areas as the criteria require.