Isn't it plausible that DeepZenGo simply isn't at the level of play of AlphaGo? Facebook's DarkForest is also a competent AI, but even it does not compete yet.
It's hard to compare, as even Deep Zen runs on much more...modest...hardware than the match version of AlphaGo did. It's conceivable that AlphaGo would get clobbered on Zen's hardware, or that Zen would clobber AlphaGo if that team had Google-eseque resources available. But there's no reason to believe we'd ever see such a fair comparison. It makes no sense for either team!
Zen is commercial software sold to people with a normal PC, so obviously reaching AlphaGo performance on AlphaGo hardware does their customers no good either.
It's for sure stronger than Darkforest. Some free PC programs like Leela have reached Darkforest levels on normal PC hardware (Darkforest used 44 NVIDIA Tesla cards, Leela uses a single normal "gaming" GPU), and Zen is several stones stronger than Leela.
A rather similar hardware with a recent version of the same program plays on the KGS server. Seems to be about what would be 10d and is among the top players on the server (was 1., now seems to be 3.).
You can see the graph of its playstrength; its quite impressive to see the gains Zen got from implementing ValueNet (and previously with PolicyNet) really.
CPU: 2 x Intel Xeon E5-2699v4 (44 cores/2.2 GHz)
GPU: 4 x nVidia Titan X (Pascal )
RAM: 128GB
So rather similar.
~10d is way way lower than what AlphaGo had to be to play on par with one of the top players of Go. Pro ranks start around or just above the KGS 9d (which is around what a 7d rank would be for regular amateurs in most world amateur organisations). A rough approximation is that you can take 3-4 pro ranks for each dan rank/stone of handicap (well, to the extent one can even approximate strength by the rank, as a rank once won is kept for life, so as you age and young pro players pass you by, your rating is getting inflated). So apriori you could expect Zen to play around middling pro ranks, and maybe around the strength of the v13 of AlphaGo, the one that beat Fan Hui, and described in the Nature paper. And that's about what his play with Chikun indicates as well.
Cho Chikun's rating is 3239, while Lee Sedol's is 3508. so around 300 elo difference, which seems a conservative estimate of the difference scale (and the results of the two matches do indicate, if weakly, that the actual difference is even larger). A 230 point gap corresponds to a 79% probability of winning.
For comparison, distributed AlphaGo v13, the one that played vs Fan Hui is 250 elo points stronger than a single machine version of the same version of AlphaGo.
Also if that would carry any weight with you, you could just take it on the say-so of Myungwan Kim 9p, who was doing commentary on both DeepZen and AlphaGo matches. Really the difference in strength seems rather obvious to such a strong go player, as you can see in the way he criticized some DeepZen's moves, and how impressed he was with so many of AlphaGo's.
My own personal opinion is embodied in the phrase, "You don't stop playing games because you get old, you get old because you stop playing games."
I'm fairly close to retirement, and have friends who are getting there (or retired already). I really enjoy programming, and intend to continue indefinitely; most of my friends tolerate it as a job, and look forward to not doing it when they retire.
So it's the old "follow your passion" thing, again, I guess... I have that curiosity about such things, they don't.
