https://doi.org/10.31449/inf.v42i3.2465 Informatica 42 (2018) 285–289 285 
IJCAI 2018 – Chinese Dominance Established 
 
Matjaž Gams 
Jožef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia 
E-mail: matjaz.gams@ijs.si 
 
Editorial 
 
In July 2018 in Stockholm, ICML, AAMAS, 
ICCBR and SoCS joined with IJCAI and ECAI to 
establish the first major worldwide AI event. This 
paper is about the resulting IJCAI-ECCAI event [1]. 
The 27th International Joint Conference on 
Artificial Intelligence and the 23rd European 
Conference on Artificial Intelligence merged with 
the other events to form a single conference. 
Around 7000 participants divided their time 
between these conferences over 14 days as one fee 
covered the entrance to all the events. As a 
consequence, several researchers attended several 
conferences, which in itself was a major 
achievement. Namely, the conferences and even the 
individual sections of the conferences are becoming 
so specialized that AI researchers are becoming 
oblivious to the achievements being made in a 
related area, leading to specialization and small 
incremental improvements, thereby deterring major 
innovations. Fortunately, in 2018 there was a 
serious attempt to reintegrate the field. 
For the organizers, the super-event joint 
conferences represented a huge effort, but 
everything ran smoothly – albeit with a couple of 
small exceptions, as usual. One of them was the 
initial robot dance, where a Nao robot performed a 
predefined sequence of moves, which the human 
dancer enriched with dynamics and scope. The 
glitch was a loss of sound during the event 
(deliberate or by accident?). Added to this, the lack 
of any AI in the performance was a huge issue for 
many of the participants; in particular, the absence 
of true AI, one of the central themes of the 
conference. However, the artistic impression was 
there. Perhaps not surprisingly, as the small Nao 
robot was clearly physically and dynamically very 
much inferior to the flexible human dancer, a kind 
of reverse of David and Goliath seemed to be taking 
place. Also, the big 1000+ lecture rooms were 
organized in such a way that at no time was 
everybody sitting down, instead there seemed to be 
5-10 people in motion at any moment. That aside, 
Stockholm is a traditional, open, metropolitan city 
that has hosted conferences for up to several tens of 
thousands of participants before, and the AI 
organizers have extensive experience as well; so by 
any measure the event must be considered as an 
organizational success. 
The IJCAI-ECCAI joint event involved a record 
3470 submitted papers: 37 % more than in 2017, 
while the 2017 event was 11% up on the previous 
year, confirming the steady growth from 2007 on. 
AI continues to progress as a scientific field and as 
an area of human interest. 
The first major technical impression in 2018 
was that Chinese dominance has finally been 
established. Eclatantly! In 2017, 37% of the papers 
were Chinese, while a year later this figure was 
46%. Only 9% increase, one might say, but the 
2017 conference was in Australia, with strong 
Chinese ties, while Stockholm is in Europe, and it 
was a joint European and international IJCAI 
conference, meaning around half of the event was 
basically a European conference. Despite that, 
European and American papers constituted around 
20% each, while several authors, in particular from 
the USA, were also Chinese. Astonishment and 
admiration are the right words to describe this 
Chinese success. 
The more detailed numbers are as follows: from 
the 710 accepted papers (21% acceptance rate), 325 
came from China, 129 from EU (UK 37, France 22, 
Italy 18, Germany 15, Austria 12), USA 122, 
Singapore 26, Australia 23, Japan 17, Israel 13, etc. 
When asked if it is reasonable to limit non-
European papers at least for the ECCAI conference, 
say to 50%, several of the researchers expressed 
concern that it would mean that several of the best 
non-European papers would then end up being 
presented at other conferences. Several of the 
Chinese papers were indeed of high quality, 
demonstrating Chinese innovation, good education 
and the major support for AI in China. There were 
some concerns that the Chinese papers often follow 
a pattern with a specific idea, lots of complicated 
mathematics and an unverified empirical 
comparison. But that is true for many other papers 
as well. It should be noted, in addition, that due to 
several national European research policies, it is 
often nearly worthless for domestic evaluations to 
publish a paper at IJCAI or ECAI, since all that 
counts for these researchers are journal 
publications. The absence of more high-quality 

286 Informatica 42 (2018) 285–289 M. Gams 
 
European papers might therefore be partially 
attributed to the strange European scientific 
policies. Some of these issues were discussed at the 
panels, as presented in Figures 1 and 2.  
 
Figure 1: At a panel about European AI, the 
importance of the field for European progress was 
clearly established. 
In some presentation, e.g., the one shown in 
Figure 1, the European position and the self-
evaluation were not exactly in accordance with the 
percentage of conference papers. Some other 
positions even sounded a bit like a clip from a 
galaxy far, far away. But in reality, the panels were 
of high quality and several essential issues and 
initiatives were raised. Several panelists mentioned 
that there is no AI coordination in Europe, even 
though the EU is still the no. 1 world economy. In 
terms of AI funding, the USA prevails over China 
2:1, and China prevails over the EU, again by 2:1. 
Such estimates might be misleading since the 
nominal comparison took place – instead, real 
economy (how many kilograms of sugar or of steel) 
already puts China above USA in terms of scientific 
funding. 
There are two important differences between the 
USA and the EU: the USA executes bold 
international policies, whereas the EU finds its soft 
approach is sometimes hurting its economy and 
society. The EU used to be no.1 in computer 
science; now it is no. 3. Lots of this falling behind 
was not necessary at all; instead there are subjective 
leadership reasons for the decline, e.g., the EU 
patent system is enormously complicated and 
bureaucratic compared to the American one. 
Another problem: the UK has the best European AI 
based on many criteria, and so Brexit will make this 
situation worse for the EU. Whereas top EU 
projects like H2020 represent world-class research, 
and the EU is still leading in many areas of business 
and science, the strong scientific funding for key 
areas as well as policies to support them are lacking. 
 
Figure 2: EU strategy involves three elements: 
science/technology, socio-economic changes and 
the social framework. 
While the EU is as concerned with legal issues 
as it is with the research, China has significantly 
improved its AI efforts through governmental and 
private funding, and there is no major rift in the 
government. The democracies of several European 
countries and the USA are torn apart because of 
ideological and political antagonisms, instead of 
focusing on technical progress. For Chinese 
researchers, the road to success and obtaining a 
good position at home is to publish at major AI 
conferences, in major journals and join established 
researcher teams in the USA or Europe. For 
Europeans, it is possible to follow the Chinese path, 
but no European country offers a several-times-
higher salary for researchers returning home, like 
China does. While the presidents of superpowers 
from the USA to Russia declare the tremendous 
importance of AI in relation to world dominance, 
the percentage of papers best demonstrates who 
supports the field the most. This is not to say that all 
major countries are not increasing their AI funds 
significantly. For example, the EU has presented its 
plans at IJCAI (Figure 3): first, a 70% increase, 
followed by a 100% and then another 100%. The 
US Department of Defense (DoD) established the 
Joint AI Center (JAIC). It will host the DoD’s 600 
AI projects with an estimated $1.7 billion over 6 
years. As predicted, AI will likely change the nature 
of welfare, along with several other fields. 
However, without sufficient AI research, nobody 
can expect to maintain its leading position in the 
world. 

IJCAI 2018 – Chinese Dominance Established Informatica 42 (2018) 285–289 287 
 
 
Figure 3: The EU will significantly increase AI 
funding. Finally. Will national governments follow? 
A closer look at the EU plan reveals that there 
are several new ideas, as presented in Figure 4. 
Among others, the EU will fund the open AI 
platform, which is at least partially influenced by 
Elon Musk’s, which by the way won the first 5 vs 5 
Dota2 game with expert players (some small 
additional limitations). The EU plan was probably 
the major AI strategy presented at the conference. 
While China does it on its own and the USA 
allocates most funds to military applications, the EU 
is focused on a public, general, AI-boosting plan to 
benefit everybody. That is for sure great news, not 
only for AI in Europe, but for humanity as a whole.  
 
Figure 4: The EU strategy introduces several 
integrating EU components, including the AI 
toolbox and the Network of Digital Innovation 
Hubs. Unfortunately, many of the most advanced 
AI hubs are in the UK. 
Several new mechanisms like CLAIRE are 
already active (https://claire-ai.org/): “an initiative by 
the European AI community that seeks to 
strengthen European excellence in AI research and 
innovation.” 
“If Europe were to fall behind in AI technology, 
we would be likely to face challenging economic 
consequences, an academic brain drain, reduced 
transparency, and increasing dependency on foreign 
technologies, products and values. The CLAIRE 
initiative presents a proposal to avoid that.” 
“The CLAIRE initiative aims to establish a pan-
European network of Centres of Excellence in AI, 
strategically located throughout Europe, and a new, 
central facility with a state-of-the-art, “Google-
scale”, CERN-like infrastructure – the CLAIRE 
Hub – that will promote new and existing talent and 
provide a focal point for the exchange and 
interaction of researchers at all stages of their 
careers, across all areas of AI. The CLAIRE Hub 
will not be an elitist AI institute with a permanent 
scientific staff, but an environment where Europe’s 
brightest minds in AI meet and work for limited 
periods of time. This will increase the flow of 
knowledge among European researchers and back 
to their home institutions.” 
Maybe, we should also remember the times 
when science was not a business, when we 
researched not for the purpose of cash, but for 
reasons of fundamental curiosity, a desire to 
improve our knowledge. Some spirit of that kind is 
still observable at the conferences and was also 
demonstrated, for example, by the computer chess 
tournament. During the breaks many participants 
occasionally stopped by and observed the most 
interesting matches. Komodo won the World 
Computer Chess Championship 2018 after a play-
off with GridGinkgo. In third place was Jonny, due 
to a win over Leela Chess Zero. The latter was 
observed with much interest, due to having 
implemented AlphaZero for the PC. It was not a 
match for the best programs, instead it played out 
 
Figure 5: The EU expects that AI progress 
will bring several benefits, from the economic 
impacts to solutions to the societal challenges. 
There are several areas that will see major 
advances in the near future, such as 
healthcare. 
 

288 Informatica 42 (2018) 285–289 M. Gams 
 
very differently – intuitively, lucidly and error-
prone. Obviously, it lacked the power of the Google 
computers to validate its fancy ideas, often in the 
form of sacrifices. Figure 6 shows the Komodo 
team, who received the Shannon Trophy (and 
replica) from the chairman of the ICGA David 
Levy. 
 
Figure 6: Komodo was again the computer chess 
champion on PCs. Leela Chess Zero, a PC version 
of the Alpha Zero, played lucidly, but had no 
chance against the hard logic of Komodo. 
In Stockholm 2018 the social meetings of 
societies were boosting the exchange of information 
and cooperation, be it inside the EU or international 
societies. For example, the EurAI meeting (Figure 
7), IJCAI AI societies meeting, IFIP meetings, etc. 
The IJCAI report should first of all be about 
scientific achievements. In 2018 there was distinct, 
albeit rather expected, progress. Indeed, there were 
plenty of reasonably novel improvements, and 
indeed the major theme was a challenging one: How 
to grow a mind, a true AI - but that was it. Quite 
enough for many, but a bit too classical for others. 
Furthermore, the AI influence on our everyday life 
has already achieved much greater impact than 
generally anticipated in the public opinion: every 
day AI makes around 100 trillion decisions, 
meaning it is thoroughly embedded into our society. 
Coupled with other ICT achievements, human 
society long ago developed into an information 
society – an integration of humans and ICT 
systems, and an integration of human society and 
technology. This is one of the reasons why nobody 
understands what is actually going on – social 
scientists understand society, while engineers and 
technical scientists understand technological 
systems, and finally nobody understand the two 
embedded and integrated into one unity – kind of 
Borg stuff, just that the unifying essence is the web 
and ICT and AI services. Another analogy is related 
to computer chess – when humans play based on 
their own brains, the inferiority and inability to 
understand complex relations are evident. Only 
coupled with powerful computers and advanced AI 
programs can we hope to decipher the societal 
changes and trends, and propose good solutions. 
With regards to the novel applications, Tambe’s 
group stood out from many – their security AI 
designing daily schedules for airports, harbors and 
other relevant facilities is employed at several 
locations worldwide. It has even been given to 60 
wildlife parks to cope better with poachers. That is 
one of the successful applications, accompanied by 
a huge mass of new research systems, e.g., a novel 
HW and SW embedded system connected to the 
patient’s spine that enables a paralyzed patient to 
stand up. 
New classes of applications are emerging, e.g., 
in visual tasks. DNNs can transform a human face 
into another, even create a new face never seen 
before. An animal, say a horse, can be camouflaged 
into zebra stripes and it can move freely around in a 
simulated video. Systems speak perfectly and listen 
better than humans; they can sharpen a picture or 
translate from voice online. Google search is using 
DNNs to capture the best answer to a question.  
On the other hand, there are seemingly bizarre 
simple problems that researchers have a hard time 
dealing with. While it is generally accepted that 
DNNs outperform humans in visual classification 
tasks, it is still a big problem to transfer one ML 
system based on examples from a specific hospital 
and specific scanning devices to another. The 
technical differences are small, causing human 
experts no problem, but for the DNNs these small 
details significantly impair the quality. Until 
recently, that is. At the conference, several solutions 
related to transfer learning, general AI and also real 
AI were presented and discussed. Why should AI 
systems not learn like children, gathering 
knowledge and learning from there on with a small 
number of examples, even a single one? 

IJCAI 2018 – Chinese Dominance Established Informatica 42 (2018) 285–289 289 
 
There is shallow, i.e., current AI, deep AI (also 
claimed as shallow AI), real AI, and fake AI. The 
last one refers to chatbots, i.e., virtual assistants, 
where human operators often jump in 
communications and leave users under the 
impression that it is AI on its own. The real AI was 
one of the major themes of the conference, which is 
quite a big difference from the previous 
conferences, where the primary goal was to 
complete the tasks better than expert humans, be it 
chess or detecting malignant tissue patterns. Now 
the task is different – perform at the level of 
children aged a couple of years. While supervised 
learning clearly achieves top performance, 
compared to humans it needs far too many 
examples, which is not acceptable, at least for the 
slow humans. Similarly, reinforcement learning 
needs way too many trials. Furthermore, machines 
do not have common sense compared even to young 
children. 
In terms of the ban on autonomous weapons, 
more and more societies and countries are joining 
the ban. EurAI, as the union of all AI European 
societies and the second largest AI society in the 
world, also supports the ban. 
 
Figure 7: More and more societies and countries are 
joining the ban on autonomous weapons. EurAI, as 
the union of all AI European societies, also joined 
the efforts. 
In 2018, the debate on banning autonomous 
weapons was held in the UN and in the European 
Parliament:  
https://www.stopkillerrobots.org/2018/07/parliamen
ts-2/. The list of institutions supporting the ban is 
here: https://www.stopkillerrobots.org/coalition/. 
 
Figure 8: Tegmark’s view of the AI field. 
Another interesting approach is to generate 
general AI, as was the case in the 2018 IJCAI 
conference. Currently, the majority opinion among 
AI researchers is that general AI is possibly 10 
years away. It is probably not that the AI 
community lacks computer power or finances; it is 
the novel ideas that we are striving for. There is also 
a reasonable consensus that AI can, could, should 
and will help humans solve major human societal 
problems. Scientists should avoid the politics, 
especially the discrepancies between different 
ideological or political tracks, and defer from 
attacking colleagues along these lines. Science 
should be kept as separate as possible from politics 
and ideology. With these words from Tegmark 
(Figure 8) we look into a bright EU AI, AI and 
human future. 
References  
[1] IJCAI 2018 conference (https://www.ijcai-18.org/). 
  
 
Figure 7: Children learn in a very different way to 
AI. Why not copy them? 
 

290 Informatica 42 (2018) 285–289 M. Gams