Intelligence and the AI

Here at the OEG, a research group where I currently do my research (and where we also eat jamón and cakes), every once in a while we tend to informally discuss topics or research papers that we find interesting. This time, inspired by this interview with Douglas Hofstadter and in the presence of coffee and chocolate, we talked about artificial intelligence (AI) and discussed a couple of interesting questions raised in the interview, such as what AI really is and is there actually any intelligence in it. 

Photo: Esther Lozano (@esloho)

Throughout my career I got to know various definitions of the AI. This doesn’t mean, however, that some definitions are right and some are wrong; rather, the context of one definition is different than of the other. Having said that, in order to approach any discussion about the nature and success of the AI, it is necessary to first be clear which of these definitions we are discussing.

One of the definitions of the AI is saying that the AI is about machines and computer programs that express intelligent behaviour and are able to solve problems that are typically solved only by humans. In this context, there are not so much interesting things to say; we all agreed that this definitions is the one that best suits to what AI really is today, and that in this sense we have gone through a lot of progress and achieved remarkable results in the past several decades. We have programs that can beat us in chess, we have programs that can translate text for us, we have robots that can go get us a sandwich, and we even have self-driving cars.

Much more interesting part of the discussion comes if we embrace the point of view in which every field of science has one big question that it is trying to answer. The big question in the AI is whether it is possible to create consciousness and intelligence outside of a living organism and, in this context, it is exactly what AI should try to achieve. The key difference with the previous approach is that here we are talking about not just behaving intelligently, but actually being intelligent. We are talking about a machine that thinks.

If we consider this second definition, the problem with the AI is that there is very little intelligence in it. The way that computer programs solve problems is far from the way people solve them. To take an example, the current AI approach for text translation works in a way that is far from how any person does it, and this is the key issue raised by Douglas Hofstadter in the interview mentioned at the beginning. On the other hand, we must admit that the goal of creating intelligence in a machine is extremely difficult and the science is not absolutely sure how the brain and intelligence work in the first place.

One of the interesting questions that arose in our discussion was whether or not we actually need to understand how the brain works, and whether or not we could replicate the brain inside the machine without such understanding. The replication of the behaviour of the human brain is exactly how Maria likes to define the AI and, according to Jorge, it might well be that there is no need to understand how the brain works but just to replicate its behaviour. A good analogy of this point of view is given by Stuart Russell and Peter Norvig in their book: “”The quest for artificial flight succeeded when the Wright brothers and others stopped imitating birds and started…learning about aerodynamics”. Airplanes don’t fly the wings, why should computers think?

No matter how much you like or not this second AI approach, the fact is that such AI is not just extremely difficult to achieve, but there is very little effort directed towards this achievement. The fundamental research in the AI lacks a larger community and dedication of more than a handful of scientists and research groups. If we want to see computer programs intelligent in a way that people are, able to think, learn, and apply the knowledge in new and complex situations, a lot more effort has to be dedicated in the fundamental AI research. And this is something that all of us in this discussion agreed on and would like to see it happening.

I strongly recommend you to read the interview that sparked our discussion on this topic. Also, the book by Douglas Hofstadter “Godel, Escher, Bach: An Eternal Golden Braid” is highly recommendable.

Many thanks to Maria, Esther, Jorge, Pablo, Alejandro, Freddy, Mariano and Jose for participating in this discussion.

The Nobel Prize in Physics and What Is So Wrong About It

The Royal Swedish Academy of Sciences announced today that the winners of the 2013 Nobel Prize in Physics are Francois Englert and Peter Higgs. The prize goes for “”the theoretical discovery of a mechanism that contributes to our understanding of the origin of mass of subatomic particles, and which recently was confirmed through the discovery of the predicted fundamental particle, by the ATLAS and CMS experiments at CERN’s Large Hadron Collider”.

Sean Carroll has a nice post on the Nobel Prize and I couldn’t agree more with him. One thing that is sure is that Englert and Higgs absolutely deserve the prize. However, as Sean argues, some facts about the Nobel Prize are really annoying.

First of all, the prize is limited to only 3 persons, which is not how science is done nowadays. Usually, there are many scientists that make contribution to a scientific discovery, both independently or in a collaboration. In particular case of the Higgs boson, seven physicists get the credit for its theoretical foundation.

Second, the limit for the number of prize winners and the fact that it can’t be awarded to organizations is especially bad for experimental scientists because experimental science usually involves even more people or whole institutions. The experimental confirmation of the Higgs boson is equally important to science as it is its theoretical foundation, and it involved nothing less than the most complex machine humans ever built – the LHC, in which thousands of scientists and engineers from all over the world collaborated on.

The Nobel Prize is a great way to honor and recognize when good science is done. But nevertheless, we should not focus on the prize itself. Richard Feynman gave a really nice speech when he received the Nobel Prize in 1965, and you can see what he thought of it in a video below. The real thing in any scientific discovery should be the “pleasure of finding the thing out“, and we should definitely “endeavor to honor what was actually accomplished, not just who gets the gold medals“.

Rubbish Science, Journal Citation Report, and Impact Factors

A former professor and a friend of mine, Dragan Djuric, recently published a paper in a journal called Metalurgia International (Romania)*. Nothing would be so special about it without this two important things: first, the paper is scientifically worth nothing as it is just a bunch of rubbish; and second, the mentioned journal is** on Thompson Reuters JCR list! Sounds almost impossible, especially after you take a look at the paper.

Just the title of the publication – “Evaluation of Transformative Hermeneutic Heuristics for Processing Random Data” makes you think about what the heck is going on. If you decide to give it a shot, you will see an endless rubbish tightened into a number of jokes. To give you an idea, the list of references contains (among others) a 2012 paper by Laplace published in the “Proceedings of the Joint Somalia Conference on Potential Theory and Pragmatic Practice“; or one published in the “Journal of Modern Illogical Studies“. Going further, one of the conclusions says that “TV exposure with high wife education and low spouse education, influences a positive opinion on the EU“. Or how about this paragraph:

As we will soon see, the goals of this section are manifold. Our evaluation could represent a valuable research contribution in and of itself.

The first experimental results came from 2500 trial runs, and were not reproducible. The next batch of results come from only 50 trial runs, and were not reproducible. Continuing with this rationale, the many discontinuities in the graphs point to improved precision introduced with our decision tree algorithms. Such a hypothesis at first glance seems unexpected but fell in line with our expectations. As hypothesized, the final run was sufficiently consistent, which shows the useful convergence of our heuristics.

Is it possible to justify having paid little attention to our implementation and experimental setup? Yes, but only in theory. Our evaluation strives to make these points clear.

An important question to ask is how did such a journal end up in the JCR list in the first place. Apparently, the only thing Dragan had to do to publish this is to pay a publishing fee. Sadly, it was that easy and the paper got published without any corrections.

Impact factor for "Metalurgia International" journal in the Web of Knowledge

Impact factor for “Metalurgia International” journal in the Web of Knowledge

But the biggest problem perhaps is not just that (at least) one JCR journal actually publishes papers without reviewing them. The biggest problem is that there is already a number of professors and faculty personnel in Serbia (and it might as well be the case that similar examples exist in other countries too) that got their positions and national project funds by gaining points publishing in MI or similar journals. Someone even bothered to list most of the Serbian authors in MI, as apparently there have been a large number of them so far. And those authors were even shameless to start a petition (asking the government to protect them as “the most capable and productive part of Serbian scientific community”!), which came right after the Serbian scientific library (Kobson) announced that it rejected a high number of requests for deleting certain papers from their database. Sounds like a joke, and I wish it really was one.

If you are curious you can take a look at the paper here (and pay attention to Dragan’s fabulous false mustaches and Boris’s wig). There is also a petition to the government to (among other things) reevaluate the contribution of people who published in MI and in similar journals, and take necessary steps to set things right (I signed that one).

Apparently, the situation with publishing low quality research in “serious” journals is more serious as this is definitely not the only case (see for example here and here). Of course, the question of how to assess scientific work remains as there are already critics on journal impact factor and calls for abandoning it. It remains to be seen how the scientific community will react on this issue and whether we will see some serious changes anytime soon.

In the end, if you had fun with the paper, you can check out how Hitler reacted on the news.

* Since the news about this paper appeared, the Web page of the journal is down (but you can check the Wayback Machine).

** It is reported that MI was recently removed from the JCR list.

Semantic Web and Me

As this is my first post on this blog, I think it would be nice to say how and why I got involved into the Semantic Web.

Back in late Spring of 2008 when I was about to finish the third year of bachelor studies, I was invited to join the GOOD OLD AI Research Network, a network of really nice people passionate about Artificial Intelligence and Software Engineering. The idea was very interesting and challenging to me and I was already collaborating with the people from the network as an undergraduate teaching assistant, so I was very happy to join the team.

Shortly after, Jelena Jovanovic and Milan Stankovic introduced me to the Semantic Web, I started to learn about Linked Data, ontologies, and lots of other new things, and I did like it a lot. During the summer, together with a few other colleagues I was working on a couple of Semantic Web related projects – Smiley Ontology and Online Presence Ontology. It was the time when I had a lot of discussions with Jelena and especially Milan, who was very kind to let me bother him almost daily over the chat about a bunch of questions and doubts that I had (the other person I was bothering a lot was my mother because back then we had only dial-up connection at home and she was deprived of using the phone, which she also had to use for business communication). But it was really a nice time for me, I learned a lot and it sure was fun.

Later on, I continued to collaborate with Jelena and Milan, and Jelena supervised my thesis which was related to the online presence project. In September 2010 I moved to Madrid for Master studies and joined the Ontology Engineering Group, where I am now at the second year of PhD.

Shortly, this is how it all started. In one of the following posts I will describe the work that I did for my Master thesis, and the work that I am doing now in the OEG.