According to an article in the New York Times, drinking beer is correlated negatively with scientific productivity. What is surprising is that even small quantities of beer are correlated with decreases in productivity.

But correlation is not causality. They have not shown that drinking beer makes you less productive. They have shown that people drinking beer are less productive. (It is not the same!)

Source: Scott Flinn.

On my blog, the best content is in the comments. Sébastien reminded me of this fact today by offering a link to an article in Physics Today by Lee Smolin. The gist of the paper is that scientists feel a lot of pressure to follow the lead of powerful senior scientists. It is much easier to be productive when you follow established techniques. Any prospective Einstein is crushed by the system.

There are many reasons for rejecting a paper. The authors might have failed to communicate their results efficiently. There may be a flaw in the science. Or the authors might have cheated. These flaws come from oversights, incompetence, and lack of ethics. But most importantly, they may all be motivated by the greed to publish more and faster.

Today is a bad day. I reviewed or rereviewed 5 papers from 3 different sources today. The best of these papers is a case of self-plagiarism. Three of the five papers were written by inexperienced students, or they appear to be, with minimal or no supervision from a senior researcher. The last one might have made a good blog post.

I believe that we are due for a revolution in science. We need to definitively stop counting the number of papers people produce. This game has run its course. If I interpret what I read correctly, it has become quite a bit counterproductive.

I propose that people list their two biggest accomplishments. It could be an experiment or a theorem they proved. To improve your case, you need to outdo one of your two biggest accomplishments to date. It does not matter if you publish 50 papers a year: you only improve your status if you outdo yourself in a big way.

Students can get started quickly. Senior researchers will have a harder time making progress. I submit to you that industry already works this way. Senior engineers are only as reputed as their two biggest most-difficult projects. It does not matter if you completed 120 small projects.

C.V.s would now fit in a single page. Tell us where you work, where you got your Ph.D. and what the two big things you did are. That is it.

Michael Nielsen posted a link to a paper in Science stating that the lone scientist is outgunned by teams and collaboration. Keith Sawyer supports this claim and gives more details.

In Computer Science, there are so few single-author papers — how can you do a sane analysis? I propose the following comparison.

Single-author papers are riskier, but can be more original:

• You can work on weird or highly risky projects: you only have to convince yourself.
• You can fool yourself and waste much time.
• You have to wait for formal peer review before you get feedback. Fast feedback tends to help you improve faster.
• You have to increase your breadth. You may need to run your own experiments as well derive new theories.
• Except for a few reviewers, you may end up being the sole human being to know about your contribution.

Multi-author papers are safer, but can get bureaucratic:

• You have to choose safe topics if you want others to be interested.
• If the idea is bad and you are not too intimidating, your partners let you know early on.
• You can learn fast from the expertise of your co-workers.
• You often do not fully appreciate and understand the work delegated to others. You tend to focus on what you know best and become more specialized.
• Your partners are likely to cite your joint work and tell others about it.

Clearly, if your goal is to build up your career, I would say that working with others is a good idea. However, producing a paper on your own shows that you do not need to rely on others. It shows your independence. It shows your breadth.

I have always had solo projects. They tend to be more painful, but somehow, I always have good reasons to push them forward. Mostly, it allows me to produce work without compromising on my (crazy) ideas. I can take risks without feeling guilty. Sometimes I get lucky. Sometimes I waste months in vain.

My theory is that solo authors take more risks and more often end up with poor, unciteable work especially if they lack maturity and skills. However, I predict that if you focus on senior researchers, you will find that solo papers are as good as collaborative work. The only exception to this rule would be senior researchers who enter a new field.

Long ago, search engines stopped using the metadata available in the header of HTML pages, because people would lie or enter misleading data by mistake. Many web sites still provide Dublin Core metadata as part of their HTML, but this data is known to be misleading, incomplete and wrong. There is no evidence that metadata can enhance search. Period.

Nevertheless, Yahoo ! announced that it is going to enhance its search results with RDF metadata. They give linkedin as an example: apparently, linkedin pages are filled with metadata waiting to be exploited. Using this metadata is great idea because linkedin can be trusted. Some other things would make sense, like GeoRSS. It would be great to know where some pages say they live.

Extracting metadata from one trusted web site is one thing. Exploiting the metadata out there is another.

A few things should be pointed out:

• As far as I can tell, Yahoo! is not talking about using metadata to improve its result sets in general. It would fail. They merely want to better describe the links found and maybe provide specialized services. If I were them, I would go around and entice various important web sites (amazon to begin with!) to provide more trusted metadata. They probably have been doing just that.
• Beside some specific instances, I do not see how it will make their search engine better than Google. No matter what, the vast majority of web sites will contain no metadata, or wrong metadata.
• There is no talk of non-trivial inference engines. Yahoo! still won’t be able to tell you whether G. W. Bush is a drunk or not.
• Graduate students worldwide, stay calm. I could not find one occurrence of the word ontology in Yahoo!’s post. They are talking about RDF, not OWL. So you can stop describing the whole world in a RDF graph.

Parand obviously meant to get me to react. He had one eye-opening experience while chatting with a famous professor. He reaches the conclusion that heavy-league schools are necessary to get this kind of learning experience. Mostly, his conclusion seems to depend on two facts.

• Physical presence matters and it cannot be replaced by electronic means. If this were false, you could just go on the Web and interact with super-smart people from all over the world without having to pay Harvard-level tuitions. Well, I do claim it is largely false. The main differences between physical presence and e-Learning are bandwidth and latency, but these are improving rapidly. By the next decade, you will be able to chat with dozens of people on large high-resolution screens with high quality sound. You will have convenient and highly interactive tools. Physical presence will grow irrelevant unless you need to touch the skin and smell your professors.
• Professors at non-heavy-league schools do not know as much. Paul Graham said it best: “you have to go pretty far down the list of colleges before you stop finding smart professors in the math department.” The smart local professor may not be as connected and he may not benefit from as much marketing, but he can dazzle most students.

More generally, if you want to know how to get really smart, go watch what really smart people do. How does the famous professor learn? Does he spend days in lecture halls listening to some colleague? Nah! I bet you will find him interacting with some of the smartest people in the world every day, and spending a lot of time working in his office, crouched over his desk. My point is that you do not get smart by sitting in lecture halls. You get smarter by working at it. Smartness is not contagious, at least not by physical contact.

Of course, perception can play tricks on you. If you know that the person in front of you is famous, you may feel like you are learning much more just by being in his presence. People may perceive you to be smarter because you studied with such a famous person. By knowing more famous people, you may also have more self-confidence. So there are economically sensible reasons to attend heavy-league schools, but they mostly have to do with branding.

Disclaimer: I am a graduate of what I consider a Canadian heavy-league school. It did help me get my first professorship.

I routinely get emails from unknown graduate students who ask me to help them. Most of these emails are interesting. Unfortunately, some are unacceptably rude.

What is ok:

• Can I get an electronic copy of your paper?
• Do you have the source code or the data for this paper?
• This new paper claims to do better than your algorithm, what do you think about their work?
• In Algorithm 1, isn’t there a missing loop at the end?

What is not ok:

• This friend of yours has written a paper, can I get a copy?
• Can I get an implementation of this standard algorithm?
• I don’t understand this standard algorithm, can you explain it to me?
• I need to adapt your algorithm to my own problem, can you do it?
• It is urgent, I need you to… (Hint: if it is urgent, don’t email a stranger about it.)

What you must understand is that if a researcher fails to answer your query, he is not being rude. An email to a researcher is a bottle in the ocean, you may get an answer, you may not.

One key quality that all competent researchers share is that they are able to deal efficiently with distractions. They may choose not to return an email or a phone call, or not to wash, to get their work done. Researchers are very good at fooling themselves into thinking that their current research is the most important thing they must do.

Oh! And whoever asked me for an implementation of Godin’s lattice-construction algorithm, and got a polite answer saying I had none to offer, you were out-of-line in telling me I was rude. You did get a polite answer. I will not stop my work to implement a standard algorithm for your own selfish needs.