Monday, 6 August 2007

The Physics Mafia

In 2002, the South African government re-calibrated funding per student in universities. Instead of all science students attracting the same government subsidy, science students were divided into 3 categories: subjects that were expensive to teach (Physics, Chemistry, life sciences), subjects with less intensive lab requirements (Mathematics) and subjects that are really cheap to teach (Computer Science).

Computer Science extra cheap to teach – cheaper than pure maths? Did you hear that right?

A moment's reflection will show that this is idiotic. Computer science at its most theoretical is pure maths. At its most applied, it is a kind of engineering (which is in the higher category, along with mathematical sciences – not the highest category, with physics etc.).

Despite numerous protests, this idiocy has persisted.

What's going on: is the SA government run by idiots?

It seems not (entirely): this information comes from the universities which are run by idiots; the government has only been stupid in believing them.

What these rankings represent is not what these subjects should cost to teach, but what they have historically cost.

Physics has been expensive for some time because it attracts low student numbers, has a relatively high fraction of senior academics (hence with more expensive salaries) and most physics departments have a generous complement of lab staff. Computer Science, on the other hand, is a relatively new discipline, without many senior academics and without lab staff.

The lack of lab staff is the most serious differentiator, and feeds back to the other major difference in numbers of senior academics. For a typical Physics first year class, the lecturer just shows up for lectures. Lab staff take care of pracs, and tutors take care of tutorials. Since the fundamentals are largely settled (most Physics developed in the last 200 years is considered advanced, and is not seen in big classes), the same lab exercises can be recycled every year, and lab equipment can last decades. Lab staff do not need specialist skills in demand in industry, and can be trained up off the street. This means that running large classes (e.g., Physics for engineers) is not particularly onerous. Up the scale, Physics is not particularly popular as a major, so level 2 and above classes are small, and lecturing small classes of keen students is a whole lot less effort than handling big classes of students of wide variations in ability and motivation. And of course running labs is a whole lot easier even in this scenario with trained lab staff. The result? Physics academics have a fair amount of free time to write papers and build their CVs.

In Computer Science, the situation is reversed. The subject (despite the odd downturn in demand) has been more popular than Physics for decades, resulting in much larger classes at level 2 and above. On the other hand, most Computer Science departments in South Africa have no lab staff. Their technical staff capacity is usually sufficient to maintain the servers, networks and lab computers, but not to set up and manage exercises. This puts a major additional burden on academics – at all levels from introductory to advanced classes. A moment's reflection would reveal that failing to employ support staff is a false economy: if someone with a PhD and 20 years' experience is doing work that a technician could do, you are in effect paying a technician the salary of a person with a PhD and 20 years' experience. From the point of view of academics, building a research track record is significantly more difficult than in Physics. There is very little time to spare while running courses. Only by attracting research students do you stand much chance of producing significant output. But it's much easier to attract research students if you already have interesting outputs.

So why is this happening? Physics has a cosy position in South Africa. Even in the best funded higher education systems, Physics departments have been closed because they are not attracting students (in the UK, 20 Physics departments have closed over the last decade). Yet in South Africa, universities persist in the 19th century view that Physics is the only real science (captured in Rutherford's claim that "All science is either physics or stamp collecting"). The reason? Physics academics have been very good at putting their case that they are brilliant researchers. No doubt some of them are. But I wonder how well they would do if their lab staff were removed from their undergrad classes. Perhaps we could try the experiment with just one big first year class.

So should Computer Science be cheaper than Physics, even taking all this into account? It's hard to see why. You can teach a perfectly good undergrad Physics curriculum with 20-year-old equipment and lab staff without higher degrees. Computer Science by contrast does not have settled fundamentals, and computers go obsolete fast: few industry users keep a computer longer than 3 years. Worse, the skills needed to set up and maintain computer labs are in heavy demand in industry: look for job ads with keywords like "network administrator" or "server administrator". Add in the job descriptions for the sort of lab staff taken for granted in Physics, and you have a real problem: you need skilled programmers – also a hot commodity in industry.

Should we care?

Aside from the fact that the South African government has made ICT a central component of its economic development plan, there is the general question of how best to spend limited resources. Physics could be run at an acceptable level if a lot less pleasant for academics on a significantly lower budget. Making Physics academics spend more time in undergrad labs would be an imposition on their time but there is an oversupply of Physics academics worldwide, so they would not have much option to protest. On the other hand, freeing up Computer Science academics to do more research could have a very significant effect because there is another variable not yet discussed: Computer Science research is on the whole a lot cheaper than Physics research, because many of the big breakthrough areas in Physics require extremely exotic, expensive equipment. On the other hand, one of the great success stories in Computer Science has been collapsing cost of computing, opening up an increasing range of low-cost research opportunities.

In summary: by pretending the Computer Science is very cheap to teach while Physics is expensive, South African universities are shooting themselves in the foot. It would be much cheaper to set South Africa up as a world-class centre of Computer Science research, than a world-class centre of Physics research. What's more, they could attract major government funding – instead of the current fiasco, where the government is sinking major resources into the CSIR's Meraka institute without much to show for it ... funding which should have gone to the universities.

What about the other "expensive" subjects? Physics and Chemistry are lumped together as Physical Sciences which is in itself a major inaccuracy: Chemistry (along with life sciences) has major expenses of lab consumables, justifying a higher price tag. Whether this makes high-consumable lab subjects inherently more expensive than Computer Science, where lab infrastructure is in effect a consumable remains questionable – but they at least do have a case for being more expensive than Physics or pure maths. On the other hand, these subjects also have an army of lab staff. I remember a biologist at the University of the Witwatersrand some years ago complaining how consequent on funding cuts, they no longer had a person to call on to wash their cars after a field trip. So tough. I wish I had their problems.

Why this apparent trade between Computer Science and Physics? Wouldn't everyone score if Computer Science was moved up the scale? Governments have a notorious inability to scale up university funding to changed circumstances. If a relatively popular subject attracted a higher notch in the funding formula and the overall funding didn't increase, the extra funding for that subject would have to come from other areas of the university. The obvious target would be a subject that has low student numbers and is expensive to teach ... like Physics. (To be fair, I suspect other lab subjects would also have to be cut a bit because Physics wouldn't be big enough on its own.)

By allowing physicists to get away with this unbalanced funding model, they are doing not only Computer Scientists a disservice, but also society as a whole; the differences in research costs feed directly to differences in the ability to make an industrial and societal impact. In Physics, doing something at the level of inventing the transistor requires a massively funded research lab with Nobel laureate-level scientists. The Computer Science equivalent of inventing the world-wide-web requires a commodity PC with a free operating system.

The solution? Universities should be brave and back their Computer Science departments to deliver. If a few other departments would have to pull in their belts as a consequence, they would survive. It's not as if they have anywhere else to go.

Here are a few other related articles in this series:

No comments: