The challenge of the modern scientist is to avoid career suicide

Feb 27, 2014 by Geraint Lewis And Chris Power
Modern science is no longer a solo effort. Credit: Flickr/ only_point_five

Perhaps an Albert Einstein, staring intently at a blackboard covered in incomprehensible equations, or of Alexander Fleming, hunched over the laboratory bench poring over a Petri dish?

The likelihood is that you will imagine the scientist as an individual of great intellect, grappling heroically with nature's secrets and looking for the "Eureka!" moment that will transform our understanding of the universe.

This notion of the individual effort is implicit in the everyday language of scientists themselves. We talk of Newton's Laws of Motion or Mendelian Inheritance. We have the annual pronouncements of the Nobel committee, which awards science prizes to at most three living individuals in each category.

Contemporary popular culture presents us with characters such as Big Bang Theory's Sheldon Cooper, single-mindedly and single-handedly in pursuit of a theory of everything.

But the practice of science over the last century has witnessed a significant shift from the individual to the group, as scientific research has become more specialised and the nature of research problems have become more complex, requiring increasingly sophisticated approaches.

The lone scientist appears to be almost a myth.

The rise of 'Big Science'

Much of science, as it is conducted now, is Big Science, characterised by major international collaborations supported by multi-government billion dollar investments.

Examples include the effort to build the next atom smasher to hunt for the Higgs boson, a telescope to uncover the first generation of stars or galaxies, and the technology to unravel the complex secrets of the human genome.

One of the key driving forces behind this wonderful growth in science has been the similarly spectacular growth in computer power and storage. Big Science now equals Big Data – for example, when the Square Kilometre Array starts observing the sky in 2020, it will generate more data on its first day than will have existed on the internet at that time.

Powerful supercomputers are the tool researchers use to sift through the wealth of data produced by observations of the universe, large and small.

At the same time, they are harnessed to provide insights into complex phenomena in simulated universes – from the way atoms and molecules arrange themselves on the surfaces of novel materials, to the complexity of folding proteins, and the evolution of structure in a universe dominated by and dark energy.

Big Science has resulted in a spectacular growth in our understanding of the universe, but its reliance on cutting-edge computing has presented a number of new challenges, not only in the cost and running expenses of supercomputers and massive data stores, but also in how to take advantage of this new power.

The Big Science bottleneck

Unlike general computer users – who may want to simply check email, social media or browse photos – scientists often need to get computers to do things that haven't been done before. It could anything from predicting the intricate motions of dark matter and atoms in a forming galaxy, or mining the wealth of genetic data in the field of bioinformatics.

And unlike general users, scientists seldom have off-the-shelf solutions and software packages to solve their research problems. They require new, home-grown programs that need to be written from scratch.

But the training of modern scientists poorly prepares them for such a high tech future. Studying for a traditional science degree that focuses upon theory and experiment, they get limited exposure to the computation- and data-intensive methods that underpin modern science.

This changes when they enter their postgraduate years – these scientists-in-training are now at the bleeding edge of research, but the bleeding-edge computational tools often do not exist and so they have to develop them.

The result is that many scientists-in-training are ill-equipped to write software (or code, in the everyday language of a researcher) that is fit-for-purpose. And just like driving and child rearing, they are likely to get very cross if you attempt to criticise their efforts, or suggest there is a better way of doing something.

This systemic failing is compounded by a view that the writing of good code is not so much a craft as a trivial exercise in the true effort of science (an attitude that drives us to despair).

For this reason, it is probably unsurprising that many fields are awash with poor, inefficient codes, and data-sets too extensive to be properly explored.

Coding the future

Of course, there are those to whom efficient and cutting-edge coding comes a lot more naturally. They can write the programs to simulate the Universe and take advantage of new GPU-based supercomputers, or efficiently interrogate the multi-dimensional genomic databases.

Writing such codes can be a major undertaking, consuming the entire three to four years of a PhD. For some, they are able to use their codes to obtain new scientific results.

But too often the all-consuming nature of code development means that an individual researcher may not uncover the major scientific results, missing out on the publications and citations that are the currency of modern science.

Those that can code are out of a job

Other researchers, those that just use rather than develop such codes, are able to reap the rewards, and this better paves their way into an academic career. The rewards go to those that seek to answer the questions, not those that make it happen.

With fewer publications under their belt, those that develop the tools needed by the scientific community find themselves pushed out the market, and out of academia.

Some senior academics recognise this path to career suicide, and young researchers are steered into projects with a more stable future (as stable as academic careers can be).

But we are then faced with a growing challenge on who will develop the necessary tools for Big Science to continue to flourish.

How to grow an early scientist

So, what's the answer? Clearly, science needs to make a cultural change in understanding on what makes a good modern scientist.

As well as fertilising links with our computer scientist colleagues, we need to judge early scientists on more than their paper output and citation count. We need to examine their contribution in a much broader context.

And within this context, we need to develop a career structure that rewards those who make the tools that allow Big Science to happen. Without them, supercomputers will groan with inefficient code, and we are simply going to drown in the oncoming flood of data.

Explore further: Exploring the dark universe at the speed of petaflops

add to favorites email to friend print save as pdf

Related Stories

Exploring the dark universe at the speed of petaflops

Nov 21, 2013

An astonishing 95% of our universe is made of up dark energy and dark matter. Understanding the physics of this sector is the foremost challenge in cosmology today. Sophisticated simulations of the evolution ...

A toolbox to simulate the big bang and beyond

Oct 18, 2013

The universe is a vast and mysterious place, but thanks to high-performance computing technology scientists around the world are beginning to understand it better. They are using supercomputers to simulate ...

Recommended for you

Water window imaging opportunity

3 hours ago

Ever heard of the water window? It consists of radiations in the 3.3 to 4.4 nanometre range, which are not absorbed by the water in biological tissues. New theoretical findings show that it is possible to ...

User comments : 13

Adjust slider to filter visible comments by rank

Display comments: newest first

shavera
4.8 / 5 (6) Feb 27, 2014
Eh, speaking as a physicist who moved into software just to move back into scientific software development... I'd say the big picture problem is "too many PhDs, too few research positions." We need some people who are the best at pouring over the data to pour over the data. But the code they write is so god-awful sloppy most of the time, it helps for those of us who aren't as good with the "research proper" to get better training in programming and form good infrastructure for the others.
Bonia
Feb 27, 2014
This comment has been removed by a moderator.
Bonia
Feb 27, 2014
This comment has been removed by a moderator.
Bonia
Feb 27, 2014
This comment has been removed by a moderator.
Z99
5 / 5 (2) Feb 27, 2014
My guess is that its a temporary problem. Some aspiring CS Doctoral student will create a "state-of-tje-art" assisted code development algorithm, implement it, and we'll go back to thinking up the questions rather than implementing the code to answer them. Machines are already forming hypotheses, implementing (yes:"implementing") experimental design, and making conclusions (all nascent, admitedly). Compilers are so darn effiecient now, its hard to justify learning lowere level languages for most computational tasks, this will just get better. Oh, and did I mention outsourcing this type of problem (opportunity)? One thing great about programming is that, in theory, the resources needed to write great code is affordable to any one...specifically the third world. Reminds me of why in the old USSR days, the Soviets were such a great mathematical scientists...all you needed was a chalkboard.
Captain Stumpy
5 / 5 (7) Feb 27, 2014
For those who don't believe me - this is how LENR workshops and conferences looks like ...Just show me some young people there...

@zephir
cold fusion is a dying field due to its chequered past and its inability to show any results.
I have tenure, so I don't have to worry about my reputation

this truly bothers me to no end...you mean to tell me that you actually are tenured at an institution of higher learning? do you still teach?

given your lack of ability to show empirical data supporting your pet philosophy, I would not have taken you for a real scientist.
you should know better...
are brave enough for to pursue new routes of research

and yet it appears to me the only way to truly make your mark in physics is to explain something better, find a new hypothesis/theory, and explore the unknown...

I would have thought that all the aged reaching for the fringe ideas was more about stroking the ego or recognition/publicity etc... just IMHO
Bonia
Feb 27, 2014
This comment has been removed by a moderator.
cantdrive85
2 / 5 (7) Feb 27, 2014
Only these established ones or those aged ones (who already have nothing to lose) are brave enough for to pursue new routes of research.

Then there are the Hoyles, Arps, and Laviolettes of the world...
Bonia
Feb 27, 2014
This comment has been removed by a moderator.
antialias_physorg
5 / 5 (8) Feb 27, 2014
Enjoyed the article. Mirrors my own experiences pretty well. Coding is a (necessary) tool. But in the end whether your code is good or bad doesn't matter: it matters that it does the job.
And the job of a scientist is getting paid for is NOT primarily writing code but putting their ideas to the test. Developing good coding skills takes time - time that researchers don't have.

Shortly before I started my PhD they had a part time programmer who was in charge of 'cleaning up the code' of all the other PhD students before me. Unfortunately she went on maternity leave and never came back (and eventually there was no budget to hire someone else) . But continuing that practice would have been ideal.

I have tenure...

I think my BS-meter just exploded.
Maggnus
5 / 5 (3) Feb 28, 2014
I have tenure, so I don't have to worry about my reputation.
I call bull! Prove it Zephyr.

I think my BS-meter just exploded.
Lol ya mine too, along with my keyboard from the snort of disbelief!
Maggnus
5 / 5 (4) Feb 28, 2014
Then there are the Hoyles, Arps, and Laviolettes of the world...
Yep, even very smart men can be wrong.
shavera
5 / 5 (3) Feb 28, 2014
Come on guys, if zephyr hasn't earned tenure as most well known crackpot on the internet I don't know what else he can do to earn it. He's certainly met the crackpot publishing criteria pretty well.
Q-Star
5 / 5 (5) Feb 28, 2014
I have tenure, so I don't have to worry about my reputation.
I call bull! Prove it Zephyr.

I think my BS-meter just exploded.
Lol ya mine too, along with my keyboard from the snort of disbelief!


Hey, don't be so hard on Zeph,,,, that's a quote he has posted countless times, on countless forums, he just never attributes it, just links it. I think the first time I saw him use it was over at reddit or physics forum. I might be wrong but I think the actual person who was speaking/writing was George Miley.
Maggnus
5 / 5 (3) Feb 28, 2014
Hey, don't be so hard on Zeph,,,, that's a quote he has posted countless times,
Well knock me over with a quill - my apologies Zeph, I completely misread that quote. You're absolutely right Q, that is a quote from George Miley.

And yes, Zephyr has posted it all over the place, I should have recognized it.
peter_trypsteen
5 / 5 (3) Mar 02, 2014
The answer is simple have researchers collaborate with programmers. Talking about multidisciplinary work and big teams without mentioning hiring programmers as a solution is a bit dumb though.
Institutions who have this problem might want to look into hiring a professional computer programmer to speed things up.
antialias_physorg
5 / 5 (4) Mar 02, 2014
Institutions who have this problem might want to look into hiring a professional computer programmer to speed things up.

While I agree that that should be the case I don't think it's going to happen. Research institutes have VERY limited funds. And you must be aware that for the price of one (average) programmer you can employ 3 or 4 PhD students (or 2-3 postdocs...or one tenured professor).
ralph638s
1 / 5 (1) Mar 04, 2014
So, physicists are cheaper than real software engineers. Now I understand why the Wall Street boys hire "physicists" to design their complex financial products...