As new grad students flood onto campuses across the US, I just got around to reading this piece in Science from a few of weeks ago about Roald Hoffman's idea for changing the way we support grad students in the sciences and engineering. Most S&E grad students in the US are supported by a mix of teaching assistantships (TAs), research assistanceships (RAs), and fellowships. A typical S&E grad student at an American university shows up and is supported during their their first year by a mix of university funds and pay for teaching. They then often make the transition to being supported as an RA by research funds obtained by their advisor through research grants. (Some remain as TAs - this is more common at large, public institutions with large undergraduate teaching needs.) Some relatively small fraction of S&E grad students are supported instead by fellowships, awarded competitively by agencies like NSF, DOE, DOD, NIH, etc. or by private foundations such as the Hertz Foundation.
Prof. Hoffman suggests that we should move to a system where all grad student support is fellowship-based. The idea is that this will (a) fund only the best students; (b) allow students much greater independence since an advisor will no longer be able to say "You have to do boring experiment #23 because that's what the grant that's paying your salary says we're going to do"; (c) result in better mentoring b/c faculty will no longer view students as "hands". Now, there's basically no way to see how such a drastic change in the system would ever happen, but it's worth looking at the idea.
As someone lucky enough to have a fellowship in grad school, I understand the appeal from the student side. Independence is great - it means that you and your advisor are freed from the stress of worrying that your grant won't get renewed when you're in year 3 of your program. It means that you are a free agent.
However, I think Hoffman's idea would be a disaster, for two main research-related reasons (not to mention the challenge of how you'd handle TA duties at large places that suddenly had many fewer grad students). First, there is little doubt that this would skew an already tilted system even further in favor of the top, say, 20 institutions in the country. Right now it's possible for good researchers at second tier universities to write grants, hire students, and do research. Imagine instead if the only source of student support were competitive external fellowships. It's all well and good to talk about overproduction of PhDs, and say that drastically reducing the number of grad students would be good for employment and salaries. There is a point to that. However, you would effectively end research as an enterprise at many second and third-tier schools, and there are a fair number of really good programs that would go away. Second, since federally funded fellowships would presumably only go to US citizens, this idea would drastically reduce international PhD students in S&E. That, too, would be a mess. Some of our best students are international students, and whether or not they stay in the US after their degrees, training these people is a valuable service that the current US system provides.
It is worth considering other funding schemes, though. I know that in the UK students are supported through their PhD, rather than on a schedule set by external grant deadlines. Perhaps some of my UK readers could comment on the pluses and minuses of this approach.