Comment by gumby | Hacker News Reader

Comment by gumby | Hacker News Reader

This entertaining essay captures things well without getting distracted by some of the really funny stories, many of which cannot be repeated as the protagonists are still alive.

Programming in Japan is typically a low-status job considered a blue collar occupation (Nintendo and SCEI are notable exceptions) despite Japan having produced quite a few globally notable computer scientists. I remember visiting a 5 Gen project om 1984. I was shocked by the open plan office (how did anyone get any work done?). At the row of desks was one ASCII terminal for each two desks (to a KL-20 I assume but can no longer remember). I spoke with one of the developers asking him what he thought of the environment. "Great: I only have to share the terminal with one other person".

The next year I myself was employed at MCC in their AI group (they had a database and CAD group, and perhaps one other) working for Doug Lenat and designing the Cyc system (when I got there they were working in Interlisp on Dandelions, not surprisingly as Doug's office had been around the corner from mine at PARC). The first thing I did was toss the Interlisp implementation and redo it on for a 3600.

5Gen and MCC actually were 2 and 3 of another transformational hype effort, the Center Mondial d'Informatique and Ressource Humaine (World Center for Computer Sciences and Human Resources) launched in France in 1983 on the back of an influential book by a major French intellectual, Jean-Jaques Servan-Schreiber. They had a KL-20, a Vax (with a copy of BSD I carried over there in my luggage) and a few CADR Lispms (two or four, I can't remember). With swanky offices in central Paris around the corner from the presidential palace and a lobby full of Apple 2s that anyone could walk in and try out they didn't do much for the third world I'm afraid, idealistic goals notwithstanding, but they did launch quite a few very good computer scientists, so were probably worth the effort.

So I worked for CMIRH and MCC, and spent some time at 5Gen. Fun times.

A funny thing is that both MCC and the Fifth Generation Project were fairly unsuccessful at actually launching the future. My hypothesis is that it's because they were trying to plan research, which is to say, come up with a plan up front for what they were going to do, and then carry it out, rather than improvising based on what was learned in the process. I suspect the massive amount of planning and investment required by modern drug approval is a major cause of the drug industry's dramatic lack of progress in recent decades. Or, as the Goblin Queen put it in her comment below, when industry takes over science, we lose the science.

The 1980s computer systems research projects I think of as the most successful were dramatic advances in semiconductor fabrication, TCP/IP, microcomputers, RISC, Macintosh, Emacs, Postgres, HDLs, GCC, SGI, MOSIS, SUN, TeX, Perl, Plan 9, Cedar, generational GC, OOP, NASDAQ, Pixar/CGI movies, and wavelength division multiplexing. (Maybe my vision is a bit limited and provincial, and I'd be interested to hear what others think.) A lot of these were government-funded, but not with top-down goals. (It might sound ridiculous, but I suspect Perl may have been the most important advance to come out of JPL in the 1980s.). Others were industry-funded. Two were mostly funded by one stubborn idealist, at least until you got involved.

As someone who's worked both on very successful research projects (BSD and GCC) and unsuccessful ones (Cyc) as well as biomedical research, what do you think about what makes the difference? Was it predictable in 1984 that Japan would fail?

I dunno if it was clear or not, but it wasn't really industry driving the research: it was the various governments, and, as you say, they were planning things. Well, sort of. That's not what a real technological plan is: in real life when we try to build advanced things, we identify the lacunae and develop a plan to fix those. Just like Apollo and the Manhattan project. This one was a giant punt into "shyeeet maybe this will work." Sort of like all fusion research to date, except worse, because unlike fusion, the gaps in building a 5th gen computer are not even identifiable. It might as well be "we plan on being lead in the antigravity industry by 2030."

Thanks for working on BSD and GCC (and Cyc, which is also pretty damned cool)!

Sorry, it wasn't me that worked on BSD, GCC, Cyc, and biomedical research, it was Gumby.

It seemed likely to me that Japan and in particular MITI couldn't continue on as they had been but I was an outlier and so thought I was probably wrong. Don't forget the US was just emerging from the stagflation of the 70s and while they have addressed that one macroeconomic phenomenon haven't made any progress on any of the things that lead to it.

So having said that, 5Gen was a "moonshot" from a bunch of bureaucrats who could see the writing on the wall, but mistakenly let it get hyped. Compare that to ARPA who used to place a whole bunch of batshit crazy bets that were clearly nonsensical though being done by legitimately smart people: that gave us the Internet, among other things. They also funded infrastructure to get there (e.g. MOSIS).

Since then (the now named) DARPA and corporate labs have lost interest in much blue sky work. "I don't agree that "when industry takes over science, we lose the science." Corp research gave us the transistor, semiconductor, SEM, WIMP interface etc. It could again. Pharma is one of the few fields where fundamental research is still done.

(I wouldn't say regulatory approval has had much of any influence in the cost or output of pharma research BTW; getting meaningful results is hard and if you want drugs that are better than the standard of care you need to do that work. I do think the regulatory framework around marketing and reimbursement has had a much larger impact on which API candidates get advanced into programs.

I'm biased in what I consider fundamental; I'd agree with you on TCP/IP, mpus, RISC, the Mac, generational GC (another corp research result BTW) and fab. (WDM is another corp invention about 100 years old). Some of the others I consider less fundamental or less influential (Cedar? CLU was far more influential even if no real programs were written in it).

The fact is research is hard, and skills in basic research and the skills of translational research are rarely present in the same person.

Thank you very much!

Do you think the ARPA approach is better, or better for certain cases? The accounts I've read of the invention of WIMP and transistors resemble “bunch of batshit bets by smart people”, if I have the story right, but I don't know anything about the SEM and semiconductors. (Weren't semiconductors Bose in the 1890s?)

A lot of my list clearly was pretty applied; I didn't mean to suggest they were all basic.

Cedar in particular I mentioned because it inspired Oberon, Java (né Oak), in some sense Xi (and ropes in lots of other systems), and maybe sam and the plumber, though maybe that was convergent Blit evolution. But certainly that impact is small compared to the LSP. (Were Smalltalk iterator methods convergent evolution or were they taken from CLU?)

I'm very surprised at your view on pharma. Sounds like my opinions need to change.

Well, life is a huge parallel process that requires a diversity of inputs, so you need blue sky, "stupid" stuff, boring spadework and directed development, all at once. (and those definitions change; as fields go through their "stamp collecting" modes, as taxonomy/biology did in the 18th and 19th centuries, that was exciting. High energy physics is stuck there at the moment, again).

So yes, we do need someone like the ARPA of old. Licklider, Bob Taylor, and the like are heroes in my opinion and I don't really see their like today.

I worked at MCC too, but in the human interface group. It was a lot of fun using Symbolics machines all day (the ultimate brain-to-computer interface). Despite nothing much coming from my work at MCC, I still got to write Lisp and Prolog for pretty much my whole career, and MCC set me on that path. So I'll thank them for that.

I'm also an MCC alum - I worked in the Software Technology Program. My code there had a surprisingly long life - transitioning to a spin out startup and then other companies. (See gIBIS https://en.m.wikipedia.org/wiki/Compendium_(software)#Histor... ). I too learned Lisp (and Emacs!) there - two tools that have stuck with me ever since. Many of us also started to learn Japanese there.

I am glad to hear of good MCC outcomes!

If you don't mind my asking, what are you using Prolog for these days?

I've only done one basic course on it a long time ago and really liked it but I don't think there are many places that use Prolog nowadays.

Querying graph databases. Prolog is a lousy data storage engine but it's a fantastic query language--much better than SQL, IMO.

Thanks for sharing that. After visiting the MCC offices when you were just getting started, I always maintained a curiosity about your work, enjoyed playing with Open CYC, etc.

Reply on news.ycombinator.com