Hacker News Reader

Comment by sedachv | Hacker News Reader

The Japanese government tried this exact same idea for AI in the 1980s:

http://www.nytimes.com/1992/06/05/business/fifth-generation-...

These 5 year targeted research (IMO an oxymoron) plans are always a waste of money, especially if they involve private for-profit companies. Take that money and spend it on 20 years of grants for basic computer science/robotics research in universities and then you will actually get new discoveries. South Korea's approach will only be effective in grifting tax dollars to the country's industrial conglomerates.

View on HN

They do sometimes work, see the Manhattan project for proof.

But they had a very specific target. It was not a "throw money see what happens" kind of thing

They also had several very specific pathways to the end result: if the calutrons didn't work for refinement, the thermal plant might; if the uranium didn't work, the plutonium might; if the implosion design didn't work, the gun type would definitely work; if a fission bomb didn't work, maybe Teller could jump ahead to a fusion bomb design; etc. Similarly for Apollo: many of the challenges had multiple viable approaches taken in parallel. This is part of why the Manhattan and Apollo projects worked.

In AI, it's not at all obvious what path leads to the end goal, much less there being multiple avenues each with a decent chance of working if enough money is thrown at them.

I keep quoting this, but I really liked Derek Lowe's comment on the similar "moonshot funding" proposal for cancer research by Joe Biden and co.:

"Trying to cure cancer in this way would be like trying to go to the moon without really knowing how rocket engines actually work, without being quite sure if Newton’s laws of motion would hold up, and with some real uncertainty in the position of the moon."

http://blogs.sciencemag.org/pipeline/archives/2016/01/15/moo...

And it wasn't done by throwing money at private contractors either, contractors were strictly overseen and some went oddly above and beyond e.g. on the building of Hanford

> Groves recruited DuPont in November 1942 to be the prime contractor for the construction of the plutonium production complex. DuPont was offered a standard cost plus fixed fee contract, but the President of the company, Walter S. Carpenter, Jr., wanted no profit of any kind, and asked for the proposed contract to be amended to explicitly exclude the company from acquiring any patent rights.

Yes, there was much more patriotic (for better of worse) leadership at the helm of a lot of the corporations (also: the stories of Skunk Works delivering under budget). However, everything I've read about Groves also indicated that he was an amazingly proficient administrator for a very complex project.

Disney was similar during its wartime efforts; all of its operations done in service of the military (some trivial, like ship mascots and badges; others less so like recruitment films, and yet others critical like classified training videos) at cost even though they were offered a standard premium.

The Manhattan Project was run by the US Army Corp of Engineers, not by private for-profit companies.

The incentives in the latter case are all towards grafting maximum funds from the government.

History repeats itself. Did South Korea learn anything from Japan?

To create AGI you need more than money.

I think the situation may be different now that AI is actually working.

For certain values of "actually working", it has been doing so for quite some time.

Reply on news.ycombinator.com