Hacker News Reader

Comment by wonderwonder | Hacker News Reader

I think a very interesting aspect of general AI is that while an incredibly complex technology, it is not unrealistic that it could be created for the first time in someone's home office. Unlike many other earth changing technologies there is nothing that a massive corporation has that a home tinkerer does not (besides the obvious of money and many engineers).

With the rise of cloud computing and open source; everything I need I have instant access too, all that's lacking is the core software which can be written (of course not a trivial task).

While unlikely, it is still quite amazing that in a few years an AI could awaken 3 doors down at my neighbor Bob's house. No idea what happens after that, hopefully Bob was a fan of the 3 laws and has a couple more up his sleeve.

View on HN

     [X] Open Source Tools
     [ ] Massive Data Sets
     [ ] $Millions in Computing Resources

I'd put it roughly on par with finding a general cure for cancer. While unlikely, quite amazing that the cure for one of the largest causes of death could be solved 3 doors down with lab supplies from amazon and a handful of mice.

Well the data sets are all around your house and yard. Think of baby human agi's.

And the computing costs could be boot strapped by the ai. Put it to work earning its keep and expanding resources.

> Put it to work earning its keep and expanding resources.

It's called Google, and is already filthy rich and has a ton of computation and privileged data at its disposal.

I think wonderwonder's point is that the massive data sets and millions of dollars in resources might not be necessary. I don't have an opinion on the matter, but what if AGI falls out just from a small number of critical ideas which while being complicated and non-obvious are nevertheless straightforward to code?

The computing resources are getting cheaper. There was an estimate that to get something roughly equivalent to a human brain you'd need ~100 teraflops and currently you can now get a 12 teraflop GPU from Nvidia for $1200.

Part of the reasoning for the Kurzweil turing test by 2029 type of stuff is that once human level hardware falls in price to hobbyist levels, loads of people will hack away and someone will figure it out.

There is a general cure for cancer.

Look up interdiction of telomere lengthening. Different groups are looking into how to sabotage telomerase and ALT mechanisms. If both can be achieved, then any cancer can be shut down. Those are the only ways to lengthen telomeres, and cancers cannot live without them.

Finding an ALT drug candidate is as simple as running assays on the drug libraries; the assay hasn't existed for long, which is why this hasn't been done yet in any major way. The SENS Research Foundation raised $70k last year to run a preliminary scan of a few thousand compounds. That's about what it costs these days.

So not quite garage science yet, but getting close.

You're vastly underestimating the difficulty in curing cancer. It is very easy to kill cancer cells. The problem is that we want to target cancer cells and only cancer cells. There have been several attempts to kill cancer cells by withholding key ingredients in necessary metabolic pathways, only to find that the cancer cells do a better job of scavenging those than the non-cancer cells.

2 and 3 aren't really restrictions. I can download multiple libraries worth of books in an afternoon. I can buy a PC for a thousand dollars that is basically a supercomputer compared to computers 20 years ago, or even 10 years ago.

The biggest reason I suspect it will be solved by one person is it's "just" a math problem. Most of the hardest problems in mathematics were solved by a single person. Building off of the works of others of course, but rarely in some big team or corporate project. 9 women can't have a baby in 1 month. The research process of big corporations or even academia isn't really amenable to actually solving hard problems.

Cloud computing will not help home tinkerer much if you need 100K/hour worth of compute to run an experiment.

If we move towards hierarchical model composition, you don't need to rebuild the visual object recognition module to experiment in learning spaces that incorporate visual information.

The Wright brothers were bicycle mechanics

Reply on news.ycombinator.com