Are we trying to do things the hard way?
A TED talk about remarkable low-cost Indian products—the Tata car which costs $2000 and is a real car, a $28 artificial lower leg which permits walking on rough ground, tree climbing, jumping, and running, and fast cheap drug development which starts with traditional Indian remedies. It’s an example of something to defend because the effort is to develop products that very poor people can afford, so that incremental improvements and cost-cutting aren’t good enough.
It leaves me wondering whether the process of creating FAI should be re-evaluated—whether there’s a built-in assumption of high personal costs which is unnecessary. That’s wondering, not an absolute certainty, it’s just that the $28 artificial lower leg shocked me into thinking about how much is being made harder than necessary.
Even if FAI is being worked on about as efficiently as possible, there may be a huge amount of possibility for making things easier in life generally.
It’s a talk filled with problems. The overarching problem is the halo effect. India is this guy’s home team, and he is committed to cheering for his team.
This leads him to overlook pretty much every qualifying factor. I was riding along with his viewpoint comfortably until, halfway through the section on the Nano, he said “adhesives instead of rivets.” And I said “Oh damn, there is a serious tradeoff going on here that he is glossing over.” And from that point on the magic was lost, I started questioning what he said. Once you correct for the fact that he is intentionally trying to get you to believe false things, his talk gets reduced to “here, look at some hand-picked success stories of developing cheap products.”
Not developing high-quality products cheaply. Just developing products cheaply.
AI research is probably the wrong thing to apply the (possible) lessons of these success stories to. Quality is essential, or else bad things happen. It’s new research, with big limitations on what you can do with prior art. I would agree that FAI is probably being done inefficiently (at least form the perspective of someone who knows the answer), but that doesn’t mean this is a way to do it better.
What’s the problem with adhesives vs. rivets?
Is your concern that only a few good cheap products get developed, while there’s either a lot of effort wasted on trying to meet irrationally low prices and/or creating poor quality products?
After I’d gotten a bunch of upvotes with no comments, I was wondering if I was in applause light country.
I upvoted without comment (there’s a self-negating statement!) because I want to see other people’s answers, in case there were things I’m doing the hard way.
Holding things together with metal is generally a more durable solution. There are some factors that can mitigate this, but in something like a car… well, maybe he was talking about interior bits and not the guts or body. But he gave me the impression that this replacement was widespread, which would be a strong indicator that the car was a “real car” in the sense that it has four wheels and an engine, but not in the sense that it showed other cars were “high cost.”
I’m not very concerned about problems from this, even were it to become widespread—capitalism is pretty good at what it does. My point was that I don’t feel like this demonstrates that we’re stuck doing things the hard way; it at best opens up a new market niche and at worst, doesn’t. It seems to be just normal capitalism making tradeoffs to explore the market, even if the speaker sometimes hides the tradeoffs.
I think we need information about whether adhesives have improved enough to be satisfactory for structural joins in a car.
Capitalism (whether normal or not) is affected by what people believe is possible and worth doing. I believe that opportunities to make profits frequently don’t get noticed for a while until someone thinks to look in that direction.
I strongly agree with the general point, but isn’t the idea of building a brain in a box in a basement already wildly cheap?
I don’t think NancyLebovitz means “cheap” in the material resources sense. The idea of finding the “key” to general intelligence is cheap, as is any other idea. The expertise to execute correctly is the expensive part. Is there a cheaper way to go about gathering the ability to do it, or a way to do it for which the ability is cheaper to garner? Whole brain emulation may be an example of the latter.
Thing is, we’ve been doing cheap AI for decades. It’s more often called narrow AI, and it shows up in everything from video games to OCR to spam filters: do just what is needed to accomplish a specific task at minimum cost.
The idea of AGI is to make a bigger investment in order to solve a wider range of problems up front, instead of having to redo everything by hand in every particular project.
I’ve heard about the $2000 car before. It can get you from place to place, but it doesn’t meet U.S. safety standards.
A big reason we do things the hard, expensive way is that government tends to make low-end choices illegal.
The Tata Nano (and automobile safety features generally) is a great example. What generally happens is that firms introduce advanced safety features in luxury cars and they generally become widespread. However, low-end cars don’t have them yet. The government declares these cars “unsafe” and makes them illegal, even if they were safer than the average of only a few years ago.
There are countless examples of low-cost, low-end choices being made illegal, like mini-meds under PPACA . I suspect the $28 leg wouldn’t pass regulatory muster in the U.S. either.
Applying something like this to computing, I’m reminded of the stuff Chuck Moore has done with Forth. Here’s one article about the sort of radical minimalism he does, and about some problems with it.
He has recently designed processors, but I haven’t looked into them enough to tell if they are a similarly strange and minimal as his software.
I’ve skimmed the piece about Forth. I now realize that I’d wildly underestimated the difference a significant increase (maybe 50 IQ points) in typical intelligence would make.
In software engineering and related to Forth, there is a huge and not always appreciated gap going from programs which can be managed by a single individual to programs that need multiple programmers to develop. The already larger complexity of the program that requires many programmers is further compounded by the fact that none of the programmers comprehend the entire thing as well as a single programmer would, and therefore can’t do all the good design decisions that take the entire system into account that a single programmer would.
A Forth story from 1995 is about this. The author was massively productive developing in Forth through the 80s, but only as far as he was able to be the single principal developer of his aggressively whole system optimized programs. Once the 90s came, he realized he could no longer manage all the complexity contemporary software required, and decided that Forth wasn’t a viable option compared to languages that provided more fixed idiom for multiple developers.
I’m not sure if working memory, IQ and the end result in system complexity are quite aligned here. Just being able to fit more knowledge of the program in your head seems like it should help, even though your IQ stays the same (then the bottleneck becomes that a single programmer can type only so fast, so maybe a hyperbolic time chamber would help). IQ will let you discover compact designs that may subsume more laborious solutions, maybe there’s a single clever design that is a generalization of all the 17 special cases the programmer with the inferior design needs to implement by hand. (Or not. Some problems can be just plain complex. Sometimes the problem isn’t even in code, you necessarily need a Byzantine business logic program to model Byzantine legislation, but now the problem starts with the design of the law, not with the design of the program.) But high IQ can also help a programmer keep juggling all the complexity that emerges from a design that should have been strangled in the crib, when another programmer would have become overwhelmed faster and started looking for simpler ways to do the thing. High IQ can also create straight up signaling perversities with complexity, like how we use natural language with needlessly complicated grammar to gauge and signal intelligence.
When designing systems, the power of the abstractions you use limits the pool of programmers you can use who can work with those abstractions, and the amount of complexity you have, generally the less abstraction, the more complexity, brings you up against the hard limit of the system no longer fitting in any single individual’s head, after which you’ve got a complex in your hands and you’d better have a really clever idea of which joints in it to carve to keep the complexity from exploding into a ball of horror where semicommunicative subdevelopers duplicate each others’ effort.
I keep thinking that there should be a science of this.
Possibly relevant: Physicist Marcin Jakubowski is trying to develop a bootstrapping industry where you can build industrial artifacts like farm machinery using a production chain that’s completely open source and can be implemented locally without commercial entities being necessarily involved.