If you build the chassis out of molded plastic I’m not using it.
2 Likes
In that case, the robots need to be reprogrammed, and a human needs to do the reprogramming. AI isn’t magic. Machine learning isn’t even learning, it’s just a set of statistical models, basis functions etc. that can hopefully approximate the problem space in a stable way without overfitting to the data. Nothing matches humanlike intelligence, especially as it relates to pattern recognition and problem solving.
I asked about what happens if the task becomes obsolete. As in, the whole task. You can’t simply teach a chess AI to play checkers, you need to design a checkers AI from (mostly) bottom-up instead.
Robots are currently just tools that make human jobs easier. They don’t actually replace humans outright. Humans are still required to maintain the robots. Maybe not the same humans, and certainly not without extra training, but we can’t take the human out of the loop yet, or maybe even ever.
4 Likes
It’s been forever since I read it, I just recall they had a ban on computers and used special people(autistic people?) to crunch numbers and do complex math problems…
3 Likes
They did that after the AIs ran everything for humanity, lessening their “humanity,” and then the Orange Catholics led an anti-machine crusade to smash the AIs and return humanity to its natural state. Only Ix continued to deal in advanced machines…
5 Likes
Interesting trying to define AI though.
2 Likes
Pfft, not likely; the reasons they are followers of the drumpf are the very reasons they will merely shake their fists at their tv’s even as the last job in their region disappears.
4 Likes
Ultimately when self-driving cars take over and are networked, accidents will tend towards zero (let’s face it, more than 90% of accidents are actually on the human error-stupidity spectrum.) Which means that cars will not need to be designed for frequent repair. The main reason for using pressing rather than moulding is that, compared say to a mobile phone, car sections are mostly extremely thin, though Smart did experiment with injection moulded panels.
It’s possible that small future city cars could be injection moulded over a magnesium alloy frame for longevity, lightness and strength. You’ve just eliminated robot welding and robot painting. Mind you, I’ll try most things in the transport line but I wouldn’t drive a Smart in the US because, US redneck drivers.
5 Likes
This ignores the possibility of printing the panels without any dedicated tooling, or at least printing the tooling. There’s more friction towards thinking of general purpose AI’s and general purpose fabrication, a la Diamond Age’s Matter compilers or Trek’s Replicator, than I would have thought from such a crowd. I’m no computer scientist but I can’t imagine why a powerful AI couldn’t be programmed to play both chess AND checkers without a hardware redesign any more than a current PC needs different hardware to do CAD & desktop publishing. Things change, my 1st machine, an XT clone, needed a $200 math co-processing chip added to be able to run CAD software.
1 Like
Well I’m safe from automation for the foreseeable future; it’s hard to imagine a robotic bike mechanic 
4 Likes
It should be pointed out that of that list of “safe” jobs, the Java application developer, internet security specialist, statistical analyst, and data mining specialist can be outsourced right now. There’s nothing particularly special about my talents that prevents well educated foreigners being hired for 1/4 the cost today. The only reason it hasn’t happened is that those skills are still in fairly short supply. But that is eminently fixable.
The reality is that we’ve had an extraordinarily good run and lived like kings for last almost 100 years. But it was an fluke heavily built on being given a huge depopulated continent with no existential enemies as direct neighbours. We’re still filthy rich, but I think we have to accept that our standard of living is going to drop, eventually meeting the rising standard of living everywhere else. Once we’re all earning a decent $10-20K/year (excepting the billionaires, of course), then we might see wide spread, global, and for the first time, fair income growth.
Alternatively, by that time, robots may turn from hype to reality, in which case 98% of humanity becomes about as economically viable as horses. And since super-men and women don’t breed true, you can’t even have a Libertarian robotic paradise of the super-elites. Statistically their kids won’t have the talent/brains to be economically viable.
5 Likes
Programming.
Or are you saying they’re not smart enough?
I don’t understand the view that automation isnt coming for your job. It’s coming for everyone’s, with the only question being timescale. Let’s be abundantly clear about this: There is no limiting principle that makes a computer incapable of performing the same tasks as a human. There may be physical limits on processing power and memory, but the vey existence of human brains demonstrates there is at least one way to fit that into something about the size of a skull.
People are perhaps confused in some part because machines have not yet invented other machines. At least one supercomputer was able to indepedently determine that F=ma from observations of a pendulum. This is the foundation of classical physics. There is no reason for this time not to come. Humans are not special. This has been perhaps the singular consistent determination of every science through history. Astronomers discovered that very little revolves around us, chemists that we’re not made of anything special, biologists that we’re simply more sophisticated primates.
Why should software engineering and robotics discover anything different?
9 Likes
I would require pretty good evidence of that. How did it come up with the idea of a pendulum? How did it make it? How did it sense its behaviour? Did it place optical sensors itself or wind the coils for a magnetic pickup?
HP used to have a calculator program that took data as an input and came up with a best fit equation; it didn’t need a supercomputer (they are mostly designed for massively parallel tasks like FEA, in any case.) I used this once to infuriate someone; long story short he had been telling someone else that his judgement was better because he had an MSc; I took the data from his thesis and fed it into the HP program, which in minutes had the (moderately complex) equation that he had spent months working out. But that isn’t AI in any shape or form, and it was running on something with a clock speed in the low megahertz.
1 Like
Science 03 Apr 2009:
Vol. 324, Issue 5923, pp. 81-85
DOI: 10.1126/science.1165893
And the article immediately following it. Company website for the creators of the AI: http://www.nutonian.com/
They fed a computer data and it used evolutionary search to determine the laws. Apparently they used a Cornell supercomputer, but the software itself may be able to run on a number of machines. To be clear, I’m not saying that a robot got curious and started playing with pendulums, I’m saying that a set of algorithms tasked with discovering general laws from data did so by iteration to find general equations of motion, lagrangians, and hamiltonians. That’s not the same as curve fitting. It did so on the order of minutes to hours, depending on the kind of problem.
This software isn’t replacing anyone, but the building blocks come first. The mistake is in assuming that current technological limitations are future technological limitations. The 47th or 159th generation of this robot is going to be what replaces people doing that kind of work.Autopilot and ILS landing didn’t replace pilots, but drones are going to, and there’s a plausible argument that they already have.
1 Like
But surely you agree that your statement of what happened is very misleading:
“observations of a pendulum” seems pretty declarative to me.
I’ve had so many vendors try to buzzword me into accepting a massive overstatement of what their product can do in “AI” terms that my bullshit detector needle is permanently wrapped round the support pin at the end of the scale where AI is concerned.
From your link:
“Developed in Cornell’s Artificial Intelligence Lab by two of the “World’s Most Powerful Data Scientists” (Forbes), Machine Intelligence leverages an “evolutionary” approach to model creation, testing billions of potential models per second, and converging on the simplest, most accurate ones that explain your data. M.I. makes no prior assumptions about the data set, instead fitting models to the data dynamically.”
In short, it’s still a kind of curve fitting on steroids using blunt compute power on a pre-existing data set. And who produced the data? Hint: not the computer. Though in this case, I suggest it is you who overclaimed.
1 Like
There are some differences between capability and practicality. Emulation can involve making a map which is to scale larger than the territory it represents. Human cognition can be modelled, but it comes at a huge expense because binary models work unlike brains. It is tempting to lump all kinds of signal processing together and solve them with the same tool, but different tools will be better natively suited to their own distinct domains.
The efficient path to AI I think is not to copy humans, which serves no purpose other than ego, and pursue what the actual strengths of autonomous digital systems may be.
Even the idea of “everybody’s job” suggests a quaint primate hierarchy of society. If you have your own job, your own goals, then you decide what tools are used to achieve it. What one does with one’s time and effort is only the concern of anybody else to the extent that one negotiates their explicit involvement.
1 Like
I think you’re making far too much of the word “observations.” It’s data. Robots collect data all the time, in this case it was fed to the computer. There is functionally no difference. What humans do is also often reducible to “curve fitting on steriods,” depending on the level of flippancy desired.
3 Likes
[edit - there is a big difference between “collecting data” (i.e. active) and “fed to” (passive).
I disagree. Any fool can be fed data, progress comes from knowing what data to collect and how to collect it. There is a big functional difference. Structured data results in information. As an example, it’s worth reviewing how CERN collects data from the big experiments. The pinch point at which the data stream becomes manageable is the one at which physicist-devised rules allow the system to identify “interesting” events and discard uninteresting ones.
I spent more or less the last ten years of my career working on a large scale data collection system in which, to manage the data transfer and to get information out, an exact set of rules (which have to be kept updated as the monitored systems change) defines what data is requested from remote stations, how it is transformed and how it is presented. That is the functional difference; and I worked out the system when asked to devise it basically by working through the CERN literature.
3 Likes
Can’t say to smarts, but certainly they haven’t the education, nor the inclination to education for the most part. Calling someone stupid when they’re ignorant is like calling a car without gas slow, it may well be but you’ll never know till you try and fill it up.
6 Likes
