Slashdot Mirror
When Things Start to Think
One underlying theme dear to Gershenfeld's heart is the death of traditional academic distinctions between physics and engineering, or between academia and commerce. Applied research is real research.
Another major theme is that older technologies should be treated with respect as we seek to supplement or replace them. For example, a laptop's display is much harder to read in most light than the paper in a book.
The book starts by drawing a contrast between Digital Revolution and Digital Evolution. Digital Revolution is the already-tired metaphor for universal connectivity to infinite information and memory via personal computers, the Internet, etc. Digital Evolution describes a more democratic future, from Gershenfeld's point of view, when computers are so smart, cheap, and ubiquitous that they do many ordinary chores to help ordinary people. When things talk to things, human beings are set free to do work they find more appealing.
"What are things that think?" asks the first section of the book.
Gershenfeld's whizbang examples won't be big news to Slashdot readers. My favorite, the Personal Fabricator, ("a printer that outputs working things instead of static objects")-- whose relationship to a full machine shop analog is like that of the Personal Computer to the old-fashioned mainframe. Gershenfeld actually has one of these in his lab (it outputs plastic doohickeys)--seeing it was one of the high points of my visit there.
"Why should things think?" asks the second section.
My favorite here is the Bill of Rights for machine users. (In true Baby-Boom style, it's of list of wants arbitrarily declared to be rights.) "You have the right to
Have information available when you want it, where you want it, and in the form you want it
Be protected from sending or receiving information that you don't want
Use technology without attending to its needs"
Under the heading "Bad Words," Gershenfeld offers a snide but useful summary of many high-tech pop-sci buzzwords, showing how they get misused by people who don't understand their real content or context.
"How will things that think be developed?"
By making them small and cheap. By getting industry to pay the bills for targeted, practical research, using the Media Lab model TTT ("Things That Think.") By reorganizing education on the model of the Media Lab, where students learn things as they need them for practical projects, not all at once in a huge, abstract lump.
The book concludes with directions to various websites, including the Physics and Media Group (One of their projects these days is "Intrabody Signaling.") Slashdotters might also be interested in Gershenfeld's textbooks The Nature of Mathematical Modeling and The Physics of Information Technology.
You can purchase When Things Start To Think from bn.com, and Amazon has the book paperback discounted to $11.20. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.
I mean, come'on. We have pattern recognition, and bots that have huge libraries of information. We aren't anywhere near true AI, and won't be for several decades, unless some huge breakthrough occurs in learning algorithms.
I think we are going to look back a hundered years from now and say how silly we were to ever believe computers could think like we do.
How is a computer program ever going to adopt abstract thinking and creativity? Is a computer program ever going to invent mathematics without previous knowledge of it just because it finds it to be a useful utility for solving problems?
Heck, if someone could write a decent language translation program I might think there is a hope.
Humans already have loads of free time now and what do we do? We piss it away watching Jerry Springer and WWF eating cheezy poof's on the sofa turning into fat slobs.
For me, I'd rather spend a little more time outside and with real people instead of wiring myself more than I already am.
Technology has it's place...serving me not usurping me.
"TV, a medium as it is neither rare nor well done." Ernie Kovacs
...when computers are so smart, cheap, and ubiquitous that they do many ordinary chores to help ordinary people. When things talk to things, human beings are set free to do work they find more appealing.
This is the same old nonsense that's been touted ever since the age of the washing machine. Considering the thousands of labor-saving devices we've acquired throughout the 20th century, by this logic we ought to be living lives of perfect leisure now. But this isn't what happens. In industrial societies, "labor-saving" devices don't. Work expands to fill the time available. When things think, I'm sure you and I will be freed from the tedious chores of cooking, driving, cleaning, and living. We can become machines ourselves, consumed with work until we burn out or die.
(More at Talbot's Netfuture, if you're interested.)
He who refuses to do arithmetic is doomed to talk nonsense.
How will things that think be developed?"
By making them small and cheap.
The invisible addendum to this sentence is expendable. Small, cheap, and expendable - the mantra of the Japanese economy. Someday we'll be so deep in silicon poisoning that it will be a worldwide crisis, and we'll have to have a resolution like the Kyoto Protocol so that our president can ignore it. But like our automobile industry fifty years ago, we should march relentlessly ahead with abandon until we reach a crisis point, rather than attempt to head it off now.
If machines could truly think they would be screaming at us: "Don't Throw Us Out!!!".
Dr. Joseph Hairston
Superintendent, CCBC
To quote Joe vs. the Volcano: '99% of people go through life asleep; the remaining 1% walk around in a state of constant amazement.'
To add to that I'd say: 99% of people *think* they're awake; the remaining 1% know they've got some waking up to do.
There you have it, your Zen moment of the day.
To be quite honest, if I'm still waiting for a Photoshop render, or a level to load in RTCW, our machines aren't ready to think.
"Thinking" has been ascribed to mechanical devices for quite some time. Watt's flyball governor for steam engines yielded such comments in its day. Railroad switch and signal interlocking systems were said to "think" early in the 20th century. At that level, we can do "things that think".
But strong AI seems further away than ever. After years in the AI field, and having met most of the big names, I'm now convinced that we don't have a clue. Logic-based AI hit a wall decades ago; mapping the world into the right formalism is the hard part, not crunching on the formalism. Hill-climbing in spaces dominated by local minima (which includes neural nets, genetic algorithms, and simulated annealing) works for a while, but doesn't self-improve indefinitely. Reactive, bottom-up systems without world models (i.e. Brooks) can do insect-level stuff, but don't progress beyond that point.
I personally think that we now know enough to start developing something with a good "lizard brain", with balance, coordination, and a local world model. That may be useful, but it's still a long way from strong AI. And even that's very hard. But we're seeing the beginnings of it from game developers and from a very few good robot groups.
Related to this is that we don't really understand how evolution works, either. We seem to understand how variation and selection result in minor changes, but we don't understand the mechanism that produces major improvements. If we did, genetic algorithm systems would work a lot better. (Koza's been working on systems that evolve "subroutines" for a while now, trying to crack this, but hasn't made a breakthrough.)
It's very frustrating.