Internet Time Blog: Accelerating Change 2005 (live)

Accelerating Change 2005
Stanford

This year's theme is AI-IA. Artificial intelligence ("AI"), broadly defined, improves the intelligence and autonomy of our technology. Intelligence amplification ("IA") empowers human beings and their social, political, and economic environments.

KnowledgeBase: slides, pdfs, Q&A

John Smart. How many people have read The Age of Spiritual Machines? Half the hands go up. This is the only conference dedicated to a multidisciplinary look at accelerating technological change. We aim to move the dialogue beyond… Awareness, education, advocacy. Future Salons popping up all over.

370 people are here. World today is a network more than a hierarchy, and amazingly enough, this has come about in the last decade.

Today you’ll hear about the singularity, age of spiritual machines, Ray Kurzweil’s new book, think of ourselves as ambassadors to the future, Google OS?, AI, a new model of neural network, more connections in our brains than there are atoms in the universe, Terry Winograd on design, birds of a feather (this is a working conference).

Verner Vinge
vinge@cs.sdsu.edu
www.rohan.sdsu.edu/faculty/vinge
www.rohan.sdsa.edu/facluty/vinge/misc/ac2oo5

Exponential. You hear people say “That’s exponential.” Exponential growth doesn’t last very long. Perhaps a catastrophic crash. Or the levelling off of saturation. Moore’s Law predicts computers with more oomph than the brain. And what happens one cycle of Moore’s Law after that? A technological singularity. Life is only the prologue to human intelligence.

What if the singularity doesn’t happen? Could be a Russian roulette situation.

Ray Kurzweil
When Humans Transcend Biology

www.KurzweilAI.net
www.KurzweilAI.net/pps/ACC2005

What will protect us from strong, predatory AI?

Soft technology, we can plan for, like we prepared for Kristina… (laughter). People intuitively think of the future as a straight-line projection of today. The Paradigm Shift rate is doubling every decade. Lots of logarithmic growth. Life at -10¹⁰. Six epochs of evolution: chem., DNA, brains, tech, tech+human intel, universe wakes up. Computer evolution similarly exponential. You must see the slides Ray's a firehose.

Informal learning note: Ray Kurzweil is easily one of the brightest people on the planet. Why then did he show us 88 powerpoint slides, almost all of them complicated graphs of exponential curves? In terms of audience comprehension, 3 or 4 slides would have been plenty. Even though I was sitting in the center of the first row, my brain seized up, and I slept through most of Ray's presentation.

Panel

Neil Jacobstein, Teknowledge
njacobstein@teknowledge.com

The Evolution of AI Applications

“If it works, it isn’t considered AI.” (laughter) Pattern recognition, control, date interpretation, design, search, training, etc. Many ways of looking at AI.

Federated services on the web a la Kevin Kelly’s Wired article in August 05. Nanotech:matter::computing:information. Advanced molecular mfg will eventually provide personal computing devices with a billion Pentium class processors.

“What is the use of the ever-faster, ever-slicker, more nearly perfect implementation of rotten plans?” asks Stafford Beer.

Falling Water. Designed late, non-standard components, no continuous improvement. Residents lived with many bugs but were asked to take ownership of them by its authoritarian architect. Over 40 years, Toyota workers have made 20 million suggestions, going from making junk vehicles to top in the world.

Rules: Balance long and short term pragmatics. Combine new and mainstream AI tech for specific application solutions vs picking the “one best way.”

Patrick Lincoln

SRI

Main point of AI is for intelligence augmentation. Tools for this growing in power exponentially. Capability gap between computing capability and human capability. Foolish to bet your life on an iffy system; insane to bet the life of society at large. Forty years ago we deal with a dozen variables; now we can handle 10 million.

But systems are too complex for a personal to understand. Hence, we must cross the abstraction barrier. Property-sensitive abstractions. Scaling up and down, slicing…

Effort required for high assurance analysis. 80s: huge teams required. Now: takes an hour. We must support designer and user interaction at the right level of abstraction at the right time. Extremely powerful symbolic reasoning tools make this almost possible. We are accelerating > Moore’s Law. Most pressing problem in the world: “Collectively getting better at solving urgent and important problems.” Engelbart, March 1951.

Peter Norvig

Google

AI in the middle. Machine learning. Knowledge engineering. AI in the middle.

Machine learning: don’t know how to do it.

Knowledge engineering: missing key stuff in 1991. costs $10K/page.

Search: n

Mediocrity augmentation

Bruno Olshausen
Neuroscience and Future Prospects for Intelligent Systems

baolshausen@berkeley.edu
redwood.ucdavis.edu/bruno

Helen Wills Neuroscience Institute, UC Berkeley

Theoretical neuroscience seeks to understand intelligence by studying these:

Psychology: behavior, perception/cognition, performance characteristics
Neurobiology: neurons, neural response properties, signaling mechanisms, synaptic transmission
Math/computer science: information, signal transformations, efficient representation

Reality is a hallucination, most of which comes from the brain and not the retina.

Dimensions

Spatially invariant or specific
Slow or fast changing
Objects vs features/details

Esther

Peter Barrett
Chief of Engineering
Microsoft Television
IPTV

“IPTV is better TV.” Broadband network for t.v., data, and telecomm. No room on the coax for a hundred channels of HDTV. Ditto satellite. Overall broadband capacity is growing fast and will dwarf traditional channels within five years; video bandwidth is declining.

Origins of tv. IP since ’70. Now many varieties. Live 8: phone-cams, commentary.

The Long Tail: Gilligan’s Island episode 73, the “Citizen Kane” of television. Baby pictures as killer app.

Mark Finnern & Zack Lynch

Terry Winograd
Teaching Innovation
Stanford D-School

University looking for mini-singularities. A new insight. Engagement in the world.

What is design?

Professionals, e.g. interior designers.
Set of skills learned in a common way, e.g. studios
Methodology for working “Conversation with the materials” (Schoen), “Iterative plan-making” “Iterative user-based prototyping” “Integrative thinking”

Design thinking: innovative, user-centered, iterative, integrative, reflective

“Enlightened trial and error is more effective than perfect intellect”