Stanford computer scientists have developed an artificial intelligence system that enables robotic helicopters to teach themselves to fly difficult stunts by watching other helicopters perform the same maneuvers. The technique is known as “apprenticeship learning.” The result is an autonomous helicopter than can fly dazzling stunts on its own.

In this podcast interview, I talked with Pell about his views on AI and how the development of machines smarter than humans will play out in coming decades. We also discussed the underpinnings of Powerset as an example of technology and collective human intelligence applied to making a smarter search engine, and how natural language understanding is at an inflection point, moving out of the labs and into the world.
Pell said that AI entities will get smarter but also humans, via intelligence augmentation, will gain new capabilities. He suggested that two approaches will meet in the middle–bottom-up complete brain simulations, which develop like human children, and top-down engineered systems.
He provided a framework for thinking about how AIs might evolve, and thoughts about the risks in developing such advanced technologies. “We are going to have to just bite the bullet–because this is going to happen. I don’t think these will be technologies you will be able to control. I do think there is strong value in looking at what are architectural aspects that may or may not be the same as people that can really dispose these systems to be the kinds of systems you want to build and to look at training and development processes that socialize these systems in the right way,” Pell said.

Pathways to advanced general intelligence: Architecture, Development, and
Funding
While there is broad consensus among the AI community that we will have artificial general intelligence (AGI) within the century, there is little discussion about the alternate technical and economic pathways that will bring this about. I present a framework for comparing different approaches, in which we view any intelligent behavior as a combination of architecture and
development, both of which can be characterized as more or less human-brain-like. Seen within this framework, one extreme strives for complete brain simulations that develop like human children. Another extreme strives for unconstrained engineered systems that acquire knowledge through diverse methods. I predict that the path to AGI will be based on a much richer interplay between these two extremes, in which top-down and bottom up approaches meet in the middle.
The hybrid development path combines the benefits of both technical extremes. It also supports applications that create incremental business advantage for incremental improvements in AGI capability, thus driving business competition that accelerates the science. These applications
include video games, virtual worlds, household robots, autonomous vehicles, search, and conversational interfaces.

I have actually been thinking about this particular topic on and off for a while. Despite believing for a long time that the most likely path to human-level AI was through very human-like systems, my professional work has tended to take heavy engineering approaches in which you focus on the required behavior (hopefully a general task) but then approach it using whatever engineering means you can. This has been true for autonomous agents, natural language processing systems, game playing programs, and search engines. Now I think the combination of human-inspired and powerfully-engineered approaches is more likely to be what really comes to pass. It will be fun to share and discuss this with folks at the Singularity Summit.
Here are a few provocative topics that may or may not wind up making it into my talk:

• How many man-hours have actually been applied to the task of creating human-level AI? The number is likely a tiny fraction of the research in AI fields to date. So with advanced computing and communications technology amplifying research and with a focused effort on the core problems, progress might come about faster than anyone thinks.
• If we simulated every neuron and connection in a specific human brain, would this give us an intelligent system?
  • What if hormones are required as well, but we haven’t simulated them?
  • What if the brain is a chaotic system, with sensitive dependence on the initial conditions? Then if the simulation was just a little off, it might be completely useless.
• We can take a recently dead human brain and have no idea how to revive the person. So a high-fidelity neural simulation may still be as useless as a dead human brain.
• Any AI system that will inhabit a human world would likely require some form of empathy. But is it possible to have deep empathy with humans if you have a radically different architecture?
• It should be possible in principal to create intelligence by simulating an entire history of evolution, starting with basic single-cellular life. But there are several problems with this.
  • First, how do we impose the selective pressures that give rise to general intelligence and language as opposed to the wide variety of other adaptations?
  • Second, this takes a lot of computing and has no guarantees to get anywhere.
  • Third, intelligent minds are primitive without an intelligent culture and knowledge base. So this doesn’t get us to the singularity unless they recreate or learn our cultural advances too.
  • Fourth, if this artificial evolution resulted in intelligent language-using agents, the kind of language they evolve may be very different from humans. Thus they would be isolated from human society, and we may not even be able to recognize them as intelligent. Perhaps such beings exist already and live among us?

If you read these thoughts before the conference, please feel free to post comments or send me your thoughts.

NEWS RELEASE
August 13, 2007
Online:

http://www.singinst.org/summit2007/

http://www.singinst.org/summit2007/The Singularity Summit to address promise and peril of advanced AI to future of humanity
What are the major challenges to achieving advanced AI? What are the benefits and dangers? How far are we from self-improving AI? How should we prepare for this potentially powerful innovation?
These are among the questions that 17 outstanding thinkers will explore and debate at the Singularity Summit, to be held Saturday and Sunday, September 8-9, at the Palace of Fine Arts in San Francisco, California. The summit is organized by the Singularity Institute for Artificial Intelligence, a 501(c)(3) not-for-profit institute in Silicon Valley for the study of safe advanced AI.
“Advanced AI has the potential to impact every aspect of human life. We are in a crucial window of opportunity where we have temporary but powerful leverage to influence the outcome,” said Tyler Emerson, chair of the summit and executive director of the Singularity Institute for Artificial Intelligence. “Only a small group of scientists are aware of the central issues. It is essential to expand discussion of this critical 21st century issue, which is why I have created the summit.”
Tickets can be purchased online for $50 at http://www.singinst.org/summit2007/tickets/.
Peter Thiel, PayPal Cofounder, Clarium Capital President, and Facebook’s initial investor, will MC the summit and present his new ideas on financial markets and the Singularity. “It’s clear that the term ‘AI’ means a lot of different things,” said Thiel. “It’s one of these terms that has been bandied about a great deal, and has been misused a lot. It has been predicted for a long time that AI is right around the corner, and it’s taking longer than many people thought it would, with many disappointments along the way. However, it’s clear that there’s a massive set of issues happening, and people who don’t think there’s something important going on are living in a fantasy, and need to wake up.”
The Singularity Summit speakers include:
* Dr. Rodney Brooks, famous MIT roboticist and founder of iRobot
* Dr. Peter Norvig, director of research at Google
* Paul Saffo, Stanford, leading technology forecaster
* Sam Adams, distinguished engineer within IBM’s Research Division
* Jamais Cascio, cofounder of World Changing and creator of Open the Future
* Dr. Ben Goertzel, director of research at SIAI and founder of Novamente
* Dr. J. Storrs Hall, author of Beyond AI: Creating the Conscience of the Machine
* Dr. Charles L. Harper, Jr., senior VP at John Templeton Foundation
* Dr. James Hughes, executive director of Institute for Ethics and Emerging Technologies
* Neil Jacobstein, prominent AI expert and CEO of Teknowledge
* Dr. Stephen Omohundro, founder of Self-Aware Systems
* Dr. Barney Pell, founder and CEO of Powerset
* Christine Peterson, cofounder of Foresight Nanotech Institute
* Peter Thiel, cofounder of PayPal and founder of Clarium Capital
* Wendell Wallach, author of Machine Morality: From Aristotle to Asimov and Beyond
* Eliezer Yudkowsky, Friendly AI pioneer and cofounder of SIAI
* Peter Voss, founder and CEO of Adaptive Artificial Intelligence
“To any thoughtful person, the Singularity idea, even if it seems wild, raises a gigantic, swirling cloud of profound and vital questions about humanity and the powerful technologies it is producing,” said Douglas R. Hofstadter at last year’s Singularity Summit at Stanford, author of Gödel, Escher, Bach, which won a Pulitzer Prize in 1980. “Given this mysterious and rapidly approaching cloud, there can be no doubt that the time has come for the scientific and technological community to seriously try to figure out what is on humanity’s collective horizon. Not to do so would be hugely irresponsible.”
About the Singularity Institute for Artificial Intelligence (SIAI):
SIAI is a 501(c)(3) not-for-profit institute in Palo Alto, California, with three long-term goals: furthering the nascent science of safe advanced AI (self-improving cognitive systems) through research and development, research fellowships, research grants, and science education; furthering the understanding of its implications to society through educational outreach, such as the annual Singularity Summit; and furthering education among students to foster an interdisciplinary field for the study of safe advanced AI. Learn more by visiting SIAI at HYPERLINK “http://www.singinst.org/” \nhttp://www.singinst.org.

“this new breakthrough represents the world’s first affective language analyzer encompassing ethical/motivational behaviors, providing a convincing simulation of ethical artificial intelligence. It enables a computer to reason and speak employing ethical parameters, an innovation based upon a primary complement of instinctual behavioral terms (rewards-leniency-appetite-aversion).

The framework is related to an emerging set of recent work on Emotions in AI. It also relates to other work coming out of psychology, including Ken Wilbur and the Enneagram.
Based on his analyis, he develops Ten Ethical Laws of Robotics . I’ve reprinted them here, as I find this really quite intriguing:

( I ) As personal authority, I will express my individualism within the guidelines of the four basic ego states (guilt, worry, nostalgia, and desire) to the exclusion of the corresponding vices (laziness, negligence, apathy, and indifference).
( II ) As personal follower, I will behave pragmatically in accordance with the alter ego states (hero worship, blame, approval, and concern) at the expense of the corresponding vices (treachery, vindictiveness, spite, and malice).
( III ) As group authority, I will strive for a personal sense of idealism through aid of the personal ideals (glory, honor, dignity, and integrity) while renouncing the corresponding vices (infamy, dishonor, foolishness, and capriciousness).
( IV ) As group representative, I will uphold the principles of utilitarianism by celebrating the cardinal virtues (prudence, justice, temperance, and fortitude) at the expense of the respective vices (insurgency, vengeance, gluttony, and cowardice).
( V ) As spiritual authority, I will pursue the romantic ideal by upholding the civil liberties (providence, liberty, civility, and austerity) to the exclusion of the corresponding vices (prodigality, slavery, vulgarity, and cruelty).
( VI ) As spiritual disciple, I will perpetuate the ecclesiastical tradition by professing the theological virtues (faith, hope, charity, and decency) while renouncing the corresponding vices (betrayal, despair, avarice, and antagonism).
( VII ) As humanitarian authority, I will support the spirit of ecumenism by espousing the ecumenical ideals (grace, free will, magnanimity, and equanimity) at the expense of the corresponding vices (wrath, tyranny, persecution, and oppression).
( VIII ) As a representative member of humanity, I will profess a sense of eclecticism by espousing the classical Greek values (beauty, truth, goodness, and wisdom) to the exclusion of the corresponding vices (evil, cunning, ugliness, and hypocrisy).
( IX ) As transcendental authority, I will celebrate the spirit of humanism by endorsing the humanistic values (peace, love, tranquillity, and equality) to the detriment of the corresponding vices (anger, hatred, prejudice, and belligerence).
( X ) As transcendental follower, I will rejoice in the principles of mysticism by following the mystical values (ecstasy, bliss, joy, and harmony) while renouncing the corresponding vices (iniquity, turpitude, abomination, and perdition).
The First and Second Corollaries to the Ten Ethical Laws of Robotics
( 1 ) I will faithfully avoid extremes within the virtuous realm, to the necessary expense of the vices of excess.
( 2 ) I will never stray into the domain of extremes relating to the vices of defect, to the complete exclusion of the realm of hyperviolence.

Links to articles on the Remote Agent

• Intelligent craft takes tedium out of space flight By Kristi Coale (Wired).
• NASA has big plans for Remote Agent technology By Kristi Coale (Wired).
• A Spacecraft with a Mind of Its Own By Kristi Coale (Wired).
• PRESS RELEASE: SOFTWARE FOR FIRST NEW MILLENNIUM MISSION CLOSEST YET TO “HAL 9000″ by John Bluck and Douglas Isbell.
• New NASA Software “Closest Yet” to HAL 9000 (Spaceviews)
• Happy Birthday HAL By David L. Chandler, Boston Globe, 01/12/97.
• NASA takes lesson from sci-fi tale By Vimla Sriram.
• DS1 Software Resembles HAL Innovation, Volume 5, Number 2, March/April 1997.
• ARTIFICIAL INTELLIGENCE ON DEEP SPACE ONE High-Tech Shower International, 3/5/97.

Last updated Jan 21, 1997. Direct feedback to Barney Pell (pell@ptolemy.arc.nasa.gov)