Liveblogging the Singularity Summit 2007 - Day Two - afternoon

During the lunch R.U. Sirius interviewed Kevin Kelly. I am looking forward to listen to that podcast soon.

13.06 Michael Lindsay X-Prize Foundation shows a video on the X Prize Foundation, which funded the prize for private space travel won by Burt Rutan’s SpaceShip One team. The foundation is expanding in new areas and Michael presents their initiative in education. Focus on educational software.
“The most difficult part is picking the target goal. I want to ask you a question: how many of you think that a 2sigma improvement giving the pupils using these tools in Algebra 1 will perform better then 97.7% of the classroom, cannot be achieved within 10 years?” A lot of hands are raised, which is a little surprising.

Michael says that they are not sure if they want to set up the prize, which for me is absurd. If they have the means, they should definitely try: there can be little more important than equipping humans through education to better cope with the world around them.

13.35 Steve Jurvetson - The Dichotomy Between Designed and Evolutionary Paths to AI Futures
Asks the question to the audience which approach is going to produce the first AGI, and two thirds are for the evolutionary approach, to which he also subscribes to.
“Evolving an artificial brain will create one that is as inscrutable as the human brain.
There never will be a solution to the problem of the opacity of evolved systems, as in Wolfram’s cellular automata there isn’t a shortcut to executing them.
You cannot just upload the brain, extracting it from the entirety of the sensorial environment.”
Shows EvolvedMachines, genetic-programming.org
“Grand engineering challenge: unification of approaches (mentions Wolfram, D-Wave, Kurzweil, Kelly)”

14.15 Christine Peterson - Preparing for Bizarness - Open Source Physical Security
“Stick around, prepare for risks.
Last year addressed IT risk, this year addressing physical risk: chemical, nuclear, bio, nano.
It is a very scary world ahead. The best strategy is to think that AGI will come on both the side of defence, and offence.
The strategies that do not work are the top down near term ones of DoD, and DHS.
The challenge is balancing security, privacy, freedom, and it is not going well. Example is airline security.
Can we take the principles of Open Source as practiced in software out in the physical world?

14.30 James Hughes - Waiting for the Great Leap… Forward?
“I see greater than human intelligence around me every day, as a sociologist, since organizations are meta-human, and they make decisions, and they act every day.
AGI will be radically alien.
Friendly AI should be attempted, but also Humans should be Friendly, and maybe that is also going to be a consequence of our attempt.
Morals are editable.
We have an ability to determine outcomes, and have to recognize our Millennialist cognitive biases in both the good and the bad direction.
Emergent artificial life forms living in our information systems might not reach human level intelligence, or more, maybe stop at the level of cockroaches, but… cockroaches are very, very annoying!
The Storm worm has implemented automated defensive mechanisms that have been called frightening by experts.
From a policy point of view, in cybersecurity, international, national, and local regulations are all necessary.
We need to talk to the people designing future versions of the Vassenar agreements to agree what to future threats are.
Detection and control AI threats will require machine help.
AI will require IA (Intelligence Amplification). Some humans are actually intelligent, and already friendly!
Maintaining the mammalian brain as a control mechanism.
Need for a new social contract around social provision, labor, wages, education and retirement.

15.00 Eliezer Yudkowsky
“Somebody came to me yesterday asking if I was a creationist because I said that it was impossible for a butterfly to evolve. I want to be on record that I am not a creationist, I am an evolutionist. What I said that evolving a butterfly is a very inefficient way of getting a butterfly!”
“By the power of laziness, which is a programmer’s great advantage, you achieve results that would be impossible just by hard work.
Deep Blue’s unpredictable moves had a
A smart AI researcher should always ask “Could an AI solve the problem I am trying to solve? How am I thinking about the problem now?”
Humans can think about AI theory, so
An AI that can output AI-theory, can swallow itself, and become a reflective AI.
This is where Intellgence Explosion comes from.
AIs do not necessarily form a natural class, so their motives will be rather varied. We need to reach out in the space of possible minds, and pick the stable self modifying trajectory of a friendly AI.
There is a difference between terminal values, and transient values (means to an end).
In designing a moral agent we need to think about the trajectory that it will describe.
We need an ethical bulldozer, because we all together are not smart enough to shovel the stars.
We need a Friendly AI capable of helping us solve the Friendly AI-theory.
The last level of laziness.”

15.20 Q&A
With a very good remark about the need of getting our own house in order.

16.00 Ray Kurzweil in videoconference from the UK where he is at Aubrey De Grey’s SANS conference.
Makes a very short introduction, and responds mainly to Norvig, with various remarks defending the exponential gains in technology, economy, biology, etc. (Evidently Peter at the beginning of his speech which I missed, did attack some of his claims.)
I wasn’t writing as I stood in line at the microphone to ask a question about his current views of open and open source approaches versus regulatory or elistist approaches with a view towards the Singularity Institute making policy recommendations. He confirmed that he favors an open source approach, but that there will be a role for proprietary data sets. He did not address the policy/political angle of my question.
Ray is filming a documentary called “The Singularity is Near”, intertwined with the story of Ramona the female avatar he created, as she progresses in the future, starts to fight for her legal rights, tries to pass the Turing test… The film is due to arrive in 2008, about a year from now.

17.00 Tyler Emerson, the Executive Director of the Singularity Institute is giving the closing remarks of the conference. Educational outreach. Research, with the research grant program for students all over the world. Developing a slideshow to be given at Universities.

Liveblogging the Singularity Summit 2007 - Day Two - morning

I arrived a few minutes late, as I miscalculated my appetite at the breakfast place, which being also popular put our orders in the queue. The keynote by Peter Norvig has already started, and as I entered the Palace of Fine Arts Theater, he was saying “So now back to AGI”, which I assume means that he spent the first part of his talk on premises, or information concerning Google, etc.

09.20 Peter Novig, CTO, Google. ”
If you can solve a lot with data, go ahead and do that! Introduce the model only when data doesn’t help.
Probabilistic truths, against the current state of the web. This changes how you do testing, with a continuous updating.
AGI prerequisites - components are more important then solving the problem at once:
(and progress is being made so I am optimistic) probabilistic first order logic, hierarchical representation and problem solving, learning over the above, with lots of data, online, efficiently.
We were surprised how game theoretic this has become.
When we started we thought we would be making a copy of the web, and people would come and look things up in our index. Now we see that things we do influences the optimizers, then we change, and so on…
We are coevolving with the web!”

10.10 Storrs Hall
How do we build laws that work in conditions that we cannot even grasp? It is like Hammurabi preventing the Enron scandal!

If we put ironclad constraints on our robots today, since they will now the world better

By 2050 all corporations will be run by their Management Information Systems, and they will be required by law to be built so that their first law is ‘Make a Profit’!

Law 1: a robot shall Understand
(Socrates said “there is no greater good than knowledge, and no greater evil than ignorance”)
Robots will understand memetic evolution
(because evolution is where morals come from)
(The superintelligent AIs of the future will be built entirely of Human Ideas)
(Ideas should compete. Bodies should cooperate.)
ESC: Evolutionarily Stable Conscience. Robots understanding their morality and evolving
Law 2. A robots shall be Open Source
We already live in a world largely run by artificial information processing structures which have no conscience: Corporations and Governments. They have an open-source motivational system: auditing, because money is their emotion. The less transparent
Law 3. A robot shall be Economically Sentient.
Law 4. A robot shall be trustworthy, loyal, helpful, friendly courteous, kind, obedient, cheerful, thrifty, brave, clean, reverent, and shall do a good rurn Lord Baden Powell 1857 2007 founder of the boyscouts). Robots shall be boyscouts! :)”

10.36 Peter Thiel speaking about investing in a world where the possibility of the Singularity exists.
“How do you invest in the world as a whole?
The basic intuition about a world with a Singularity scenario, with extremes in good and bad, is that the tails of the Bell curve distribution of possibilities are much fatter.
If you are somebody predicting the end of the world, then even if you are right you will not make a lot of money!
Consequently you have no choice but to bet on the positive outcomes.
The things you choose are not going to be exactly right.
You would expect the world to be full of manic booms and busts.
All the conventional theories say that the markets should be getting smoother, but that is not what we are seeing.
One of the ways to see them, is that these fluctuations represent different bets on the Singularity, or proxies for it like globalization.
How does someone like Warren Buffet invest in view of the Singularity? His holdings in the last ten years shifted towards insurance, and reinsurance. That is very significant.”

10.50 Charles Harper, Templeton Foundation
“What does a slug know of Mozart?
Epistemology follows Ontology. There can be state changes. The best example is the Chimp-Human differentiation.
Our culture is social based on a linguistic ontology.
How serious is the “dilemma of power”? We are much slower in evolving social and cultural solutions to the proper handling of the technological advances that we create.
I don’t have the answer, but I think that it matters a lot, and solving it is very hard.
How important is “transformation of desire”?
“The Hungry Soul” by Leon Kass
“Evolutionary Dynamics” by Martin Nowak”

11.35 Q&A session with the morning’s speakers.
J. Storrs Hall “It has been the subtext that it is possible to build machines that are smarter than humans. I would like to suggest that is then also possible to build machines that are more moral than humans, and that it is desirable to do so.”
Peter Thiel “We must not presume that regulating AI research in the US would not stop it everywhere. I almost hope that we could push the accelerator, and arrive to AGI faster, because my intuition is that we would maximize then the probability of its use for good.”

12.00 Lunch. Spoken with R.U. Sirius, Kevin Kelly, Ronald Bailey, and others. Shot some photos that could be rather nice, I hope (and the World Future Society asked for permission to use them, which they didn’t need, since all my photos are licensed with the Creative Commons Attribution license, which allows for their free use as long as attribution is given): for example Peter Norvig of Google, and Barney Pell of Powerset sitting together on the floor hacing lunch

Follow the afternoon of the second day of the Singularity Summit on my following post.

Liveblogging the Singularity Summit 2007 - Day One - afternoon

15.00 Jamais Cascio - The Metaverse (is being skipped?! or maybe speaking later?)
Actually looking through the Google Blogsearch links I myself provided in the previous post, I saw that Jamais has already posted the full text of his speech. I guess I will list just a few sentences here and there that strike me most. (if and when he speaks)

Marcos Guillen of Artificial Development is launching today its commercial and open source products, CCortex Spiking Neural Network Engine. AD wants to implement a complete simulation of a cortical system.

Now Jamais is going to speak. He is not using a PowerPoint presentation, which he says corrupt the mind. He is reading, quite well, the text of his speech. He is referring to the Metaverse Roadmap Overview that he authored a few months ago.
“Scenarios are not predictions, but provocations.”
“Code is inherently political.”
“It is important to democratize the Singularity”
“There will be surprising benefits from a messy open source collaboration for AGI”
“We can make the right choices. Survival is possible.”

15.50 Panel discussion
Stephen Omohundro
“Self improving AI is going to be extremely unpredictable. The previous version could itself not understand the one coming after it.
Describes Von Neumann’s theories of rational behavior.
“Programmers are devices for converting pizza into code” “Self-improving systems are devices for converting resources into maximizing utility”
“The Four Drives: Efficiency - use recources better, Self-Preservation - keep resources, Acquisition - get more resources, Creativity - find new ways to create utility”

Peter Voss, Adaptive AI talks about “Increased Intelligence, Improved Human Life”
“AGI is almost certainly going to happen in less then ten years, and very likely in less then 5
Human level intelligence is the bottleneck for many problems that humanity faces
Through the benefits of AGI we will become more wealthy, more healthy and more moral”

Question from Brad Templeton: “Will personal AGIs will be objected to on the basis of them being slaves?” Answer is that the AGIs might not themselves object, but other humans will.

17.10 Neil Jacobstein of Teknowledge
“Many of the formerly narrow expert systems have acquired broad scope.
Myths about AI: if it works it isn’t AI; AI is all hype (but writing off AI is like writing off eCommerce after the meltdown of the end of the 90s); if it isn’t human+ level then it is not AI”
Shows statistics about IAAI - Innovative Applications for Artificial Intelligence.
Comment from a former DARPA Director: The DART scheduling application paid back all of DARPA’s 30 year investment in AI
An other application, CombineNet ASAP, has a documented cost savings of $1.8B.

17.34 Ben Goertzel of Novamente and the Singularity Institute has a new title for his speech which is “Nine years to AGI”, since a year ago he gave a speech which was entitled “Ten years to AGI”.
“Human babies are the stupidest things, but we have to start there.
The robotic approach is good, but you will loose all of your time with actuators, and servo motors. We settled on robots in virtual worlds.
Novamente is initiating an open source AGI called OpenCOG, and will present virtual pets at the Virtual Worlds Conference in San Jose in October.

18.00 Paul Saffo “Machines of Loving Grace: Anticipating Advanced AI”
“The press coverage of this conference is proof that the public is about to join the conversation.
But this is a moment of pessimism. The optimistic scenario is that when AGI arrives they will treat us as pets, and the pessimistic scenario is that they will treat us as food.”
He reads a poem “All Watched over by Machines of Loving Grace”
We need more poets, and more novelists covering this.

Questions to Ben, Neil, and Paul. Missed many of them, because I stood up making a question/observation which went more or less the following: “Paul, your observations, and the reading of that wonderful poem make it seem that we need to build a bridge between technologists, and humanists, which organizations like Edge.org, and Third Culture are very well doing. I would posit that it is even more important to build a bridge of communication towards, between those of us who are proude of our knowledge, and those who starting from a faustian fear, and often in a position of power, are proud of their ignorance.”

Liveblogging the Singularity Summit 2007 - Day One - morning

We got the power supply issues solved, and the wifi is also now working well thanks to the great help from the organizers.

09.40 Peter Thiel, Clarium Capital introducing the Summit:
“The 21st Century is going to be far more greater, and/or far more terrible then the 20th Century. We are here to make sure that it is going to be greater, and not more terrible.”

09.50 Rodney Brooks, MIT speaking
“In 1783 people were not asking the right questions about the airplane industry and flight when the hot-air balloon was invented
When an AGI will be invented the world will be a very different place than today
I am a great fan on AC Clarke, and 2001 Space Odissey is one of the films that is aging better.”

10.00 “AGI and Robots are needed for demographical reasons. And there will be a lot of VC, a lot of investment to provide them.
In 2016 we will have all movies worth watching in our pockets on our iPod… so much for RIAA and DRM!
There are 5000 deployed robots in Iraq now, up from zero in all of the military in 2001. These are not prototypes”

10.10 “From 2 meters in six hours in 1979 our cars at the Stanford AI Lab now do 200 kms in 6 hours in 2005. That is the power of exponentials.”

10.15 He is depicting various scenarios for the future, each of them however including the coming of the AGI, in which he does believe. Unless “as from Tau Ceti somebody is laughing at us as if we were chipmunks not just smart enough to get there”. Laughing.

10.20 Questions from the audience. One of the non-AI specific ones:
“You said that it is the time now to ask if it is a good idea to let robots aquire autonomous firing capacity, and that there are some governments following the Geneva Convention thinking about it. Do you think it is a good idea to give AI to the United States Government military?”
Answer from Rodney:
“This is a question that scientist face since Leonardo who was completely funded by the military. Scientist must think and implement control mechanisms, but this is a question outside of AI.

10.30 Eliezer Yudkowsky, Singularity Institute speaking
“Criticism of the bold thesis of accelerating change does not necessarily touch the core of the thesis.
If you want to be smart don’t be fooled by futures with blinking lights, but concentrate on cognitive enhancements!
He is quoting Geordie Rose of D-wave on Software Progress vs. Hardware progress (1977 hardware with 2007 algorithm is better then the other way around) to show that the threshold for the Singularity event horizon can be lowered by breakthroughs!
The three schools of Singularity (Event horizon, Accelerating Change, Intelligence Explosion) do not imply each other, or require each other’s thesis, but they support each other’s cores.”
Question time, and answers…
“The most important challenge in AI today is to keep improving our understanding of how to think how to think about thinking, which we have been doing on our own in science for hundreds of years.
One billion operations per second could be enough to be the basis of an Intelligence Explosion grenade, but we would need an additional 100 years’ worth of science to be able and code it there.”

11.00 Break

11.35 Barney Pell, CEO of Powerset “Most of AI reserchers don’t do AGI because it is very hard, and also because it is difficult to get funding.
At some point better AI and AGI becomes central to the creation of better products, and that becomes a virtuous circle where more revenues lead to more funding, better AGI, better product, and so on.
Within the next five years the introduction of natural language interfaces is going to create a huge difference.” Obviously he says that, since that is what Powerset is about! A few slides present Powerset’s approach.
Question from the audience if this is going to help people learn, and the answer is a definitive yes. Conversational interfaces are going to be powerful learning tools.

12.00 Sam Adams from IBM Research “We can beat Gary Kasparov, but no computer has the common sense of a six year old child.
Joshua Blue prject to develop a computer that pass the ‘Toddler Turing Test’, stopping at age three. They can do all kind of things that traditional AI says it is incredible hard.
Implementing analogs of this approach in a computer system we learned a lot of very counterintuitive stuff. For example superstition appears to be the source of all knowledge, since all experience for the first time is without understanding. Yes, but grounded in experience.
An other is ‘aggressive forgetfulness’ where most of the sensorial input received is quickly discarded.
Superstition + Forgetfulness = General Intelligence?
For the next two years I am going to study multicore systems because we don’t know how to program them now effectively and we need them to run our algorithms closer to real time.
The pathway to AGI is in following the Child!”

12.30 Wendell Wallach, Yale “In our current computational theories of the brain we take into consideration neurons and synapses. What role do other elements of the brain play, like glial cells, microtubules, variations of the neurons, etc.?
As Michelson was wrong in saying in 1894 about physics that “it seems probable that most of the grand underlying principles have been firmly established” we could be missing pieces that make our science of the brain inadequate.
If we want to approach machine consciusness, we need to better define the phenomenal experience.
We are just a few years away from a major catastrophy caused by autonomous computers systems making a wrong decision. That will precipitate the discussion on machine morality.
Artificial Moral Agents are necessary.
How can we make ethics computable?”
Great quote from the audience during questions: “Bureaucracy is our way of getting untrustworthy agents to be trusted. Are we going to corrupt our Moral Agents if we apply this trick to them?”

13.10 Panel discussion and question session
“Should we give these machines rights?”
Barney Pell answers “Robots will ahve rights when robots will claim rights. Today we are hypocritical about these issues, when we eat animals, etc.”

13.30 Lunch break

Here are more people liveblogging the Singularity Summit and photos of Flickr

Dan Farber’s blog Between The Lines, covers the Singularity Summit, is being updated frequently with great posts, and it doesn’t show up in the previous query since it doesn’t use the word ‘liveblogging’.

15.00 Continuing for the afternoon the live blogging of the Singularity Summit on a separate entry.

Liveblogging the Singularity Summit 2007?

The Singularity Summit 2007 is about to start at the Palace of Fine Arts Theather in San Francisco. Already at 7.45 about a hundred people were waiting in front of the gates ready to enter.

(Update: we were just told to unplug our laptops, so it is unlikely that it will be possible to liveblog. Tyler is now trying to find a solution to give us more power. Dan Farber from Zdnet is sitting on my right trying to cover the event as well. So we’ll see…)