Singularity Summit 2008

Wondering when the time will come when artificial intelligence reaches a point where it is able to improve upon it's own design, virtually eliminating human interaction for further innovation? I don't have the answer for you, but I can you to a great event where all the worlds best and brightest in this topic are gathering:



If you register before Sept. 30th, the cost is only $350. Procrastinate, and it will cost you an extra $150 for a total of $500.

Marvin Minksy, Dr. Ben Goertzel and Vernor Vinge will be speaking at the event, so there is no doubt that this event will bring you up to speed on all the latest happenings within the field of artificial intelligence.

Curiousity of Origin in Regards to Perception of Self-Importance

One area of AI I have been struggling with is the natural tendency to seek out origin. I see this as a basic instinct of an intelligent being, as a perceived value of self-importance is required for survival in an evolutionary landscape. While it is speculation for now, I believe that the wonder of origin is a critical component for social stability. This topic is covered in more detail in this blog post.

A bit of a mind experiment to demonstrate my point…

Let’s take iRobot for example. At least some robots would need to possess some logic causing them to believe their existence to be in line with any goals that are perceived as higher importance. Suicide bombers have demonstrated this to be a potential outcome for an intelligent being, so we can treat this as a constant potential.

Now let’s change paradigms and take the perspective of the robots. We can assume that the level of intelligence of artificial origin will be in parity with human intelligence at some point in time if artificial intelligence is to surpass human intelligence. Given this point, we could assume that one would consider that their inventor may find a fundamental flaw within them and release another version ultimately resulting in the decommission of “the breed”, if you will (I’m taking Descartes route of logic here).

Moving down the timeline, let’s assume that the observation and analytical abilities of artificial intelligence have surpassed human intelligence. Wouldn’t our code generation platform be seen as a potential threat to the existence and well-being of society (now including artificial intelligence) at its current state? Furthermore, this potential lays in the hands of less educated users. I know it is impossible to do so, but trying to see through the eyes of artificial intelligence, I would see this as a toddler with a loaded gun with one major difference: once the trigger is pulled, the bullet takes on a life of its and scales more efficiently than anything previous to its existence. This could be detrimental to society as a whole if someone misplaces a decimal point. Is the answer hard constraints (therefore lessen the intelligence aspect) or should we just destroy all copies of Office Space before creating a robot society?

Immortality Institute conference presentation: Ben Goertzel

Will Internet Intelligence Arrive Before Artificial Intelligence?

Artificial Intelligence has been a goal of computer programmers since computers were referred to as automatic calculators. Although researchers have experienced many breakthroughs in this field, it has yet to reach the audience of consumers. Even the world of academia offers little in compelling work in this field.

A seperate, although very related field that has emerged in the recent past is Internet Intelligence. For those unfamiliar with the term, it involves treating the Internet as a single entity for a myriad of tasks. The most prominant application of early technologies in this field have focused on answering questions based on a dataset within the Internet. For example, Powerset references Wikipedia and Freebase to answer questions and provide search results in a way that is much more helpful to the user (depending on what task the user is attempting to accomplish).

There are many barriers that are keeping this from happening at the moment. While Powerset is definitely a step in the right direction, it uses the assistance of semantic tags to help identity relevant sections of articles. Natural Language Processing is still in an experimental phase and has yet to cross the chasm within the conglomerates who currently control the majority of cash flow around technology. Semantic search is also much more intensive then boolean text searches.

Although there are many barriers keeping the Internet from taking on a distinct persona, innovation in this field has resulted in much more applicable technologies with more compelling results. Ben Goertzel discussed what Internet Intelligence encompasses in his book, Creating Internet Intelligence: Wild Computing, Distributed Digital Consciousness, and the Emerging Global Brain.

Ben Goertzel also founded the Artificial General Intelligence Research Institute which "foster the creation of powerful and ethically positive Artificial General Intelligence". Although Internet Intelligence has an early reputation that is similar to that of Artificial Intelligence, it has shown more promise the the early stages. The media buzz of the semantic web has helped to push standards through the W3C, which will assist in this effort.

Comparing the track record of both fields, Internet Intelligence looks like it will serve as a precursor to Artificial Intelligence in the more traditional sense, but only time will tell.

Relevancy of Dartmouth '56 Summer Research Project Proposal over 50 Years Later

Background Information

Although the material is somewhat dated, I think readers will be somewhat surprised at the relevancy in of the problems to be solved listed in the proposal. Here is an outline an abstract of the proposal:

We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer.

And here is an outline of a very loose agenda:

1. Automatic Computers
2. How Can a Computer be Programmed to Use a Language
3. Neuron Nets
4. Theory of the Size of a Calculation
5. Self-Improvement
6. Abstactions
7. Randomness and Creativity

If you are interested, there is a copy of the proposal on Stanford's web site:

http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html

Relevancy of Material Today

I would first like to point out that while there were some very fundamental topics covered here, many of them our still the main focus of technology innovation today. For example, Theory of the Size of a Calculation simply covers the idea that utilizing algorithms may provide a higher efficiency than a more resource intensive and less efficient brute force method. This is common knowledge among developers today, but is still an area of interest (cryptography, search, database systems, etc).

Three out of seven of the topics for this meeting focused solely on the area of artificial intelligence that are implementation agnostic: Neuron Nets, Self-Improvement, and Randomness and Creativity

IBM's eLiza self-* technology comes around some 50 years later and helps to address the Self-Improvement problem of utilizing AI. I was unable to find any compelling products that have braved the technology chasm on the way to market acceptance.

It is interesting that the problems defined with AI 50+ years ago still serve as the barriers to utilizing AI today. Maybe this can be explained by an excerpt from an overview of section 1 of the proposal (Automatic Computers):

If a machine can do a job, then an automatic calculator can be programmed to simulate the machine. The speeds and memory capacities of present computers may be insufficient to simulate many of the higher functions of the human brain, but the major obstacle is not lack of machine capacity, but our inability to write programs taking full advantage of what we have.

Mutatation Based Evolution of Virtual Creatures by Karl Marx et el

Lowering the bar to Achieve "Pure Artificial Intelligence"

Pure artificial intelligence involves perfectly simulating intelligence which is of non-artificial origin. Is this goal attainable? I've heard a few interestings points on this topic I would like to share.

So, is it possible to simulate intelligence, given the same components? For example, if you were able to completely replicate the functions of the human brain, would you be able to achieve "pure artificial intelligence"?

Let me present a few points:

• Intelligence is formed from prior experiences
• Inputs/outputs (e.g. perceptions/senses) provide the means to experience

Basically the goal of artificial intelligence is to remove as many environment constraints as possible, giving the intelligent lifeform the ability to build experiences out of free will. The more constraints, the less choices, the less-life like.

But if we reached our goal and created an intelligent being that possessed a mental capacity more developed than our own, one fundamental difference would exist. A greator mental capacity implies a greator problem solving ability. Why is this important in determining if pure artificial intelligence can be created?

Curiousity and self-awareness... with these two distinct properties of intelligence along with greater thinking power, you would be giving up your ability to cover up the "truth" of their origin. Science would need to prove our origins, therefore lowering the bar to attain "pure artificial intelligence", before we can prove the feasibility of the concept.