Wednesday, September 12, 2007

Will the next Singularity Summit be overrun by a rogue AI?

It seems entirely reasonable to expect that an integration of cumulative knowledge can produce an uber-humanoid when the processing power advances enough. This cyber-consultant may not be able feel emotion the same way as a normal human, but that does not seem to be necessary for amazing intellectual productivity gains to be realized. How often do real flesh and blood guru's get hired to provide advice and expertise without appearing to have ever left the library for anything but a star trek convention.
What does a great CEO or medical doctor know that can't be codified? Perception is the current hurdle being addressed and if that is mastered such tasks as financial analysis, medical prognosis, product engineering etc may become nothing but the embodiment of a warm breathing human mediator and the cold calculations of the uber-cyber-geek. That may seem scary until you realize how many medical errors are made a year and how much misallocation of resources stifles economic growth.

"...Who is to say there are not principles of biology and physics that we don't fully understand yet that will act like a brick wall on the development of information tech?" John

This is a good point to consider, but will intelligence be limited in the same way as physical laws? Some say processing power limitations can only be circumvented by nanotech and/or quantum computing. An intelligent agent may be unable to explain what a prop plane will feel like above the sound barrier (a physical limit) but knowledge often makes decisions easier. Having a large body of experience with flying objects behavior is not as useful as a law of aerodynamics. But a super AI may be limited in its ability to predict human behavior, like the explosive demand for tickle-me-elmos, because the AI's capacity to predict is limited by the subject matter, the chaotic nature of normal humans.

Here are some other questions to ponder...What ethical considerations should we apply to people who use the accumulated corpus of human knowledge to create a comprehensive uber-AI? What patent rights are reasonable for an advanced algorithm that can beat the stock market or look further down the iterative line of invention?

If AI is able to think exponentially and create what will that do to our notion of intellectual property rights?

What if the AI determines eugenics etc is the best social policy?

If the individual agent is able to make the best decisions within a distributed manner then it is highly likely that the uber-intellect will acknowledge that. The ability to buy cigarettes and porn does not have to be restricted by an uber-intellect unless it was given the authority to make policy decisions and decided so… and that may well be a scary form of government. Would you agree that an invisible hand is already limited by product availability, and social norms. AND these previous actions of the invisible hand have created a cumulative current state of affairs with positive and negative limitations? An uber-intellect may be able to create prosperity and subsequent increases in individual choice.

also I am not sure Adam Smith would necessarily understand our use of "invisible hand"

Processing power does influence what it considered intelligent behavior because within a certain time and space any task that will be judged intelligent (turing test) is going to be limited by mental resources available to evaluate associations and determine actions within a dynamic environment. The real world is fast paced and real world decisions exist within a quasi-infinite problem space. Chess itself requires incredible processing power in order to span even a segment of the state space and is not dynamic like most tasks.

What does a dog with super computational ability do with itself, who knows, a cognitive scientist might argue it depends on what part of the brain this super dog got upgraded, IF the dog is capable of reflecting on its thinking it may be able to do more than just calculate the distance squared to the next dogs ass. What is different between the human brain and the dog brain…size, structures, innate algorithms?

Processing power is necessary for human-like intelligence, but not sufficient.

Was Alex intelligent?

Alex is intelligent according to a measurement of intelligence that is satisfied by an ability to “combine the different labels in his vocabulary to request, confuse, or categorize over 100 different things," Pepperberg said. "So he understood labels from materials, shapes, colors, things like that."

definition of intelligence on wikipedia:
The definition of intelligence has long been a matter of controversy.
At least two major "consensus" definitions of intelligence have been proposed. First, from Intelligence: Knowns and Unknowns, a report of a task force convened by the American Psychological Association in 1995:

Individuals differ from one another in their ability to understand complex ideas, to adapt effectively to the environment, to learn from experience, to engage in various forms of reasoning, to overcome obstacles by taking thought. Although these individual differences can be substantial, they are never entirely consistent: a given person’s intellectual performance will vary on different occasions, in different domains, as judged by different criteria. Concepts of "intelligence" are attempts to clarify and organize this complex set of phenomena. Although considerable clarity has been achieved in some areas, no such conceptualization has yet answered all the important questions and none commands universal assent. Indeed, when two dozen prominent theorists were recently asked to define intelligence, they gave two dozen somewhat different definitions.[1]

A second definition of intelligence comes from "Mainstream Science on Intelligence", which was signed by 52 intelligence researchers in 1994:

a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings—"catching on", "making sense" of things, or "figuring out" what to do.[2]

http://en.wikipedia.org/wiki/Intelligence

not wanting to sound like Ramona, but a thermostat is intelligent if the definition of intelligence is an objects ability to react to stimuli in a manner satisfying some task goal, ie measure current temperature and adjust processes in order to adjust temperature to that of a goal state (ex. beautiful 65 degree WI evening).

but not intelligent like a retarded 5 year old.

Is my ability to respond to your questions more than what could be achieved by an advanced conversationalist expert system? How would you test me to see if I am a human instead of a DARPA cyber-turing-bot seeking out AI skeptics?

If hungry, eat
If tired, sleep
If hot, remove hand
If pleasurable, smile
If pleasurable
And contested (turbulence detected)
Display poker face
send signal to nearest person for optimum attitude, ascend or descend to determined attitude or until within optimal comfort zone.

“by an advanced conversationalist expert system” I mean a CES with ADD and questionable social skills..

“computers are very poor at doing things we've done since the beginning of human existence: language, image recognition-- deriving meaning from imagery, understanding, emotional cognition”

computers are great at language, they are programmed with various levels of languages that must communicate between an OS, various software apps, a user input, external computers in the network, all with a thermostat like awareness of processing capacity. They are not good at human language and the hodge-podge of associations/connotations with which we relate our language to individual experiences/perceptions in the real world.
…besides how often have you tried to explain abstract terms like love with little success.

…technology is making big advances in image recognition, not necessarily via the same method as human sight,…besides I doubt you would suggest visual imagery is necessary for intelligence, unless you wanted to be bombarded by angry audiomails from visually-impaired bloggers. Emotional cognition…I don’t understand myself let alone other people, maybe you know a genius shrink you can put me in touch with.

Objective Art? I have nodded several times when a pretty art student explained the meaning of some “contextual impressionist art” only to be reminded of unique perspective and an ego's willingness to defer to the id.

satisfice...satisfice…satisfice

"Kurzweil talks about a computer program able to 'simulate' a human mind in 10 years; in what way? Will it go out on dates? Will it have emotional meltdowns? All the best work in cognitive science recently (Le Doux, D'Amasio, Ramachandran) has been on the importance of emotions to mentality and the importance of bodily states to emotions." Stephan Johnson

I agree that 10 years is unlikely to produce the technology required to comprehensively simulate the human condition with regard to emotion and feel. But is one human mind ever able to simulate another to any certain degree? And what about cognitive scientists who stimulate a portion of the brain with electrodes and the subject reports various moods and memory recalls…what are dreams, flashbacks and the ghost limb phenomenon of amputees. Is all past experience relevant to every new conscious experience? Besides it would be a troublesome hurdle if AI was expected to simulate the 13 year old female in order to gain respectability.

Mark Bahner
Are you arguing for a Chinese Room perspective or against it? My novice understanding of current translation software would cause me to consider it little more than a set of word exchanges, with some syntax rules. Ultimately it is unable to convey meaning between complicated sentences and subtle uses of semantic meaning, with any high degree of accuracy. What about discussion context, semantic drift, and measurements of the “fusion of horizons.”

probably all quibbles over differences between strong and weak AI again.

testin, I don't know about Paul but I am not all together clear on what your asking in your questions.

No comments: