Meshing Developmental Evolution and Technology

Jerry23 writes "IT Conversations has free audio of a very provocative talk by futurist and developmental systems theorist, John Smart. He weaves a big-picture narrative featuring developmental evolution, technological acceleration, computational autonomy, the emergence and behavior of human social systems, why prediction has such a poor history, the unique growth properties of Information and Communication Technology and the limitations of biotech, finally culminating in his case for the inevitability of digital personality capture and a ubiquitous Linguistic User Interface. Among many other things, he asks 'What will Windows (and the Google Browser) of 2015 look like?'"

Comments

Singularity

[PxM]

This is a good time as any to mention Vinge's Singularity. The main topic is AI, but he also talks about IA or Intelligence Amplification. The DM in the article is a type of IA for communications systems between people. It would merge the useful parts of online communications such as active logging without the problematic impersonal problems that are sometimes caused. This gets extended further when people are connected 24/7 and they have the ability to treat the real world and the wired world much more similarly

--
Want a free iPod?
Or try a free Nintendo DS, GC, PS2, Xbox. (you only need 4 referrals)
Wired article as proof

Technological SIngularity

[bstadil]

Google Technological Singularity M or use the Wikipedia link. Speculating on anything after machine intelligences starts improving itself is futile. ETA 2060

Key breakthrough: the Intentional Stance

[G4from128k]

One key breakthrough will be to give computers the ability to take an intentional stance (short definition or longer essay) with regard to users. If Google could infer why I am searching instead of just what I am searching, it would be able to do a much better job. This would move from search-as-data-retrieval to search-as-intelligent-dialog.

I'm not sure if this can happen by 2015, but it seems like a key goal that is much more important than adding "Genuine People Personalities" to computers

Re:Key breakthrough: the Intentional Stance

[quokkapox]

I agree, integrating feedback from real humans is badly needed to improve search results. For instance, we know Google can't tell if you clicked a link in the search result page directly, but it can learn from how people behave as they hit the cache for different resulting links. If you hit the cache for result #1, and then quickly back out of that page and hit result #3, and then your session ends, or you revise your query, Google (and the other search engines) need to be able to learn from your behavior, and similar behavior exhibited by other humans.

They're probably working on that. And they, unlike microsoft, have the software to run the massive computations required to implement this type of machine learning. That would be my 20% project, anyway.

LUIs and the K-Prize

[Baldrson]

Primitive LUIs exist today in interfaces like Google, but will become dramatically more powerful over the next few decades.

I am quite excited by the confluence of advances in prize awards for technology advancement, and advances with the theory of compression. I'm convinced that if a substantial prize award can be created for dramatic advances in natural language text compression, it will lead directly to a solution to the most critical aspect of the "AI problem" -- that being the problem of the explosion of words without concomitant understanding. I had high hopes for the Internet being the new Gutenberg press leading to a new enlightenment but I'm concerned that without dramatic advances in AI to correlate the huge corpus being generated, the benefits of the new enlightenment may be too long in coming to save us from ourselves.

My work on a legislative proposal for fusion technology prizes was picked up by one of the founders of the Tokamak program. The more recent X-Prize award has a renewed the popularity of such prizes.

As a consequence I've been suggesting the creation of a new prize based on Kolmogorov complexity. As argued by Mahoney in "Text Compression as a Test for Artificial Intelligence":

"The Turing test for artificial intelligence is widely accepted, but is subjective, qualitative, non-repeatable, and difficult to implement. An alternative test without these drawbacks is to insert a machine's language model into a predictive encoder and compress a corpus of natural language text. A ratio of 1.3 bits per character or less indicates that the machine has AI."

A simple prize criterion would be for the first program producing a major natural language text corpus, with the size of the program being less than 1.3 bits per character of the produced corpus. Smaller intermediate prizes would help spur broader interest.

LUInterface technology entails Personality

[theDunedan]

I was expecting Smart to make a connection between LUI'S and Personality Capture, but if he did I missed it. It has to do with the notion that Natural Language Processing (which he pointed out is such a challenge) is naturally done by personalities. Okay, perhaps it is not a link with the "capture" part of Personality Capture, but we capture things now with computers. The link is with the "personality" part.

Language processing is based on life experience. In order for a neural net to learn language it must have inputs such that it can understand a concept such as "select", "walk", or "win." For a computer to understand "select" might be pretty easy. The trainer could say "I select a file" while he uses a mouse to select a file. The neural net could interface with that. But more sophisticated interfaces will be required to provide the nn with "context" for less computer-like concepts. We could put the nn into a robot that walks (like recently discussed on /.), along with visual processing so that it can experience walking see other things walking. Then when it hears references to "walk" it can make the connection.

(Yes, people who are unable to walk from infancy can speak intelligently about walking. Blind people can speak of and understand seeing. But those objections miss the point of a lack of "context input." As I understand it, a totally blind person does not know what "red" really is. (If I am misspeaking on this point, I apologize, especially to blind people or their close friends.))

Now consider this sentence, which is spelled phonetically:

"wonwonwonandwonwontoo"

Pretend that you heard it spoken instead of saw it written. The proficienct English speaker would realize several things. First, she would parse that into individual words:

"won won won and won won too"

Then she would do a lot of fast computation work to try different parts of speech for each word such that the sentence fits semantically and grammatically into rules of English. She might then write the sentence on paper as:

"One won one and one won two."

And that is enough for her to understand that the sentence is a fragment of a larger text, a newpaper report on the dog show or something. This is because the sentence has a lot of ellipses in it with anaphoric references being elided. Since references must be present, there must be more text associated with the sentence. With the references put back into the sentence it would read

"One person won one prize and one person won two prizes."

or "one dog won one bone . . ." or something.

The proficient English speaker would not even be thinking about "anaphoric elliptical references." She would "just know."

All of these levels of computation go on in our brains constantly when we participate in all forms of communication. And in order for a LUI to work properly, the machine will also have to be able to do the same thing. Yet without other faculties (such as visual processing, mobility, etc.) these things can not be learned either. Hence, Linguistic User Interfaces and Personality emulation are intrinsically linked (and pretty darn tough).

the Dunedan