Thursday, July 10, 2014

A leap in artificial intelligence? - Reflections

Reflections on achieving Artificial Intelligence


Before providing feedback on the progress made in building Gaia, I want to share today some other reflections not yet expressed in previous posts.



Image credit: http://www.cs4fn.org/ai/


For me one of the most important questions we should ask ourselves in relation to AI is what such an AI should do.

Answering such a question will lead to a lot of things that are actually possible (and some not achieved yet). To name only a few of the existing ones:

  • Acquiring information about our world - necessary for all the rest (Google's Knowledge Graph, Cyc, Semantic Network Technologies, ...)
  • Answering spoken or written questions (Google, WolframAlpha, Apple, ...)
  • Translation of spoken or written text (Google, Microsoft, ...)
  • Information quality evaluation (e.g. spam detection, authorship, authority) (a long list)
  • Sentiment analysis (FaceBook, Inria (Poppy), ...)
  • Providing contextual tuned information (all big search engines, output facial expression in robots, ...)
  • Identifying objects in real world environments (all modern robots - including cars, ...)
  • ...
To be able to provide such services big data is collected and one or more machine learning techniques are applied. Targeting the intended service.
All these services are improved permanently (although some improvements or experiments are questionable as the recent manipulation of sentiments by FaceBook).


As far as I'm aware of there are only a few efforts to interconnect all these kinds of services and the underlying data. But they will. And soon. And at a large scale.
We will all have some sort of personal digital assistant available through a more or less obtrusive wearable. And switching to the non-wearables when available.

Such an assistant will be able to provide, at a certain point in time, a choice out of a fixed set of services. Of course new versions will provide enhanced and extended capabilities. But it will remain a fixed list at any point in time.

If you would want to think in terms of learning in this context, two views should be combined.
Learning what the meaning of data is will come from the combination of these huge data silos. Within each silo the meaning is somewhat implicit. Learning how to do something comes from the organizations that introduce new services or combine existing ones (the teachers).


Gaia is conceived in a way that she can learn from the beginning for and the How and the What.
And without a teacher.
She is actually learning how to handle the various data streams in order to define what the meaning is of those streams. This is the focus of the part I'm building now.

A similar learning paradigm will be applied for learning on the output side of the spectrum: what to do or what information to provide and how to achieve this.

She might even learn that there are well-known services that do certain tasks well and use them.
Or she might create her own stand-alone specialized child services once she has figured out the what and how of those services.


Concluding, I like to see both approaches of Artificial Intelligence as complementary. Especially in a short-, mid- and long term perspective.


Ronald Poell
2014-07-10

Edit 2014-07-17: Corrected the image problem.



Support

If you like to support the development of this new kind of AI you can donate Bitcoins (or fragments of it) at

1HadqhKD5EdCDpTorgxP9P8akKqycEUkwk


When Gaia will be more mature and she will be able to interact with the environment she might also need to spent money. She is not there yet but Bitcoin donations for her can already been done at:

1PyyRd6k5kSuuErAnJNpvdoNCW5ZPgD9fb