Meta Skills

Meta Skills |

As Kevin Kelly depicted in his book, What Technology Wants (2010), the technium has its own autonomy and evolves with human beings, the similar discussion occurs in Meta Skills by Marty Neumeier. Especially, Kevin and Marty both mentioned that the future of technology must head to the destination, which has three characteristics:

1. Generating information more easily and with more quality

2. Spreading information more quickly and more widely

3. Accumulating information with more accessibility and with more quantity

Meta Skills, Marty Neumeier, 2012

While heading to this destination, the machine is getting "smarter" and "human-like". Hence the machines can help us deal with more information, and the information technologies such as Big Data, Cloud Brains, Search Engines, Internet of Things and Robots require smarter machines.

So, what are "smarter" and "human-like" mean?

Marty mentioned that machine Learns to make only from making. The mechanism is similar to "machine learning". Machine learning requires a series of training patterns, the design of training pattern is a tough task for engineers. Taking image process as an example, the training patterns include tons of images and each image has a topic. During the training process, engineer may show a picture of dog to machine, and tell it this is the image of a dog. Then, engineer shows more and more images of dogs with different scales and different angles to machine. After thousands or even millions of iterations, the machine starts to build up a recognition system of a dog. Not finish yet, the second phase for machine learning is that the machine begins to train itself. Engineers connect the machine to the internet, and the machine start to search all the pictures of dog on the internet. The machine filters out those pictures without dogs and test its recognition system to see if any mismatches happened. After billions of iterations again, the machine formally operates in the real world, meanwhile it continues training itself without any breaking. It feeds the errors and the mismatches back to its system and re-write the source code by itself. Eventually, the machine knows how to recognize the images of dogs and also how to fix the system, and also, how to train itself! 

I think this example is good to illustrate the details about how machine gets smarter. Getting smarter is just the first phase, getting smarter by itself is a Meta Skill. I believe that the machine will evolve into the second level intelligence, meanwhile, human beings have to evolve to second level too (or 3rd level). 

Image Recognition - Semantic Segmentation, Chenxi Zhang, Liang Wang, and Ruigang Yang, 2010

Also, we apply this concept to many other technologies such as digital communication. We training the communication system by sending a series of encoded message through the channel which has noise and interference and decoding the messages at the receiver terminal. Then, check the bit error rate (BER) feedback those information to transmitter, and the transmitter can adjust the amplitude or the bandwidth for next transmission. Furthermore, digital communication is entering a new age that is mesh structure such as LDPC code. The mesh structure is also mentioned by Kevin and Marty respectively in their articles. The each transmitter sends the message to every receiver (in a possible range) and each receiver get all the possible messages from all possible transmitter. And, if some message is broken during the channel, other receivers can offer an unbroken one. So, those receiver are connected together and communicated to each other to achieve higher resolution. Most important of all, the single training evolves into group training. The BER can thus approximately reduced to Shannon's Boundary which is the limit of communication. (Claude Shannon was the father of modern communication theory, who was graduated from University of Michigan, too!).

Low Density Parity Check Code, 1963