First steps and first questions AGTIntellegence network

Don’t go where the road leads.
Go where there is no road and leave your mark.
©Ralph Waldo Emerson

©P.Drobyshev, D.Novhorodkina

Let us first focus on the fundamental differences between AGTInw and conventional neural networks.
In a conventional neural network, a neuron is an array/summarizer of data over which a fast algorithm runs, emulating the training and operation of the neural network according to a given operating scheme and architecture. Pluses – very fast connections between neurons and assignment of weights/fast in operation in given modes/good parallelization/other minuses.
In AGTInw, on the contrary, each neuron (Agent) is a person with a rather powerful computer, but the connections between neurons happen via the Internet – not fast, at different speeds, with communication breaks.
Therefore, the task is to study the solutions used in distributed computing in the Internet and to propose/refine them so that they would suit us taking into account further development, functionality and scaling of our network.

Next, to start the Network and the AGTInw application, we need to create at least one or two algorithms in a network of the simplest kind, with the following properties:

  • urgent need for many people
  • the quality of the result is one to two orders of magnitude higher than existing analogues
  • time to obtain a result is not more than 10 times higher on average than analogues
  • the price of the result is low – about 1 AGTI

At once we see the search for information about a person (“find me”), search for airline ticket, tourist route, real estate, correct translation; perhaps there are other options; we also see the work with BIGDATA, confidential search (for example, dirt) well and look for more options. Once launched, apparently one thing will cling to another and networks will become easier to build.

The next important problem is subnetwork training. If the process of forming the network architecture is clear enough, the process of training, forming the weights is an open question. Apparently, it will be solved first for the first algorithms and gradually improved.

In general, the development plan is this – from a small embryo brain germ by developing and building subnetworks to grow the brain of a lizard, monkey, baby and then a global superintelligence.
We’re looking forward to ideas and questions!