Information Technology Reference
In-Depth Information
4 Comparison with Neural Networks
We provide a general comparison between the essence of AKL and ANN. Note, this
is meant to be only a very general, although accurate, contrast between the proposed
structure, AKL, and the long-existing well-known structure of ANN.
ANN is a structure that aids in decision making. After training, it answers, essen-
tially, “yes”-“no” questions, upon presentation of input for a task. The task has to be
within the domain that the ANN has been trained. That is, the ANN does not have
general knowledge, neither does it have the ability for creativity and for combining
multi-domain knowledge. In this sense, ANN is a specialist (rather than a polymath)
that continuously hones its skill to perform a certain task, without any creative abili-
ties . An ANN may increase its knowledge, but the newly acquired knowledge is
strictly confined within the boundaries of performing better the same task that it used
to perform before.
AKL is a structure that accommodates learning, but no decision making. An AKL
does not point to a single piece of knowledge; on the contrary, it spans several domains
of knowledge and this knowledge is expanded. The expansion is done with the incor-
poration of new K-lines into the existing AKL graph, regardless of the knowledge
domain from where those K-lines come from. By incorporating more K-lines, AKL
builds on existing knowledge and expands its ability of many different alternative
“ways to think”. As such, it facilitates creativity (i.e., allowing the generation of new
“ways to think”), by allowing the formation of paths comprised from edges from dif-
ferent K-lines (as for example, cases (a) and (d) of Figure 4). The number of new
“ways to think” is, in most cases, significantly greater than the number of the original
“ways to think” that were used to form the AKL graph due to the periodic fun-outs that
we encounter at K-line intersections. In this sense, AKL is the creative polymath that
continuously expands its knowledge and increases its chances for creative thinking .
5 Conclusion
We present AKL, a novel approach in how to use K-lines. We illustrate via two appli-
cations of AKL that the proposed method can solve a problem that no other known
method can solve as easily, as well as how an AKL can be used to facilitate a novel
way of machine learning. It is hoped that the proposed structure is a potential tool for
building intelligent systems, complementary to ANN.
Our future plans include refining the proposed method and investigating its potential
for machine learning (such as the TTT example outlines here) and for artificial creativ-
ity . For machine learning , our immediate plan is to formulate a way to automate the
testing of the TTT example described earlier. Note, testing of our methodology as de-
scribed in Section 3 is possible, but it is cumbersome since it relies on continuous hu-
man participation. Since it is desirable that there is a large number of moves and games
played in order for the AKL to acquire a substantial amount of knowledge that can be
used by the machine, it is impractical to utilize an actual human for this purpose. There-
fore, the input of a human participant has to be automated. Fortunately, for the TTT
game outlines here, the entire knowledge space can be generated since this particular
game is fairly small. We are currently working on formulating the problem of how this
Search WWH ::

Custom Search