Database Reference
In-Depth Information
their lives in the meantime before the go-ahead is given for “theoretically rigorous”
simulation? Of course not.
In the case of realtime recommendation engines too, there are still many
questions left unanswered. We will address these head on. We will also always
clearly emphasize empirical assumptions such as the Markov property or probabil-
ity assumptions. For one thing, it would be naive and wrong to seek to derive
everything in science purely in mathematical terms and to eliminate the necessary
empirical component (expressions such as “scientifically deduced” should always
sound alarm bells). And for another, it is important to understand about assumptions
so that in individual cases, the applicability of the recommendation method can
be verified in practice. That is why a methodologically rigorous procedure as
described in the introduction is essential: a stepwise approach to a self-learning
recommendation engine.
It is also clear that new ideas and methods, such as reinforcement learning for
recommendation engines as described here, usually need to mature for years before
they are suitable for practical application. The initial euphoria, especially when
everything seems to be “mathematically sound” and proven, is usually followed by
disillusionment in practice, with countless setbacks. But practical problems should
also be regarded as an opportunity, because tackling them often leads to the most
exciting theoretical advances. And when the method is finally ready for commercial
application, this is often followed by a dramatic breakthrough.
Finally, let's pick up once more on some critical points regarding the general use
of recommendation engines (and of realtime analytics). This brings us back first
of all to the “cybernetic control” of the Soviet planned economy envisaged by the
OGAS project. Soviet economists blamed its failure on its inconsistent and piece-
meal implementation, and this has been a constant source of regret. Even now the
legend still lingers on in Russia that the Soviet economy would have developed
differently if only OGAS had been implemented consistently. As a consequence,
the “theory of economic control” - now opportunistically extended to include a
synthesis of market and planned economy - is undergoing a real revival in the
search for a “third way.” Ultimately, however, this is more about reinvigorating the
failed concept of the planned economy. The growing importance of cybernetics in
modern Russian economics is clearly a retrograde step (which does not mean to say
that the use of cybernetic approaches in economics is inherently wrong).
As we mentioned earlier, it is true that OGAS was not implemented correctly.
But it is also true that the entire concept was misguided. For one thing, predicting
key indicators in economics is difficult over the long term, and predicting an entire
economic system is impossible. The idea of controlling it completely is even more
absurd. Not to mention the fact that in a (market) economy, the state can never set
out to exercise control over the economy.
For that reason, the “father of cybernetics” Norbert Wiener excluded economics
and sociology entirely from the remit of cybernetics as a highly mathematized
science [Wien64]:
