The Polanyi paradox
Polanyi’s paradox from1966 is not truly a paradox, as what it reflects, rather than a contradiction, is a difficulty, a barrier of overcoming the development of artificial intelligence and automation. And since 1966 a lot has changed. Important technological advances have taken place and different strategies have been implemented to try to overcome this difficulty.
HOW TO “DODGE” POLANYI’S PARADOX?
We have already seen that Polanyi’s paradox shows the difficulty of automating a task that is easy for us to carry out, but difficult to explain. There have been two main strategies to overcome this difficulty.
CONTROLLING THE ENVIRONMENT
Controlling the environment; so that it is easier for a machine to perform a task. The machines work with relatively simple routines, but they find it difficult to adapt to changes in the environment. If you simplify the environment, you facilitate automation. A simple example of “simplifying the environment” can be train tracks. The train does not have to overcome obstacles of the terrain, only circulate the tracks. Another interesting example are the Kiva robots that Amazon uses for its warehouses. In the video we see how the warehouse environment has been simplified so that the robots can transport the shelves containing the products. However, it is human workers who load the products onto these shelves or choose the product from each of them that must be added to a specific order.
“TEACH” THE MACHINE
The second strategy is to try to “teach” the machine to make decisions as a human expert would. How? In contrast to the “top-down” programming strategies (from rules to results), we turn to the “bottom-up” strategies of Machine Learning (based on the example data, we train the machines to interpret the rules). In the new data-based economy, we can find examples of ML application practically everywhere. Recommendation systems, recognition of images, text or sounds etc. While the first strategy tried to adapt the environment to the limitations of the machine, in this second strategy, it is the machine that adapts to the difficulties of the environment, it “learns” from it, training through data. This development has been possible due to the greater availability of training data and processing capacity of the systems.
However, although machines can perform tasks that are impossible for humans, such as processing huge amounts of data to, for example, correlate our genome with that of other species, or certain biological variables with drugs that can cure a certain disease, all this is only a small part of what can be called real human intelligence.