Here is the forword to his book:
http://www.papert.org/articles/GearsOfMyChildhood.html
From the link:
* One day I was surprised to discover that some adults--even most adults--did not understand or even care about the magic of the gears. I no longer think much about gears, but I have never turned away from the questions that started with that discovery: How could what was so simple for me be incomprehensible to other people? My proud father suggested "being clever" as an explanation. But I was painfully aware that some people who could not understand the differential could easily do things I found much more difficult. Slowly I began to formulate what I still consider the fundamental fact about learning: Anything is easy if you can assimilate it to your collection of models. If you can't, anything can be painfully difficult. Here too I was developing a way of thinking that would be resonant with Piaget's. The understanding of learning must be genetic. It must refer to the genesis of knowledge. What an individual can learn, and how he learns it, depends on what models he has available. This raises, recursively, the question of how he learned these models. Thus the "laws of learning" must be about how intellectual structures grow out of one another and about how, in the process, they acquire both logical and emotional form.*
This kind of programming environment despite the stars in here scolding it is exactly how to teach IMHO.
What this allow children to do is to tinker with words and through an input/output model allow them to learn about variables.
There is no right or wrong output here (perhaps besides gramatically) there is only input/output.
The basic of computer-science is it not?