Thread by @DellAnnaLuca

ON ALTERNATION

Often, the best approach is not A or B, but an alternation of A and B.

For example, action and reflection. Or study and application.

It’s not about balance, but about making hypotheses and verifying them.

Thread (1/N)

For example, the human cortex 🧠 is based on the principle of *alternating* two operations: expansion and compression of information.

Only one operation would not be able to achieve any meaningful result.

Instead,

3/ Instead, the cortex alternates these two steps:
- It expands information across all possible hypotheses, relevant or irrelevant.
- It compresses information, by eliminating the hypotheses that are proved wrong by sensorial stimuli and/or experience.

This “brainstorm” is 💯

4/ In appearance, it is a very inefficient process.

It zigs & zags instead of taking a straight route.
It creates data points that that have a high chance of being discarded.

And yet, the human brain is the most efficient computing device we know of.

All thanks to alternation.

5/ (More details on the process in my paper “Techniques for the emergence of meaning in ML” on Luca-Dellanna.com

6/ In general, I’m under the impression that people and companies plateau in their development when they stop alternating.

7/ This is not only about action and reflection, but often about many alternative processes of similar effectiveness.

Alternation is a great way to create tentative data points and validate or invalidate them.

Short-term inefficient, long-term effective.

8/ As another example issued out of neurology, our cortex is not divided in one big area for thought and one big area for action.

Rather, every point in its surface performs both operations related to thought and action (In referring to L2/3 & L5, for those who know).

This…

9/ …This is another form of alternation: it provides a much faster feedback from thought to action back to though (fast feedback loops are effectively alternation)

10/ In sum:
- development is limited by feedback loops.
- alternation is a great way to get fast feedback loops.

11/ Learning is iteration.