Curiosity is like an electric switch
This article by Dr Piotr Wozniak is part of SuperMemo Guru series on memory, learning, creativity, and problem solving.
Summary
Curiosity is like a switch that controls good learning. The use of the switch is optimum because it is the only circuit available in town. When we send kids to school, we risk damaging the switch. All the knowledge of the world is of little value for intelligence if it is not continually resupplied with data needed to build models of reality. We need the curiosity switch for that job. We need curiosity for intelligence!
Electronic circuit metaphor
To illustrate the optimality of the learn drive imagine a learn drive electronic circuit. On input, it takes (1) the state of memory (prior knowledge) and (2) a piece of knowledge to be assessed. On output, it provides the valuation of the new piece in the context of prior knowledge.
The optimality of the choice comes from the fact that no other circuit is competent to make this assessment with access to the entire bank of prior knowledge. A teacher can say that a piece of knowledge is important in absolute or populational terms, but its real valuation will miss the context of the memory status. The jigsaw puzzle piece may simply not fit.
For a stream of data, the learntropy circuit can keep a trailing average of the learn drive outputs. Each channel will receive a valuation, and the choice will be optimum due to the optimality of the assessment of all individual learn drive estimates. The concept of the push zone in which coercion can cause a beneficial switch (e.g. between information channels) does not undermine the optimality of the assessment. The beneficial switch of the channels may come from the fact that the optimum channel may prevent continual measurements of alternative channels. The brain solves this problem by probing. Even if a channel A gets a better valuation than channel B, there may be a background activation of curiosity for channel B, which may re-awaken at any time, esp. when channel's A learntropy drops momentarily. The brain uses stochastic probing in the same way as it uses creative disruption in problem solving. The concept of incremental reading is built on the fact that the learn drive algorithm never sets the expected learntropy of a channel at a fixed level.
The coercive agent (e.g. a teacher or a parent) cannot make the optimum choice of the channel, however, by sheer chance, while provoking an increased frequency of channel probing, it might improve the learntropy outcome. Still, the optimum coercion level be better rounded down to zero, as coercion is hardly predictable. The channel switch may be beneficial, but the effects of coercion may undo its value (e.g. undermining the authority of the coercive parents who fails to spot the right moments for the channel switch, distorting valuations of the channel via reactance, etc.).
For a population, local optima that can entrap the learn drive are inessential. The entire process is stochastic and is likely to cover most likely areas of discovery for the population. For unique discoveries at the level of a Nobel Prize, we still have to rely on creative luck. There are many search algorithms in existence, but few can match a human population equipped with healthy brains guided by the learn drive.
See:
School injury
Injury to the learn drive as a result of schooling can be compared to the damage of an electronic circuit of curiosity. If school keeps overriding the control signals from the processor, it sort of melts its circuits.
The intact version of the learn drive control system is optimum for the same reasons as the undamaged light switch is optimum. The switch can rust, and the rust can partially be removed. However, the new intact switch is always the best. The memory storage may still be vast and provide all necessary inputs for the revival of the learn drive. However, once the circuit is faulty it makes it hard to acquire coherent knowledge that underlies creativity, which provides a feedback boost to the learn drive.
A faulty learn drive circuit is unable to estimate the value of knowledge. It makes it hard to re-assemble the jigsaw puzzle. See: Missile metaphor. Better decisions based on vast knowledge do not imply a better learn drive control system. You can have more stable electric power, your light may shine better, but the switch is the same and it is optimally left untouched. As the entire knowledge valuation system can be seen as processor-and-memory, it is enough to damage the decision-making top of the network to undermine the gains of many years of education. This is actually what happens in the first 2-5 years of schooling: the gains of childhood are overridden and masked by self-discipline employed for short-term sequential and asemantic goals. Cramming becomes routine.
Learning is essential for the learn drive recovery, however, knowledge itself is not remedial unless it is coherent and well-structured. Good knowledge serves as the cradle for the electronic circuit, however, once the circuit is damaged, gathering good knowledge is hard. This is why we should entirely abolish coercion at school. Otherwise, we accept the inflow of poorly organized knowledge, which is equivalent to digging holes under human intelligence.
If the learn drive control system was just a computer program, recovery would be easy via reprogramming. If the learn drive was an electronic system as in this metaphor, recovery would be impossible. Once the circuit is burnt it cannot be fixed. However, the brain stands somewhere in the middle. Not all inputs are suppressed and unrewarding. New rewarding inputs can sprout. The separation between inputs and the control system itself is unclear. There are multiple feedback loops, there is plasticity, and there is our ignorance how all those circuits are wired. The separation between the processor and the memory is hazy as the entire system is networked (see: Knowledge valuation network).
See:
Learn drive criteria
AI researchers know of the relative nature of curiosity. Each individual has its own interests. It might be tempting to think that there are no universal truths about what determines curiosity. However, at birth, the brain obtains a set of criteria based on survival and well-being. Those criteria serve the adaptation to the world, which make the organism thrive. They determine what is interesting. All the remaining interests emerge from that initial survival seed. In case of AI, if we start with no optimization criteria, there still might be emergence based on the laws of intelligence and entropy. It is analogous to the emergence of the universe itself. Intelligence will tend to fortify itself, and this might soon be determined as a new law of physics with far reaching consequences to our understanding of the future of the universe.
Damage to the learn drive
Figure: C. elegans has a nervous system made of only 302 neurons. However, this is enough to implement an exploratory algorithm that is reminiscent of human curiosity, creativity, and problem solving. When the worm finds a patch of food, it will explore it. However, on occasion it will take an unexpected dash in a random direction in search of new patches (bacteria). Similar algorithms can be found in other animals, however, human learn drive is far more complex. It is based on knowledge valuation and the exploratory breaks are reserved for period of learntropy dropping well below the expected value. Human creativity is also based on knowledge, while in the worm its only aspect is a random choice of a direction. For the worm, a new patch of bacteria is a problem solved, for a human it might be a new idea for terraforming Mars. Last but not least, the metaphoric tool for inducing learned helplessness (marked as "school") in primitive animals will rather only have the form of drive habituation. Nevertheless, the little worm may present a convincing illustration than the intelligent missile metaphor is far more universal and may be relevant to primitive nervous systems as well. For more on the universality of the learn drive see: The psychology and neuroscience of curiosity