The most effective machines have a memory. After all, results from the past provide the best indication of what may happen in the future. That is the conclusion of a theoretical study by Susanne Still, a computer scientist at the University of Hawaii (Manoa) and her colleagues. In principle, the outcome of this research applies to all “machines”, ranging from proteins to complex computers, according to Nature News.
When a machine remembers what happened to it in the past, it allows the machine to prepare, ie adapt to future circumstances. Susanne Still says she was inspired by her dance hobby, and sports in general, where better movement stands or falls with good (mental) preparation. Nevertheless, for each machine it is always necessary to consider how much energy can be spent on storing information. Only very specific information is useful for predicting the future. So a 'machine' has to make a selection.
This problem is very similar to the one that also occurs when developing a model. A good model contains only the essence of what needs to be investigated. Developing a good model is the art of omission. At the same time, you sacrifice accuracy with every simplification. It requires careful consideration of how much accuracy can be sacrificed in exchange for less complexity in the explanatory model (and thus how economically this model can handle information processing). For a 'smart' car on the road, for example, where you can install a whole battery of hard drives if necessary, this is of course much less of a problem than for a molecular machine, such as a motor protein.
Another nice clue is that this model does in fact implicitly involve Occam's razor. The model that best conforms to Occam's razor, ie is the simplest, is the most practical. She hopes that understanding the relationship between energy consumption, predictive power and memory will help scientists develop algorithms that help simplify models.
1. Susanne Still, David A. Sivak, Anthony J. Bell, Gavin E. Crooks, Thermodynamics of Prediction, Physical Review Letters, 2012, DOI: 10.1103 / PhysRevLett.109.120604
2. Susanne Still, David A. Sivak, Anthony J. Bell, Gavin E. Crooks, The thermodynamics of Prediction, arxiv.org/abs/1203.3271
3. Nature News