Sometimes I have decided to go off to a place without any preparation or map, a challenging adventure into the unknown, exploring my options such as where to stay when I got there. Such an approach offers the opportunity to gain access to hidden perils and delights that the individual would not have had if they limited themselves to a prepared plan of action.
Experts in artificial intelligence follow one of two paths, both which are pre-planned paths of action. One path is where they define and code all the rules the AI will follow, however, this limits the AI to the restrictive paradigms and rules of its creators. The second path is the use of so-called machine learning that uses vast datasets that the AI will explore and determine patterns from. The problem with machine learning is that the AI is unable to offer feedback how it arrived at a choice from the datasets it trawled, and it is limited by the prejudices, errors and quality of the dataset.
Ideally the AI is given a simple set of rules and map, then unleashed with an error checking capability into the real world, if the data it gets in the real world won’t match its rules and map, it updates the rules and map to reflect the real world. I think the use of machine learning or attempts to define and code every rule for an AI is both unnecessary and ineffective.
In addition, having an AI working alongside human counterparts who are researching the same project can offer a feedback loop for both sides. Rather than limit AI by blinding it with predefined rules and datasets, instead encouraging it to hunt, gather, explore, play and test data that it has access to in real-time, creating its own map and rules based upon what it finds is in my view a true example of AI.