Towards computers with problem solving ability - Artificial Intelligence

Conversational ability is only one ingredient of human intelligence. Another important ingredient is problem-solving ability. Humans have a unique capability for solving problems. To illustrate, let us examine seven problems that require intelligence for their solutions:

Seven problems that require intelligence to solve.

  1. Searching for a better solution.
  2. Representing knowledge.
  3. Maintaining a conversation with a human.
  4. Creating artificial life.
  5. Creating artificial intelligence.
  6. Making water flow uphill.
  7. Finding a general method of problem solving that is applicable to all of these problems.

These problems seem to be listed in increasing levels of difficulty (ignoring the last one for the time being). The first five are problems discussed by this volume. The sixth seems to be insoluble, assuming we are not allowed to use a pump. Some would argue the fifth one is as well. But are these problems really increasingly more difficult? Perhaps they might be at the same level of difficulty. If we could devise a general method of problem solving (as with the seventh item in the list), then we can use it to tackle all of these problems. Humans already have that capability, since we have the ability to devise solutions to each, evaluate where the solutions fail, repeatedly propose improvements or alternative solutions, until we eventually reach a solution that satisfies us or we give up. This ability for problem solving is a key ingredient of human intelligence.

As stated in Chapter, searching behaviour is a fundamental component of intelligent behaviour. Search algorithms in computer science and AI have been continuously developed and refined over the last few decades by many people in a sustained programme of research. The word ‘research’ itself bears testimony to the fundamental search process that is being carried out. Clearly, the ability to search for alternative solutions to a problem, including the meta search problem of searching for a better search method, is a fundamental component to human intelligence and to intelligence in general. The ability to evaluate the ‘goodness’ of a solution, and to continually seek better solutions, is also a key ingredient of problem solving.

We as humans are never satisfied. It is in our nature to strive to be better, both individually and collectively. No matter how good we might have done in the past, there will always be someone who will want to do better, for whatever reason. To illustrate, we only need to look at the ingenuity of solutions put forward by past AI researchers in the search for methods to develop Artificial Intelligence.

We can characterise problem solving as a search of the solution space. Referring back to Chapter, where the importance of movement to an agent was emphasized, we can think of improvement of a system as movement performed by an agent from one part of the solution space to another. Progress is often associated with forward movement.Commonly used English phrases reflect this analogy–for example,“we have made forward progress”;“we have to take one step back to make two steps forward”;“we need to make a breakthrough”;and “we are going down the wrong path”. We can try to improve the solutions that have been implemented as models in NetLogo in this volume. Referring back to Table, the solutions described in Chapter for the first item in the list–searching for a better solution–implement only the basic algorithms, and there are many examples of other search algorithms in the literature that lead to improved solutions in various circumstances.

The solution developed for the maze-searching problem, for example, that of searching between behaviours rather than searching between paths, is inadequate if we wish to develop human level search capabilities for an AI system. For that we need the agents to search the maze like a human would–the agent needs to examine its environment from an embodied, situated perspective using sensory-motor co-ordination and be cognitively aware of the choices when they occur, and also be aware of the past choices that have been made.For the maze problem,for example, the agent needs to realise (i.e.be cognitively aware) that there are alternative paths to follow when it reaches a junction in the maze. Reviewing the solutions developed for the second item on the list– representing knowledge– the Knowledge Representation model described in Chapter implements only the basic methods of representation and reasoning, and more sophisticated solutions exist in the literature.

Substantial research has been done in the field of knowledge representation over the last five decades, and researchers have improved and are continuing to improve the solutions. More work is still required to solve some of the deep problems associated with representing knowledge such as how symbols are grounded, and how an embodied, situated agent can automatically acquire knowledge of non-trivial concepts. This work can certainly benefit from insights from other fields such as cognitive science; Gärdenfors conceptual spaces theory is one example.


All rights reserved © 2018 Wisdom IT Services India Pvt. Ltd DMCA.com Protection Status

Artificial Intelligence Topics