9 Things I Learned About AI Problem Solving: Elements of AI

This section talked about methods of problem solving using searches, and games, with references to related fields in AI.

One

From Euler diagram perspective, computer science is a broad field that encompasses AI. Machine Learning is considered subset of AI, and Deep Learning is subset of Machine Learning. Data Science integrates the disciplines of Machine Learning, Statistics, Programming, and Databases.

And the ultimate goal for AI to achieve is Robotics. This is because it encompasses all areas of AI such as cognitive modeling, NLP (Natural Language Processing), etc.

Two

• Searches - how to do this more efficiently in various domains
• Pattern recognition - matching nose and face to create facial recognition
• Learning from experience - Approaches to AI based on connectionism and neural nets
• Search engines use AI methods to improve their ranking algorithms
• Search engines use NLP to understand human written language 
• Search engines are looking for high-quality content that engages users.

Three

Goal formulation is the first step in problem solving. Goals help organise the behaviour by the objectives it is trying to achieve and the actions it needs to consider.

Goal is set of states and transitions in which the goal is reached and satisfied.

And each action has only one outcome, which means it is deterministic, under ideal conditions.

Four

The definition of AI, human intelligence, and philosophical consciousness leads to never-ending discussions, and theories. All this may be very interesting, but has little impact on practice of implementing AI methods.

Creativity is always a human effort. The human creative innovation effort will not be surpassed by machines. This is because creative innovation / achievement operates differently for different domains such as music, art, math, games, etc.

Automated technologies will require humans to develop, market, and monitor them.

Five

Concept of Search and Planning: Search algorithms can be used to solve problems. It is a process of finding sequence of steps to solve a problem.

There are various types to use: uninformed search algorithms, and informed search algorithms

Uninformed search algorithms: these are algorithms that are given no information about the problem other than its definition.  

There are various types of uninformed search algorithms such as Breadth-First Search, Uniform cost search, Depth-First Search, etc. The web has various sites that describes the algorithms in detail as well as provide the advantages / disadvantages of each algorithm.

Informed search algorithms: can do quite well given some guidance on where to look for solutions. Uses knowledge to find steps to the solution. Typically, solution can be found quickly, at lower cost.

Six

Asking the key questions to solve problems:

• Define choices and consequences
• What is the goal? When can we consider the problem solved
• What are the sequence of actions that would lead to the identified goal

Seven

The solution of any problem is a fixed sequence of actions. The process of looking for a sequence of actions that reaches the goal is called search. A search algorithm takes a problem as input and returns a solution in the form of an action sequence. Once a solution is found, the actions it recommends can be carried out.

Eight

Different search techniques leads to different solutions. Search and games are central to research for AI methods.

Game developers have been programming software to help create virtual worlds. Researchers are trying to teach software to play more and more sophisticated video games. And this research is not to create better game playing experiences, but to explore deeper into critical thinking, problem solving and path planning that is required to play and succeed at simple games, for starters.

For example, the Atari classic, Montezuma’s Revenge, rewarded puzzle-solving capabilities, and rethinking their strategies based on changes / obstacles in front of them. This was difficult for AI agents to learn. 

Agent is something that perceives and acts in an environment.

Nine

Search algorithms, and defining the problem gives us the impression that any AI research problem can be resolved by specifying the state and transitions between them, and finding a path from current state to your goal. But this is not the case. We cannot just use brute force (exhaustive search) to find any solution.

Whatever we choose to do in one state, and transition to the next does not mean that it will always determine the outcome we want. These are limitations of search, that can be further randomised using probability.