The term artificial intelligence was minted in 1956 by Stanford’s John McCarthy. Early AI pioneers included McCarthy, Alan Turing (Cambridge), John von Neumann (Princeton), Marvin Minsky (MIT) and Allen Newell (Carnegie Mellon).
The goal of the research is to produce a unified general artificial intelligence model.
We are a long way off from realizing the goal of unified AI.
To date AI systems are best described as narrow artificial intelligence machines.
These types of machines are characterized as a device is capable of doing one task within a network of knowledge domains. The goal of keeping a self driving car on the road for example can be achieved by marshalling a network of AI enabled components focused on sensing where the car is within the current environment. The AI portion of the sensor applies rules such as “Stop Car” if the front sensor detects that a barrier is ahead.
From an application perspective, AI has evolved into the following class of applications:
1. Rules engines
A large collection of “If condition then else” logic statements to interrogate the data state and perform an action on a true or false outcome.
2. Expert systems
A type of rules engine where the rules and actions are focused on a large knowledge domain such as patient symptom diagnosis
3. Rules based machine learning aka Decision Trees
Rather than apply “If condition then else” logic to solve a problem, Decision Trees apply the most likely outcome of the “If Condition” predicate. The use of fuzzy logic enables machines to quickly react to the changing environment without having to program all the scenarios. The probability values are computed by simulating the decision tree and its outcomes. Supervised learning describes the case where the parameters are manually fed into the system. Unsupervised learning describes the case where the system parameters are obtained automatically from a component whose sole job is to compute the next best move. The problem with these types of systems is that it takes an inordinate amount of time to dynamically re-compute complex decision results.
4. Neural Networks
Vision systems use neural networks to search for outlines of an image within a photo or video. No other programming is required. The term “deep learning” refers to large neural networks which process hierarchically organized information such as a box which has a top, is coloured red and is 20 cm wide and 30 cm tall. Classification of these features as a “box with a top” help vision systems identify what the device is looking at.
The expectations of what AI can do has always outpaced reality. The AI research of the 1980’s led to pessimism in the research community and the press. The lack of new investments led to “AI Winter”.
AI Winter is around the corner. AI used in self driving cars is killing drivers and pedestrians. There is no major AI must have application.
So, will computers be able to think?
No. They are a machine created and controlled by man.