Artificial Intelligence

Artificial Intelligence (AI) was originally defined as the "the science and engineering of making intelligent machines"^[1]. The theory is that human intelligence can be reduced to a set of scientific rules that can be simulated by a machine. However, although genuine AI currently remains unachievable, AI for video games is subtly different. At its most primitive a game requires a set of algorithms that creates the illusion of intelligence in order to control the behaviour of computer-controlled characters^[2]. It is common for the characteristics of computer-controlled characters to be significantly weighted in order to create opponents that offer a challenge but do not create a one-sided or boring interaction for a human player.

Artificial Intelligence and C64 Games[edit | edit source]

There are many programs for the C64 that claim to use AI. Although many are actually a series of algorithms that can not (and do not) react to external influences, there are some that successfully strive to create rudimentary machine learning.

ELIZA[edit | edit source]

In 1965 Dr Joseph Weizenbaum created an AI psychotherapy program at MIT called ELIZA. The program works on natural language processing, comparing human inputs to programmed scripts. There is no knowledge base or understanding of human behaviour, instead the responses were achieved by pattern matching^[3]. In 1983, the Artificial Intelligence Research Group realised a version of ELIZA for the C64 which was almost identical to the original^[4].

Skool Daze[edit | edit source]

Skool Daze, released in 1984 by Microsphere is an example of a rule-based simulation^[5]. It fools the human player into believing that AI is a guiding force for the twenty-odd computer-controlled characters. Based upon existence within a virtual school building, the characters follow a predetermined school routine and do not adapt to external influences. It does appear that characters do react to the player, but these are based purely on a flag-based decisions rather than one made on patterns, data collection or weighting.

Little Computer People[edit | edit source]

One of the most popular AI programs for the C64 was Little Computer People, released in 1985 by Activision. It allowed human players to make typed requests of characters that simulated human behaviour and occupied a virtual house. The requests were not always taken, with the decision being based upon the weighting of priorities that the virtual human calculated were required to be executed^[6]. Therefore not only did the program act according to a set of internal rules, there could also be adaptation based upon external stimuli. The adverts for Little Computer People implied genuine AI, however this is not possible as the program could not withstand scientific analysis. It was a crude version of learning, as the AI only covers a very small part of the possible actions, but is often claimed to be a predecessor of modern-day virtual-life simulator games^[7].

Neuron Network Associative Memory Program[edit | edit source]

In 1987 John Walker of Fourmilab created a program for the C64 that was able to recreate a neural-network learning function using pattern-recognition^[8]. Although the brain has ten billion neurons and ten trillion connections, in only 226 lines of BASIC Walker was able to create a very small pattern recognition algorithm using tens of neurons and thousands of connections. The program works by taking any pattern (letters are used) and establishing neural connections (mimicking synapses) between every pixel in the array (mimicking neurons). The program can then identify pre-learned patterns or make approximations of new patterns. It is successful for only two or three patterns before the program breaks down; however it clearly demonstrates AI through pattern recognition.