How to create an Artificial General Intelligence

Artificial General Intelligence (AGI) is becoming more and more the focus of the AI-community. AGI is presented as improvement or advancement of AI – this is not true as AGI is from base completely different to AI. The distinction of AI and AGI is basically a philosophical one. The philosophical roots of AI are lying in the inductivism. “Inductivism is a view that argues that scientific knowledge is derived inductively from observation.”[1]. In simplified terms, this means that from a distinct number of single observations universal laws are being formed.  It is based on the assumption that it can be concluded from past events to predict future events. Summarized this leads to a generalization of individual statements. These statements are then available universally.

What does that mean for AI? AI-systems working with neural networks like Googles Deepmind for example are using the inductive form to generate a stimulus-response scheme: A specific input (visual, auditory, etc.) triggers the neural network in a specific reaction produced. The problem here is that therewith the creation of spontaneous behavior is not possible and cognitive processes cannot take place. The computer is able to learn new stuff and to use this input data to create a distinct output – of course according to the laws it has formed – but it is not able to produce new ideas and therefore it cannot react to new situations.

On contrary AGI, which has the ability to do also machine thinking (producing of new ideas, suggesting ways of improvement and reacting to completely unforeseen and new situations), is based on the empiristic deductivism. “Deductivism is the process of asserting the validity of a conclusion from a set of premises which have been allotted a truth value.”[2] The deductivistic approach is orientating on the learning behavior of human beings: “there is no such thing as instruction from without … We do not discover new facts or new effects by copying them, or by inferring them inductively from observation, or by any other method of instruction by the environment. We use, rather, the method of trial and the elimination of error”.[3] At the beginning of the process of deductivism, there are theories and hypotheses which make predictions for empirical studies. The hypothesis can be proven or disproven in further investigation.  In contrary to inductivism generalizations (= universal laws) are obtained by creativity and are always only temporary, they cannot be detected inductively. This means that they are only valid until there is a new approved hypothesis which is better!

What does that mean for AGI? The Power of neural networks used in the deductive form generates a mutation scheme: The system actively responds to the incoming stimuli (information data), builds scenarios / hypotheses and verifies these hypotheses against reality via teacher or environment and learns from it. Through a dynamic process this is later continuously replaced by an even outcome. New, unprecedented ideas can be produced and improve workflows. The computer can also react to random, unforeseen occurrences spontaneously.

What are the consequences for knowledge? Using AI knowledge takes place in a passive precipitation: Circumstances are learned codes the AI responds to them in an already learned form-> Independent thinking is not possible! In contrary to this using the deductive way knowledge is active and dynamically changes: Active adaptation, active building and hypothesis testing (the average best hypothesis wins). The computer is creative and can adapt to new situations. -> Independent thinking is possible!

Summarized the matter of creating an AGI is a philosophical one. “I am convinced that the whole problem of developing AGIs is a matter of philosophy, not computer science or neurophysiology, and that the philosophical progress that is essential to their future integration is also a prerequisite for developing them in the first place”.[4]

Isabell Kunst, 23/07/2017

[1] http://www.qualityresearchinternational.com/socialresearch/inductivism.htm

[2] http://www.qualityresearchinternational.com/socialresearch/deductivism.htm

[3] Karl Popper, https://aeon.co/essays/how-close-are-we-to-creating-artificial-intelligence

[4] David Deutsch, https://aeon.co/essays/how-close-are-we-to-creating-artificial-intelligence