Wednesday, September 28, 2016

The Humanization of Robots and Artificial Intelligence

As robots and artificial intelligence (AI) become increasingly prevalent in human society, there is a trend to make them look , talk, think and behave more like us i.e. make them unto our image.  AI software already show some  human traits. Here are a few examples:

One of the latest techniques in AI called Deep Learning Neural Networks [DLNN] (commonly used in machine vision, speech recognition and search) is actually capable of dreaming. While other AI has to be fed data for input, DLNN is also capable of generating some of its own input data and has its own ideas of what an object should look like. This it does by entering into a dream-like state. You can view some artistic/hallucinatory/horror/Salvador Dali-like dreams of DPNN contributed by Google’s AI Group at 

As a one-year old child grows its brain sees and automatically classifies objects according to their likenesses viz number of features and their degree of similarity. For example cars, trucks, cranes, fire engines etc may be classified as vehicles by his brain maybe because they all have wheels, move on roads and people sit in them.. Later on an adult will teach him that this is a truck, this is a fire engine and so on. This is exactly how a DPNN learns. The unsupervised first stage groups objects into clusters based on degree of similarity and the second supervised stage builds upon this foundation to refine and differentiate objects in each cluster.

In Nature, only members of the same species will be capable of reproduction, not to mention breeding with members of the same family is taboo. This is called Speciation. One form of AI called Genetic Algorithms (GA)  ‘breeds’ solutions to problems by having a gene pool and evolving through generations till the best (fittest) solution is output has Speciation rules in its algorithms. Problems are encoded as strings of chromosomes and breeding is by cross-over and mutation of these strings. Simulating evolution of hundreds or thousands of generations, the GA evolves to produce better and better solutions. Some uses of GA are  optimising transportation schedules, work and school time-tables, portfolio creation in financial markets etc

As in humans, the evolutionary process ‘chooses’ the fittest parents to allow them to produce children who will in general be fitter than their parents. GA also fight for the right to reproduce. This they do by fighting in tournaments where the fitness function is defined. And winners will be allowed to reproduce the next generation. Unlike humans however, the defeated will be culled. They can also practice Elitism if this is decided by their programmers. Only a select group with desired characteristics will be classified as elites and allowed to breed. Degree of social mobility can be adjusted from time to time.

Fuzzy Logic is another form of AI which basically tries to make machines  look at the world in a more human way. The human brain can look at a problem or an object in a holistic way that machines can’t. One of the reasons for this is that humans think and see things not as digitally binary or black and white but degrees of truth. Thus a man may be described as somewhat tall or the temperature may be described as cooling-and we will understand it. Fuzzy Logic enables rules to be encoded in degrees of truth by having fuzzy membership functions (see image) where characteristics can overlap. The Japanese use a lot of Fuzzy Logic in their consumer appliances from rice cookers to air conditioners, refrigerators and washing machines. And the reason why your fuzzy logic rice cooker can consistently cook a perfect pot of rice. It can measure the realtime relationship between temperature, moisture, texture, type of rice etc and make its own judgement calls as to when and how much to reduce or increase the cooking time. 

1 comment:

  1. Today's world is no doubt a robot world in which humans are working as robots. This post is much relatable for now a days. Thank you for sharing.