Artificial Intelligence

Is artificial intelligence inherently biased?

April 22, 2021

Recently, there has been a good deal of publicity and debate over bias in artificial intelligence. The question is: is artificial intelligence inherently biased?

AI is just a statistical methodology for analysing material in which the statistical methods are too complex for humans to understand readily. These statistical methods can involve neural networks or genetic algorithms or the like. As such, they are 'unsupervised', and run in some sense without human intervention or supervision.

The assertion, then, is that the computer has the 'intelligence,' but that is of course a misnomer. The computer is doing a statistical analysis in the same way, although with far more complexity, than would a calculator or an abacus.

Where bias could arise is in the data with which the algorithms are trained. Were data concerning prior medical experiments used to create models of medical tests used to train an algorithm, and there were gender biases in that data (i.e., the tests had all been performed on the inmates at a men's prison), then we could hardly do otherwise than to say that the software itself had a gender bias because it had been trained on data which itself had a bias. Thus the place in which we seek bias is in the training data, not in the algorithms themselves (which conform to our understandings of statistics, and I do not think that a concept like the normal or Gaussian curve have a bias in re social structures).

As far as for SecurAI, our algorithms are trained on very large datasets involving the base language (English), which we chose because it is the largest language in terms of words and the largest language in terms of available data for training.  If the language itself has an inherent bias encoded within the linguistic structure itself, we are put in a difficult position; how can we critique it using that language itself? Godel's theorem says that it is not possible to critique a structure -- any structure -- from within that structure itself. So my contention would be that -- while language may be an imperfect tool -- it is the only tool we have.

Published by

Curious to know more?

Connect with our experts to have a look at your business challenge.

Book a demo

Download whitepaper

Learn about how we apply our science to real world assignments.

Download whitepaper

You might also be interested in

Get the answers to your strategic questions in just one day.*