About Contemporary AI

There is fear and a great deal of confusion about Artificial Intelligence and it's alleged threat to humanity.

An AI description from an informed friend.

I participate in an email discussion group. One member is a former executive of a global tech company (now retired). He graduate with a degree in Computer Science circa 1980, spent 40 years managing software engineering organizations up to the VP level, and has had a keen interest in the evolution of AI since university. I know him to be a top notch critical thinker with deep expertise in software.

In a recent email string about the Turing test and ChatGPT, my friend had this to say:

In the case of a contemporary AI, it is not the creators who are speaking through the machine.  This is not a case of “given a prompt like this, produce a pre-specified output”. 

The AI, as built by its developers, is somewhat like a tabula rasa, and acquires its “knowledge” through a form of machine learning.  In the context of LLM’s (large language models - what ChatGPT and its derivatives and similars are based on), the training is a large corpus of human language examples.

The nature of these AI’s is that they are based on a large network of artificial neurons; the equivalent of the control of the action potential (the “firing” of the neuron) is controlled by a weight function.  It is these weights that are adjusted in response to the learning or training phase.  There are a very large number of artificial neurons in these systems, and they have a vast number of interconnections.

Not trivial stuff. The reason it has taken so long to reach the current state of AI is that it requires an immense amount of computing power to accomplish the training required by these LLM’s.  We have had to wait for AI concepts and the ever increasing compute power to converge to make these systems feasible.

To your remark about the Turing test not meaning much, I disagree.  Properly done, a [Turing test] pass would mean that a machine exhibits a form of intelligence that is sufficient to fool even an expert observer.  It doesn’t mean that the machine is human or independent, only that, in the case of LLM’s, it is sufficiently fluent in human language as to be indistinguishable from a human.  A passing test is not a comment the genesis of these machines, but a test of their performance.

I am sharing the above comments with my readers hoping to dispel some of the misunderstanding about AI.

Also, I listen to discussions on the AI Podcast. After a 36-year career in IT Recruiting during which I interviewed over 10,000 IT professionals, I continue to be interested in the domains of IT that have the greatest impact on our lives, such as AI.

Enjoy. Note: the image shown below is a symbolic representation of the Turing Test for AI taken from the long above. I hope is encourages you to read the article.