One of the buzzwords of our times is AI. There are many discrete elements that get marketed as Artificial Intelligence - language processing, image recognition, predicting when your customers pay their bills, robotics, etc.
Arguably the most important point being made is that it’s only a matter of time before machines supersede Humans in all aspects of intelligence. Google’s Ray Kurzweil has predicted 2029 to be the year this happens [1]
It’s easy to see why we would think so given the advances we’ve made in recent decades. Computer programs now routinely beat Grandmasters in Chess. Computers are better at recognizing faces and data-driven algorithms can better predict consumer behavior.
That being said, there are fundamental limitations of Universal Turing Machines - upon which ALL computing devices without exception are based - that cannot be overcome simply by faster processors and more memory. The area of Computer Science called non-computability deals with the study of problems that cannot be solved computationally. While set theory (based on axiomatic mathematics) is itself marred with paradoxes (I’ll cover some of these in separate posts) Gödel’s incompleteness theorems[2] underscore an important point about Turing Computability - Truths that are not provable in a formal system.
The Halting Problem[3] and the Tiling Problem[4] are just two of the most common examples of this.
Now what does that have to do with AI, you might ask? Well, if the ‘I’ in AI has to live up to its full promise it must address pesky issues where the human mind still reigns strong; particularly the properties of Intuition/Insight, Empathy, Guilt, Remorse, Happiness, Grief, Creativity, Aesthetics.
These are a long list of problems that humans can solve relatively easily without requiring any formal education or training. I present two for you to ruminate over -
(1) In this chess problem outlined in one of his talks, Roger Penrose points out how it takes Fritz and Stockfish - two very popular, powerful, and purpose-built Chess programs - a very long time to realize it’s a draw. While a human realizes very quickly that white can keep moving the king on black squares to get a draw.
(2) This paper claims to demonstrate an example of a consciousness process that requires quantum mechanical elements - not available in classical Turing machines.
Even if one only considers the seemingly mechanical parts of the human body, the scale of things is astonishing. Here's a Kinesin protein 'walking' a cellular structure on a microtubule inside a cell.
My position is that the topic is far from settled at this point in time. There are clear limitations to classical computing devices - perhaps quantum computing will change some of that. But more importantly, there is still a very limited understanding of the biological processes that drive life; let alone the mental 'software' that runs on the living hardware.
More on this where Penrose expresses his views on the subject: https://www.youtube.com/watch?v=hXgqik6HXc0
References:
[1] https://www.cnbc.com/2014/06/11/computers-will-be-like-humans-by-2029-googles-ray-kurzweil.html
[2] https://plato.stanford.edu/entries/goedel-incompleteness/
[3] https://brilliant.org/wiki/halting-problem/
[4] https://math.mit.edu/~rstan/papers/tilings.pdf
#AI #Artificial Intelligence