Thinking Machines: Can AI really think?
According to me, one of the hardest questions in AI is defining it. I often try to avoid this question, as I have never found an explanation that is objective. The reason for this is, defining intelligence is itself controversial. I recently conducted a poll on my instagram with two question: What is Intelligence defined by Thought Process or Action? A smart AI is the one that acts rational or human?
The results of the poll shows the ambiguity I have been talking about. Around 56% of the people believed that Intelligence is defined by Thought Process, while the rest 44% believe it is based on action. In the next question, 65% of the people believed that a smart AI is rational, while the rest believed it should be human.
It’s ironic how we intelligent beings struggle to agree on what makes us intelligent. More often than not, AI definitions vary along these lines, behavior vs thought and rational vs human. Therefore instead of trying to define AI, I am planning to ponder over a question closely associated with it, “Can Machines Think?".
Birthplace of AI
We certainly aren’t the first people to ponder this question, and certainly won’t be the last. In fact, modern approaches of AI has been built on this very question. Alan Turing, one of the pioneers of AI and a tragic hero, discussed this very question in his article ‘Computer Machinery and Intelligence’. In the foundational paper for AI, Turing rather exceptionally replaces the question of Machines Thinking, with an experiment that has come to be known as the Turing Test, or the Imitation Game.
The experiment has three entities, a human and a machine who answer the questions from an interrogator, whose objective is to identify which one is machine. The objective of the machine is to fool the interrogator into believing that it’s human. To avoid conversation, the communication between the interrogator and participants is done by typewritten or digital means. Turing goes on to discuss the various critiques for the proposed experiment, ranging from theological arguments to mathematical ones.
Machines and Consciousness
We certainly don’t have the time to go through each and every argument presented by Turing, I have linked the paper below in case you want to go through it. One argument that I would like to discuss in some detail is “The Argument from Consciousness”. A layman explanation of this argument is that a machine that passes the Turing Test, still isn’t thinking, but only simulating thinking, while a modern day AI can write a sonnet or draw a painting, it is merely replicating patterns and not because of thoughts or emotions felt. (Prof. Geoffery Jefferson)
This argument is, for me, one of the most interesting ones presented in the paper and is possibly the most difficult to answer. Turing argues that it is impossible for others to determine our intrinsic emotions and therefore an interrogator in Turing Test will not find any use of it. He does however discuss the mystery that comes with consciousness but doesn’t believe it to be necessary for intelligence. While a satisfactory answer in itself, I generally prefer to also mention an excerpt from Richard Feynman’s lecture on Heuristics.
Feynman argues that the expectation that machines should think the same way humans do is itself wrong. We agree that airplanes and birds both can fly, but the way they do it are different. If we wanted to make a machine that runs fast, we would add wheels to it and not model it completely on a cheetah. Computers already do arithmetic in a fashion very different from humans, and trying to make it more like human is taking a step backwards. Thus dismissing the question on whether machines can think based on their lack of consciouness at present, is as illogical as saying airplanes can’t fly because it doesn’t flap its wings.
Conclusion
Before writing the article I conducted a poll to get peoples opinion on the question ‘Can Machines Think?’ and the result shows the skepticism regarding the same. 48% of the voters decline the chance of machines thinking, or possibly machines thinking the way humans do, and certainly they will have a fair reasoning for the same. The debate on whether machines can think certainly isn’t going to end anytime soon. While AI has come a long way from the Turing’s era, with machines now capable of drawing and writing coherent poems and proses, there is still a lot to be done in terms of Artificial General Intelligence, before we genuinely state that ‘Machines indeed can think’.