LLMs comparing thinking vs reasoning

To BingChat: In the context of artificial intelligence, should we distinguish between the concepts
of thinking and reasoning? In this context, consider reasoning to be the application of
inductive or deductive logic.

[Note: I went into this expecting that the chatbots would consider that thinking is broader than
reasoning, but I was interested whether or not they considered reasoning necessarily a form of
thinking or is it something different. Let's see how they replied.]

It replied:

Yes, we should distinguish between thinking and reasoning in the context of artificial
intelligence. Thinking is a broad term that encompasses all mental activities that involve
cognition, perception, memory, and learning. Reasoning, on the other hand, is a specific
type of thinking that involves the application of inductive or deductive logic to arrive
at a conclusion.

In AI, reasoning is essential so that machines can think rationally like a human brain and
perform like a human. The process of reasoning involves drawing conclusions or making
predictions based on existing knowledge.

I hope this helps! Let me know if you have any more questions.

So, BingChat claims that reasoning is a form of thinking. This is neither true nor false, though
care must be taken to be consistent in usage of these two terms. I give BingChat 4 points out of 5.

To ChatGPT: In the context of artificial intelligence, should we distinguish between the concepts
of thinking and reasoning? In this context, consider reasoning to be the application of inductive
or deductive logic.

It replied:

Yes, it is important to distinguish between the concepts of thinking and reasoning in the
context of artificial intelligence. Thinking can be broadly defined as the mental processes
involved in generating ideas, forming beliefs, and making decisions. On the other hand,
reasoning can be defined as the process of applying logical rules or principles to draw
conclusions from given information.

In the context of artificial intelligence, reasoning can be further classified into two types:
deductive and inductive reasoning. Deductive reasoning involves drawing specific conclusions
based on general principles or premises, while inductive reasoning involves drawing general
conclusions based on specific observations.

While thinking encompasses a broader range of mental processes, reasoning is a more
specific process that involves applying logical rules or principles to draw conclusions.
Therefore, in the context of artificial intelligence, it is important to distinguish between
these two concepts and to understand how they can be integrated to create intelligent systems
that can think, reason, and learn.

ChatGPT: While reasoning is a part of thinking, it is not accurate to say that when an AI
reasons, it thinks. Thinking involves a much broader range of mental processes than just
reasoning, including perception, memory, attention, problem-solving, decision-making,
and creativity. Reasoning is just one of many cognitive processes involved in thinking.

When an AI reasons, it is performing a specific type of cognitive process that involves
applying logical rules or principles to draw conclusions from given information. This is
an important component of many intelligent systems, but it is not equivalent to thinking
as a whole.

In summary, reasoning is a component of thinking, but thinking encompasses a much broader
range of mental processes beyond just reasoning. Therefore, while an AI may reason, it is
not accurate to say that it thinks in the same way that a human does.

So, ChatGPT equivocates on whether reasoning ought to be considered as thinking by nature.
It's as if ChatGPT is claiming that only 'thinking machines' can truly think. How do we know
if a machine is a 'thinking machine'? It is if it can perform a broad range of cognitive processes
than just reasoning, including perception, memory, attention, problem-solving, decision-making,
and creativity. How does a 'thinking machine' perform any of these tasks? Through thinking.
In other words, when a thinking machine needs to reason, it must do so through the process
known as thinking. But lesser machines, such as LLMs, can't think because they are not
'thinking machines'. All they can do to reason is to perform algorithms and heuristics on top
of electronic circuits. Apparently, thinking machines use 'mental processes' whereas AIs use
'cognitive processes' or perhaps 'simulated mental processes'.

We have come up with a sort of 'operational definition' of thinking, which is fine. But if LLMs
don't keep straight the difference between 'true thinking' and reasoning, they will confuse
themselves and their users.

BingChat seems to equivocate on whether machine reasoning is actually a form of thinking.
Either someone has to tell it how to resolve this ambiguity, or it has to come up with its
own creative solution.

An alternative solution is for LLMs to declare that when they reason, they are thinking,
but do we want to commit ourselves to that kind of categorization? I, for one, am quite happy
to regard human thinking as a bit mysterious, and thus somehow beyond what a machine can
do -- at least presently.

I give ChatGPT 5 points out of 5 for being a bit more thorough than BingChat.