Thursday, April 20, 2023

What Does it Mean to "Think" and Why ChatGPT isn't That

 The media has gone wild over  ChatGPT and artificial intelligence. 


Many have declared them a major leap forward in thinking.  Some see it as rivaling if not replacing human thinking, potentially becoming a new tool for discovery and scientific progress.  Beyond the hype, the reality is what it does is far from what thinking is and  how scientific or other forms of discovery occur.

Philosophers have long debated what knowledge is and how we know what we know.  This is epistemology.  How we know is connected both to what can be known and to human existence.  Rene Descartes famous “I think, therefore I am” captures the idea that thinking or self-awareness is connected to  existence, whereas Martin Heidegger’s retort “I exist, therefore I think” reverses the relationship and raise questions about the limits of human understanding and what it says about who we are as people. He also suggests that merely thinking does not prove existence or that one  is human.

 Whoever is correct the point is that knowing and self-awareness are closely connected and arguably critical to what it means to be  human.  There is no indication that ChatGPT is actually self-aware in the way humans are—at best humans program it to mimic self-awareness.

More fundamentally,  ChatGPT fundamentally misunderstands first what it means to think.  Thinking is neither the mere massing together of facts not simple inductive statements.  Many philosophers once thought that ideas were mental representations of objects that exist outside our minds. We learn by sensing things.  The concept or idea of “chair”  is simply a mental  representation of some object we perceive.  Similarly, language philosophers said that words such as “chair” correspond to  real objects which words mirror or represent.

Few believe the above is true.  Philosophers David Hume and Immanuel Kant  powerfully argued that we cannot prove empiricism.  By that, how can we empirically prove that the word chair corresponds to some empirical object?  We somehow would need to get outside of our heads to prove this correspondence is t rue and that is impossible.  Kant further contended that thinking is not when ideas in our brain correspond to external objects.  We cannot prove reality exists independent of our cognition of it.  Instead, thinking or understanding occurs as a result of our minds filtering or structuring the way we see reality.  The same is true with language as Ludwig Wittgenstein pointed out—it is our filter or medium that helps us make sense of the world.

Thinking or, better yet, cognition is not an assembly of brute facts that exist our there. What counts as facts is determined by  our minds. What is a fact is contingent on context and it does not exist independent of our knowing.  Thinking is structured by what scientist Thomas Kuhn once called paradigms.  He and others argued that scientific knowledge is not merely the linear accumulation of more and more facts which lead to discoveries.  Scientific knowledge is often the shifting or rejecting of paradigms which are based on certain assumptions about the world.  For hundreds of years the Earth was assumed to be at the center of the universe and all knowledge literally revolved around that idea.  Facts were defined by a geocentric view of the world that also impacted theology and politics.

            When Copernicus proposed a heliocentric alternative it was not the result of an accumulation of facts. It was a rejection of a model premised upon a new set of facts as defined by that model. A Newtonian vision of the world had its own assumptions and definition of facts, as does a view based on what Einstein proposed or which quantum mechanics holds.

            The point is that thinking and knowledge are not the mere accumulation of pre-existing facts amassed inductively to reach new conclusions.  There  is a connection between how and what we know and what is considered knowledge.  ChatGPT and AI are pre-programed schema based on a specific definition of facts defined by a particular paradigm or set of assumptions about the world.

            ChatGPT, and perhaps much of AI right now, operates according to what was once called  the GIGO rule—garbage in, garbage out.  The answers or conclusions they reach are based on the assumptions or knowledge that goes in. This input includes the normative choices and definitions of knowledge and facts that the programmer decides are important.

            The boldest discoveries occur when paradigms are challenged and rejected.  It is about seeing the world in ways others have not.  This is what true thinking is about.   It includes creativity. This is not just true in the arts but also in science.  ChatGPT will  never be able to compose Beethoven’s Fifth Symphony, paint the Sistine Chapel, or determine E=MC2.  Each of these examples are forms of knowledge and thinking, but they are more than grabbing facts and bunching them together.

            Mere accumulation of facts is not knowledge for other reasons.  Plato once said you can have truth without knowledge and knowledge without truth.   We may have a collection of facts but they do not add up to knowledge or truths.  But even more, and again Immanuel Kant pointed this out, many important things we know are a problem of judgment or what later philosophers would call understanding.  To know is to understand and that too is more that an accumulation of facts. It is rendering a judgment, making normative choices. 

            Many of the most important choices we make are about values.  David Hume  once declared that you cannot “derive an ought from an is.”  Even if we have facts, the facts do not tell us what to do with them.  Many of the most important choices we make such as what we should ethically do cannot be made based on facts alone.  Thinking, rather cognition, is more than a robotic act of  amassing facts, is making choices.   When someone does something stupid or wrong and we say to them “What were you thinking?” It is not always a question about ignoring facts, it is often a question of exercising bad judgment or because someone did not know better.

            In a world of ethical pluralism or where not all of us agree on what is the right thing to do, or in a world of epistemological pluralism where not all of us agree on the facts, ChatGPT and AI privilege a specific world view.  This too fails to appreciate what real thinking is.

            There is an adage that says that what separates humans from animals is the ability of the former to think.  Whether animals can or cannot think is a matter for a different day.  But what separates humans from AI is the capacity to feel emotion, to empathize, and to perceive or understand the world as it is, or as it could be.  All of this is what we mean when we say we think, and this is not what ChatGPT or AI currently does.

1 comment:

  1. Forty years ago I took a course in graduate school titled, “Management of Management.” My thesis advisor recommended it even though I was in the engineering school and the course was offered in the B-school. The subject of the course to paraphrase was;

    Because the purpose of education and training is to make stupid people look smart, it is in management’s interest to be able to differentiate between “smart” people from those who only “look” smart.

    The text we used was titled: The Invisible War, Pursuing Self-Interests at Work by a couple of clowns from UCLA by the names of Samuel Culbert and John McDonough, published in 1980. The last time I looked you could buy a copy on-line for a penny and $3.99 shipping and handling. I found it completely unreadable. I still have my first edition, it is full of notes and where I had to diagramed sentences to try to figure out what the authors were trying to say.

    At almost the end of the semester, I was getting ready to go to work one morning, worrying about the grade I was going to get in a course without homework or tests. As I looking into the mirror, I had the most horrifying and self-esteem destroying revelation (epiphany?) that I had become well trained and educated to such an extent that I could differentiate people who were smart from those who “looked” smart.

    Interacting with smart people lets me see texture, colors, and tone in a world I normally see as black, white, and shades of gray. Dealing with stupid people who act smart is not nearly as good but if one is cognizant of their shortcomings that they only know or understand what they know or understand it is bearable. They may have PHD’s, a blog, and be prolific writers but they bring nothing new to the world. In the last twenty years or so, I have observed more and more stupid people who appear to be smart promoting stupid ideas and being unwilling or unable to change.

    Dealing with stupid people being stupid is the norm of our society, it provides full employment to our criminal justice and social welfare institutions. It requires a sense of tolerance, patience, and cooperation because most of them think they are smart and become offended when you point out how really stupid they are.

    While I am forced to agree with you on AI not being able to do original work, it can definitely do the work of the stupid trying to look smart. Actually, it will do it better because the AI will have access to a much larger fund of ‘facts’. Though on the other hand, because AI is just a stupid person who looks smart, who programs it (educates and trains) will determine what it knows and understands, but that is no different than what we have now.

    ReplyDelete