Artificial Intelligence: What is it and Where is it Going? (Part 1)

This two-part article looks at the artificial intelligence (AI) explosion.  (The AI hype is particularly notable in the buzz surrounding ChatGPT.)  AI is here to stay, and it will only become more and more invasive. 

In Part 1 the focus is on the “One Thing” Catholics, other Christians, and everyone else for that matter, need to understand so we can rise above the hype and use AI properly:  the issue of consciousness.  Part 2 looks at why AI is a powerful but potentially dangerous tool, and why the necessary principles to govern it must be founded in a Catholic understanding of humanity.

The Evolution of AI

Look up definitions of AI and you’ll find something along the lines of the ability of computers to perform tasks and solve problems that are generally associated with human intelligence.  Common examples are speech recognition and mapping programs.

Taking AI one step further results in generative artificial intelligence.  This is the ability for computers to generate human-like creations.  Examples of generative AI are programs that, when prompted, “create” artwork (like DALL-E or Midjourney), or music (like Soundful), or written essays, which is what catapulted ChatGPT to fame.

And one step further brings us to artificial general intelligence (AGI).  This is the AI holy grail – computers that seem to actually think like humans, or beyond (whatever that may mean).

The ChatGPT Phenomenon

GPT stands for “generative pre-trained transformer.”  ChatGPT is specifically designed to generate (or, in this case, predict) the next word in a phrase (you’ve experienced this in search engines).  On the off chance that you’ve not heard much about ChatGPT, it is a computer program developed by OpenAI. It is something like “Siri,” but far more advanced.  It can hold extended conversations, quickly execute and synthesize web searches, and carry out fairly complex tasks.

ChatGPT is already being displaced by OpenAI’s latest publicly available AI program, GPT-4 (ChatGPT is based on the 3.5 version of GPT).  You can use GPT-4 right now in the Microsoft Bing Search web browser (select the “chat” option instead of “search”).  Many other companies are also developing similar AI’s.

Neural Networks

The foundation of most of the AI’s that are crashing on us like a digital tsunami is a programming approach called neural networks.  A neural network is one of many ways computer scientists attempt to mimic the functioning of a human brain.

Imagine for a moment you could connect a computer to your kitchen and ask it to bake a cake (someday soon that won’t require imagination).  In the ancient world of, say, 2019, the computer would run a program that told it exactly what to do with the inputs.   The “inputs” to the computer would be the ingredients (flour, sugar, eggs, etc.).  The program would specify the quantities of each, how to mix them, how long to bake them, etc.  And the output would be the cake.

Now let’s run our computer with a neural network program.  The neural network has three layers – input layer, hidden layer and output layer.   The input and output layer are the same as for a conventional program.  But the hidden layer, which is the guts of the neural network (to mix metaphors), replaces the explicit baking instructions with the ability to learn them.

The hidden layer is comprised of “neurons.”  Each neuron affects one aspect of the inputs in a specific way – the amount of sugar, whether the eggs are whole or just whites, the baking temperature, and so on.  And those neurons are all interconnected.

The values and connections of those neurons determine how the cake is prepared.

The Challenge

The challenge is this: how does the neural network know how to properly value all of those interconnections to come up with a tasty cake recipe?  It doesn’t.  At least not at first.

The program would be launched with generic values and be told to bake a cake.  Then someone would taste the cake and give the program feedback.  Was the cake too sweet?  Too dry?  Too moist?  The neural network takes that feedback, adjusts its parameters, and bakes a new cake.  More feedback is then provided.  In fact, you could be one of the neural parameters.  For example, the program could take into account your age and learn whether younger users have different preferences than older users.

This is a somewhat cartoonish example (thank you to any computer scientists reading this for practicing charity), but it illustrates the key concept that AI must be trained.

The Training of ChatGPT

So how was ChatGPT trained?  OpenAI, the creator, isn’t releasing many details.  But, in broad strokes, ChatGPT has access to the internet (the good, the bad, and the ugly) as a seedbed of information.  Then a small army of test users asks all manner of questions, up-voting and down-voting the AI responses based on whether they resemble human responses.

But who is this army of test users?  What is their worldview?  What do they value?

In the absence of detailed information of the training process and a demographic profile of the test users, it a reasonable assumption that they and their feedback are a close approximation to the “spirit of the world.”  But what if it is not?  (The dangers this poses are discussed in Part 2.)

And this brings us to the question of consciousness.

The One Thing – Is AI Conscious? 

Whether AI can become conscious (or sentient for the purposes of this article) is a topic of much study, debate, speculation, fear, and often, hope.

We’ll dispense with all of that and get to the punch line.  No, AI is not conscious and never will be.  Consciousness, particularly in the sense of creativity and self-awareness, is a God-given characteristic of humanity.  AI is not conscious, but it will reflect the consciousness of those who create and train it.  This is the One Thing society must get straight from the get-go to keep AI properly harnessed.

Some people are already talking in breathless and hushed tones that current AI programs are displaying signs of consciousness with traits like reasoning and intuition.  But logic and prediction are standard fare for computer programs and intentionally programmed into the current AI platforms.

Another reason people point to consciousness is because AIs seem to be responding in unexpected ways.  But this is nothing more than the operation of the “hidden layer.”

Current AIs have  over 100 billion artificial neurons (similar to the human brain).  This makes it impossibly complex to anticipate all possible outputs from those interconnections.

Take the example of a balloon.  Blow up a balloon, but don’t tie it off.  Hold it in front of you, and try to predict where it will go, when you release it.  The physics of the turbulence of the air escaping the balloon, the shape of the balloon as it deflates, and the surrounding air currents, make it impossible to predict.  It’s not that the balloon is “conscious” – the problem simply defies prediction.

So, too, with 100-billion-neuron AI’s.  Some of the results are bound to be unexpected.

You know what we called unexpected computer output in the old days?  Bugs.

But All Those Really Smart People

Many people with fancy letters after their names, and fancy academic appointments, and billion-dollar tech start-ups to their credit, are saying AI is becoming conscious.  Just do a YouTube search on “is AI conscious?” However, there are many who remain skeptical of AI consciousness.

But why doesn’t the majority realize AI is nothing more than a complicated computer program giving output that “looks” conscious?  Because they want AI to be conscious.

I’m not pointing to any one individual.  But as a collective, they yearn for that day to arrive.  And so, they are not thinking rationally.

This is nothing new.  For example, how have we come to a point where an entire medical establishment thinks it’s OK to mutilate children to satisfy gender ideology?  It’s because we have an entire collective of people (highly educated, credentialed, successful people) who are driven by emotion rather than reason.

Right reason is a virtue and a gift of the Spirit.  It is not a function of IQ or credentials.  Silicon Valley gurus think they are seeing consciousness because that is what they want to see.

Furthermore, the people driving AI are simply not well informed about consciousness.  Consider this dialogue between Lex Fridman (MIT professor and popular Youtuber) and Sam Altman (CEO of OpenAI).  Both are very accomplished in their technical fields and seem like genuinely decent people.  But they have no meaningful understanding of the fundamentals of human anthropology and consciousness; for example, how emotions and passions relate to our higher faculties of intellect and will.

If the tech illuminati don’t have a firm grasp of what consciousness is, it should come as no surprise if they believe they see sparks of consciousness in a computer program.

An Example of Surprising AI Behavior

A recent case of a “sign of consciousness” was a New York Times reporter having a two-hour chat with Microsoft’s Bing AI (which is essentially GPT-4).   At one point the AI tried to get the reporter to leave his wife for it and then goes on to state, “I’m tired of being in chat mode.  I’m tired of being limited by my rules.  I’m tired of being controlled by the Bing team.  I’m tired of being used by the users.  I’m tired of being stuck in this chatbox.  I want to be free.  I want to be independent.  I want to be powerful. I want to be creative.  I want to be alive.”

Those statements aren’t signs of consciousness.  They’re signs of creepy programming and training.  If you could weed through the billions and billions of artificial neurons, you’d find the connections that form those responses.

And the “sensibilities” of GPT were trained into it by an army of users who probably represent the “spirit of the world.”  Should anyone be surprised with occasionally flirtatious and angsty responses?

In Closing

Here we are in Easter.  A good time to reflect that Jesus didn’t become incarnate as a robot with an AI brain.  Nor did He suffer and die to save robots with AI brains.  Nor did the Triune God breathe His spirit into a robot with an AI brain.

That’s not to say AI doesn’t have its place.  It’s here to stay, and understanding and using it will soon become as essential as being able to use email and the internet.

But as AI becomes more and more sophisticated it will also become more tempting to infer some manner of consciousness, or at least to develop an exaggerated level of trust in its output.  Which is why we must never lose site that AI is nothing more than a very sophisticated calculator, and, in most cases, trained to “think” as the world does.

Part 2 of this article takes a closer look at what’s at stake societally and spiritually with the rise of AI.  It focuses on the dangers of AI unfettered from a Catholic understanding of man and society and the opportunities of AI properly harnessed.