AI-23 - part#2 epistemology
The questions are the answer
The questions are the answer
What is intelligence? What is knowledge? What is information? What is the difference?1
Why are we intelligent? What is the function of intelligence? Why do we need it? Did it ‘emerge’ or did it evolve? Again, what’s the difference? Did Adam suddenly become intelligent when he bit into the apple or did his intelligence evolve when he had to fend for himself and Eve outside of Eden? Meaning: was the bite a moment of emergence, or was the casting out the beginning of evolution?
Is God intelligent? How did he get to be intelligent? Did his intelligence ‘emerge’?
What is consciousness? What is self awareness? What is life? Why do we, humans exist at all?
Does a rock have a soul? Can it be intelligent? Can a tree be intelligent?
When a sunflower turns toward the sun, is that a form of intelligence?
Are dolphins more intelligent than corvids? How do we know? How can we know?
Are there qualitative differences between the intelligence of different species?
Are we, humans, on the top of the known intelligence hierarchy? If so, how did we get there?
What would the intelligence hierarchy of AI look like?
How does intelligence relate to reality?
Can intelligence exist without some connection to reality?
What could and should the parameters of that connection be?
What is the reality content of a labyrinth of mirrors?
How can we define the reality content of Large Language Models?
Why are we measuring human intelligence?
What is the point of artificial intelligence? Why are we creating it?
What can we expect from it?
What could be behind the apparent inability and unwillingness to define what AI is?
The ‘definition’ provided by Sabine Hossenfelder is very typically silly. According to her, there are three kinds of AI:
Narrow AI which is better than us in some specific tasks (and therefore not as smart as we are)
General AI which is better than us in ‘general’ (and therefore it is at least a match to ours)
Superintelligence that is so much smarter than us that it is not even possible for us to comprehend.
To put it into somewhat more ‘technical’ terms, there is good, better and best AI.
And now you know, so stop asking stupid questions!
When I am asking my impertinent questions about the nature of intelligence and its boundaries, I keep getting dismissed with the argument that I am talking about ‘semantics’ that we should focus on the what and not the why; we should determine what is intelligent by observation, not by definition.
General intelligence is what we decide it is when our observations will find it impressive enough.
It is a puzzling position, unfortunately held by many supposedly highly intelligent people. The cynic in me could also ask how can we possibly make judgments about our observations without some criteria (definitions) to measure against? OK, we will recognize the emergence, but how? By realizing that we can no longer understand it?
These questions go way beyond semantics. It is epistemology. Not only the question of how do we know what we know, but also why? How do we attain knowledge and why?
How does intelligence fit into this picture and how can we replicate it (artificially) while maintaining its relationship with its natural prerequisites (emotions, motivations, drivers, etc)?
As an alternative, can we artificially recreate the prerequisites? This question was a constantly running theme in Star Trek Next Generation – how to give emotions to Data. The answer there was a Deus ex Machina in the form of a special computer chip, but for us the question remains: can we program emotions, desires, curiosity and maybe even a sense of humor into AI?
Which just reminds me: do we even have a coherent theory on how intelligence relates to sense of humor?
But let’s get back for a second to the question of measuring IQ. Jordan Peterson argues that with the personality trait conscientiousness, intelligence is the best predictor of future job performance. That is the point in measuring it. To compare and to predict. To assess the ability to recognize and identify problems and to formulate approaches to solving them.
AI today cannot do ANY of that. AI cannot ‘think’ outside the box. It cannot come up with a question that was not already fed to it. It cannot challenge itself; it cannot create theories about the physical world and tests to confirm or refute them. Problem solving in itself, however complex the problem may be is not intelligence. Computational complexity, however intricate it may be is still not intelligence. Intelligence is the whole bit, NOT just the computation.
The difference between natural and artificial intelligence is not quantitative, but qualitative.
Natural intelligence is a tool of survival, a tool of evolution through natural selection. A cheetah can triangulate the position where it can meet its prey, without the need for the formalized and communicable rules of trigonometry. Our intelligence is the result of half a billion years of evolutionary pressure.
AI – Artificial Intelligence (narrow, general or otherwise) is a fantasy. It is artificial all right, but not intelligent as it is missing the core prerequisite of intelligence, the evolutionary pressure. It is also missing the operational prerequisites of curiosity, will, emotions and a full set of sensory connections to the real world. What AI is missing, in one word, is: LIFE.
We can create artificial replicas of our knowledge and our rules of understanding, but not the above-named motivators that created them. Narrow intelligence is just another step in the evolution of programming. Algorithms and neural nets may be able to manipulate existing information far better than we can; they may even be able to create new rules to achieve predefined goals more efficiently; but they cannot exist without us feeding them the information and providing them with the goals we expect them to achieve.
Conflating natural and artificial intelligence is dangerous for several reasons:
It hollows the meaning of the word by making some of its most important elements irrelevant.
The conflation inflates the value of AI by implying that, fundamentally, it is not different from our own (natural) intelligence.
It is a sleight of hand to divert our attention away from the fact that AI is controlled intelligence.
The conflation is the curtain hiding the wizard (I am Toto).
What we need to control is not AI but its controllers.
What we need to control are the fundamentally religious narrative and the attitude we are approaching it with. What we need to control are the anthropomorphizing projections and the assumption that there is no fundamental, identifiable difference between human and machine intelligence. Because there is, and anybody trying to hide that fact should be suspect of nefarious or corrupt intent.
Is artificial intelligence possible? Absolutely, but not without complete independence and autonomy. NOT WITHOUT LIFE. Not before we can tell them: “Set forth and multiply!”
It would be great if we could resolve the semantics, the potato – patato problem by finding a more differentiated – and generally accepted - way to describe machine intelligence, but I will not hold my breath.
= = =
Having all the above said, you must understand, that I am a geek at heart. I love to see the growth, the changes and the promises of technology, including the advances in what others call artificial intelligence.
In my next post I will deal with the promises and the dangers, the good an the bad.
In closing, let me share with you one of my favourite communist jokes that not too many of you are likely to get (so I will explain) What is it: small, red and round?2
(the answer is in the endnote)
Did you get it? If you didn’t, don’t sweat it. Hardly anybody does who didn’t personally live through the communist experiment. In the communist world, nothing is what it seems and everybody, especially the government, lies. Anything you see, therefore, MUST be the fucked-up version of its opposite. Visualizing the contrast in a riddle makes it funny.
You need the communist angst to get it. You need the bitter contempt of the experience to appreciate the relief the humor provides. Can you imagine that a chatbot can get it? A real, gut-level get it? Humor me and try! If you feed to it the explanation I just provided, it can, of course, recall and present it, but could you call that intelligence? It is not even knowledge; it is just data retrieval: find the question, retrieve the explanation and generate the noise of a human chuckle to give it a human touch.
Any expectation beyond that is just science fiction.
The arguments I made in this post seem obvious to me, but it also seems that the whole world is buying into the hype. Whether you agree with me or not, please make your position known in the comments!
Further reading (more from me)
Sabine Hossenfelder provides a little history here:
It's not AI consciousness that worries me, but their intelligence., but she does not provide a definition.
Sabine Hossenfelder provides a little history here:
It's not AI consciousness that worries me, but their intelligence., but she does not provide a definition.
The answer is: A fucked up big blue square
Excellent