Here's a challenge for any artificial intelligence taking the Turing test - "Describe a flashback or déjà vu moment you recently experienced." I must say, ChatGPT failed dismally when I asked it, replying that as an AI language model it has no such personal experiences, after which it proceeded to trot out Wikipedia's descriptions of both phenomena.
Flashbacks, or déjà vus, memories of qualia experiences - are to me an intrinsic part of what it is to be me. Walking back from the shops along a farm track today, I had one of these qualia memories of driving with my parents from London to South Wales in the early 1960s, a National Benzole petrol station, ivory-painted tin shack with pumps outside, probably somewhere in Gloucestershire, on the route of the old A40. It may have been triggered by the orchards in bloom, but the memory was very clear, and led on to the memory that my mother would tell me how even at the age of three I could identify all the petrol stations by their logos - Shell, Esho (as I'd pronounce Esso), National, Cleveland and BP.
I suspect that with time AI language models will learn - or be taught - what it is that humans define as 'flashbacks' and be able to hallucinate one up on demand, making it indistinguishable from an answer given by a random human when asked the same question. Now, whether a computer can ever tell us what it feels like to be a computer, to subjectively experience the being-ness of a computer and verbalise that to humans so we can empathise with its emotions - I doubt. Can we ever really tell whether a computer understands something, or does it merely regurgitate stored data in a logical manner? And more's to the point - will computers ever learn intuition? Does one even learn to intuit?
To me qualia experiences - such as the one I'm getting right now looking out of my window on the wood next door, bright green leaves bathed in sunshine - are the quintessence of what it is to be conscious. Even with the image in front of my eyes, and the full suite of Photoshop tools at my disposal, the image below delivers only a rough approximation of what I'm experiencing when I look up at this actual scene from the digital representation of it, which I share with you, below.
Artificial consciousness, therefore, will, I believe, be out of reach to silicon-based computing. Yes, AIs will evolve - quickly - into AGIs (artificial general intelligences) - and though they may reason and learn and even create net new knowledge, the actual conscious experience of qualia will be beyond them. What of memory? ChatGPT has a record of each one of my hundreds of chats I've had with it since January 2023, but that's just data; there's no emotion there - no surprise or delight or dismay at being asked a question, like a bright pupil may have in a classroom encounter with a teacher.
Metacognition - thinking about thinking, monitoring one's own thought process, higher-level awareness that one is in the process of thinking and how that thought process can be optimised - is something that computers may be trained to learn. With leading AI researchers saying that they no longer know what's going on within the box once machine learning gets going - we see only input and output - we have little idea how metacognition is developing within large AI systems. But even advanced metacognitive skills are no proof of consciousness, only reason.
Again, I turn to ChatGPT:
I would posit that even advanced AI systems that have high levels of metacognition would still not be conscious - with the proviso that once quantum computing becomes a reality, then things might change.
Quantum computing - which depend on qubits that, unlike bits (zeros or ones), are both zeros and ones at the same time, held in that superposition until the wave function decoheres. This property is known to be able to increase computers' number-crunching capabilities exponentially. Once the engineering problems are solved (like having to operate at near absolute-zero temperature), quantum computers could represent a huge leap forward. Now, according to Penrose and Hameroff, consciousness is a quantum phenomenon that takes place within microtubules within neurons (the Orch-OR theory first posited nearly 30 years ago).
So - say some AI researchers - future AI systems based on quantum computers rather than today's silicon-based processors - could be able to display something that is a semblance of genuine consciousness.
But even Sir Roger Penrose has stated on many occasions that consciousness is not computational. And here's the nub of the so-called Hard Problem of Consciousness - whether it is an emergent epiphenomenon of evolution, something that's limited to the brains of higher-order animals on our planet (and who knows - on other planets too?) or whether consciousness permeates the Cosmos and is everywhere - or whether it pops into existence in the right conditions. I'll leave the last word to ChatGPT:
Current understanding of consciousness is still limited, and there is no consensus on what it is, how it arises, or how it can be simulated or replicated.
This time seven years ago:
Baletowa Blues
This time nine years ago:
Two rainbows
This time ten years ago:
Dandelions in bloom
This time 11 years ago:
Warsaw's city centre - a deli-free zone
This time 12 years ago:
Patching up the holes
This time 14 years ago:
In search of the sublime aesthetic
This time 16 years ago:
Flying in from the Faroes
No comments:
Post a Comment