AI meets Uncle Harald
I am not an electronics expert. But my AI friend supposedly is, so with its help I soldered together sensors, capacitors and other exotics – and a powerful battery. Then I went to Uncle Harald’s (Harald Nyborg hardware store) for a plastic box to keep my creation tightly secured againt rain and slug invasion.
It’s 8 o’clock in the morning. I am fiddling with my electronics and a promising box found on the shelves between rack 12 and 13. Ha! Lid closed. It just fit. Battery and all. It’s my magical black box! I am a genius! My black box seems carried away too, as it starts to emit a steady stream of smoke. Crossed wires. Short circuit. Panic.
I frantically hack my box into non-conducieve parts and discretely exit the store, leaving a nerdy trace of burned circuits. Back in the office, I meet my AI friend:
– “Would it be possible to add a fuse?”, I ask nonchalantly, “Just for security reasons?”
– “Sure”, it says, “In fact, that is standard in such a design”
I seriously consider degrading it from being ‘my AI fried’ to ‘my friendly AI’, while I realise that together we are sliding fast. Downwards and blindfolded.
AI works by extracting those patterns of information from the Internet that will suffice to answer your questions, whether you ask for patterns that look like electronics or patterns that follow the rules of a scientific paper. The answer it produces may be safe, meaningful or interesting but that is simply a side effect of its superpowers of pattern recognition and recombination of information. While you converse with the AI, you try to guide it to produce patterns that fill your needs, but it won’t necessarily check if the result is safe, unless it should occur to you to ask for it.
It is commonplace to find scientific papers that are pure form without content. Their very blandness apparently slips them through journal review without friction. It is ironic that the more sophisticated AI gets, the easier it will become to ‘write’ such papers. An avalanche of well-written but completely superflous – and possibly missleading – scientific-style papers is in the making. Soon these papers will begin to cite each other and lead scientists and research funds into fruitless searches through darkened rabbitholes. But one day, this head- and heartless motion through information space may in an event that sets a lab or the world on fire, just because the scientist did not think to ask AI the question that would have added a ‘fuse’.
– “Wow! This is complicated. I don’t quite understand it, but I wonder what will happen, if I connect A and B?”