Skittering on the Surface

 

We have all spent the last seventy-five years tiptoeing around the problem in AI.

Each new “breakthrough” says we can avoid doing any real work – we can use statistics or propinquity to avoid having to do what our Unconscious Mind does – a great deal of hard work. But we shoot ourselves in the foot – the strategies we adopt for handling complex text make the techniques we develop useless for handling complex text through statistics or LLM.

In Anti-Money Laundering

   (6)  This Part (other than Division 6) does not apply to a designated service covered by item 40, 42 or 44 of table 1 in section 6.

This is to help a person understand complex text – what it does not do is help any DL or LLM system to understand complex text. It violates the simplistic statistical or propinquity concepts they use.

Looking at what the Unconscious Mind must be doing is terrifying – how are we going to understand all that detail? With the serious limit on our Conscious Mind (the Four Pieces Limit), it is going to take a great effort, and it isn’t something that a team is any good at – it is not a reducible problem that a new aircraft is – and we made a real dog’s breakfast of the F-35, which should have worked well using a team approach.

But understand it we must if we are going to be successful with AI, and not circle around the “toy” end of the spectrum. It works a bit – isn’t that good enough? No, it has to be accurate and reliable, and some of the problems aren’t going to give us the luxury to learn by experience – it is not the Middle Ages any more – there is an increasing tempo and we are stuck at “how can we avoid doing any hard work”.

You have a choice - continue to play with toys in the self-imposed playpen, or face the dragon in its lair (your Unconscious Mind). Let's get to the bottom of it!