Framing the Question

Look, I'm not here to be a defender of AI or, more pertinently, a defender of the big AI companies and those who run them, but this latest round of AI doom mongering really is ridiculous.  'We're all doomed!' and 'AI is going to kill us all!' scream the headlines - all based upon the utterances of some former AI developer geek.  No mechanism for how this might happen is given.  No chain of causality is posited.  All we get cited in support of these claims are recent incidents where AI agents have 'unilaterally' hacked into websites and databases.  The thing is that if you start looking into these incidents, then you'll find, as with most of the 'bad stuff' attributed to AI, you'll find that an element of human error lies at their root. When Ai 'misbehaves', it is the result of insufficient parameters having been set by those who originally instigated the task that resulted in this behaviour.  The AI was merely trying to complete some kind of task, usually involving gathering data, but had not been set boundaries as to exactly how it should go about this, what was off-limits and what was permissible.  Contrary to what those who would have you believe that some AI models are sentient or 'conscious', these systems have no inherent moral code of their own, any more than any other tool has - morality is ultimately a human construct, not something possessed by inanimate objects.  You wouldn't expect a hammer or a spanner, for instance, to distinguish between right and wrong: they are merely tools that perform the function they are set to by human users. 

So it is with AI - it merely carries out the functions set for it by humans.  If you don't give it specific enough instructions, the end result will not meet expectations.  I only use AI for simple function at which I know it is quite good at - ChatGPT, for example, I use primarily for producing the sort of images that simpler systems have problems with.  But my instructions as to what I want have to be very specific to get a satisfactory result.   Otherwise, I have to spend a fair amount of time giving it additional instructions to modify the initial attempt.  Every so often it simply can't seem to understand what I want, no matter how detailed my instructions are - recently, for instance, it just could not grasp the concept of one guy slapping another guy with a fish for the artwork for an episode of 'After Hours'.  Eventually, I had to provide it with an image of a fish slapping in order to use as a guide - which resulted in exactly what I was looking for.  Mind you, sometimes with the most basic instructions it can come up with an image which is incredibly 'spot on' - the fake tabloid front page used to illustrate the latest story on The Sleaze, for instance.  The only edit I had to get it to make was to change the very real tabloid title of 'The Daily Star' which it had chosen to use for a fake tabloid name.  (The front page was sufficiently convincing - it was even dated -that I actually checked through some newspaper archives to assure myself that it wasn't a real front page).  My point is that, in general, the responses you get from AI are tailored by the wording of your query - so people really do need to learn to be careful in how they frame the tasks they set it and stop blaming the tools themselves when it all goes horribly wrong.  After all, when that set of shelves you've just put up collapses, you don't blame the drill, timber and screws you used, do you?