00:00Artificial intelligence is already becoming part of everyday work and public services.
00:06The bigger concern is what could come next. Systems more capable than people across many
00:13tasks with greater independence to act. Some researchers warn that could eventually threaten
00:18human control. The fact that we're starting to talk about the dangers of AI is no bad thing
00:24because quite clearly this this thing seems to have sort of looned up from nowhere. I mean we
00:28weren't talking about AI five years ago for instance so I mean it's a rapidly developing
00:33sort of science if I sort of use that expression and technology and of course that's all it is
00:39and let's face it you know a generation ago I remember sort of as a child in 1970s you know
00:43computers were sort of you know something of the future and of course the point is we rely on them
00:48for our everyday lives. Everything that we do is used as a sort of a computer and I've got one
00:52in
00:53my pocket it's called a telephone so we use it for sort of for obviously sort of talking to people
00:57doing our sort of shopping going on holiday etc etc etc so we live in a world full of sort
01:02of systems
01:03which you know dominate our sort of everyday lives. Now of course that's great because of course we
01:07always believe that we're in control. The reality is of course the systems are sort of far more
01:11complex than we know and of course the point is though we see what happens and of course we had
01:16this very recently with due to sort of a glitch of course the sort of the reasons which you know
01:20it was supposed to be one rogue plane from the military caused a complete sort of breakdown in
01:25terms of air traffic control and of course we saw what happened you know planes in the wrong place
01:29people being stuck in airports for a couple of days and so on and so forth. The concern centres on
01:35control. Safety features are testing whether increasingly capable AI systems can reliably follow human
01:43instructions and limits. Frontier models have already shown stronger abilities in areas such as cyber
01:49activity and autonomous decision making although tests are often conducted under control conditions
01:57and do not show that current AI can take over well defended systems. Some researchers put the risk of
02:04catastrophic loss in double figures others say those percentages cannot be scientifically verified
02:12because the hypothetical systems do not yet exist while pointing to immediate risks such as fraud,
02:19misinformation, cyber attacks and unreliable outputs. I'm reminded of the sort of the uh the Lord of
02:26the Rings when of course of um J.R. Tolkien he talked about sort of you know the the elves
02:29um and the sort
02:31of the the end of sort of the um the world as it were because of the um the orcs
02:35and sort of the
02:35sarum and um and slight divergence of course the line is don't put your sort of uh trust in humanity
02:40because of course they're fickle and we know that for as long as humanity being about they sort of they
02:44raise armies they want to take over territory we've now got a sort of something that potentially and of
02:50course this is what the sort of the great fear is it will start to sort of to want to
02:53take over
02:53territory and it will dominate and it may see human beings as being sort of not worthless creatures so
02:59much but something if it can't sort of dominate them in the way it wants to
03:02it'll somehow try and destroy it and of course as I've just said in terms of the systems that we
03:07rely on for our everyday sort of being including sort of our utilities which are all sort of you
03:11know they're run by sort of electronic systems as it were and computerized networks yeah if all of
03:16those closed down for whatever reason we would be in a pretty miserable sort of state and we'd be back
03:21to sort of to uh if not the stone age not far off that so it's a real difficult one
03:26and of course
03:26there's this idea that what we need to do is design systems which have sort of empathy and over
03:32the ability if you like to sort of to have a kill switch to stop it but of course that
03:36requires
03:37people to be in control if I can just add there is a great difficulty we're talking about handfuls
03:42of people Jeffrey Hinton you know developed the system but of course it's now in the hands of
03:46sort of you know global billionaires so of course what we're going to see is the sort of the you
03:51know
03:51the control was been by a handful of people and we've seen that already in technology and of course I
03:56think of the sort of the Elon Musk and the Jeff Bezos you know the sort of the um the
03:59corporate billionaires who
04:01sort of you know they dominate our lives in ways which is never intended and that's not a good
04:05thing either so clearly what's the way forward it's about regulation I suppose um although that's
04:11difficult because these things are sort of global entities but somehow or other we have to sort of
04:16design these systems that shut such a way is that they can't sort of either hijacked and of course we
04:20see absolutely increasing with sort of box from sort of malign states Russia in particular but also
04:25more particularly to allow this thing to sort of to develop a life of its own um and somehow sort
04:30of
04:31build in the sort of thing the human features but dare I say human features can be sort of good
04:35largely but of course we sort of know that human beings they can be sort of capricious at the sort
04:40of
04:40the best of times and they they need to sort of to be sort of monitored and policed um which
04:45of course is
04:45difficult there's no scientific consensus that AI will cause human extinction or that systems more
04:52capable than humans across most tasks will be built the debate will continue as models become more
04:59powerful for businesses public services and households the key questions are how much autonomy
05:07AI should have what safeguards are needed and who is responsible when systems go wrong
Comments