00:00Conor McKeown is a lecturer in digital media at the University of Stirling.
00:04Hi there, great to have you back.
00:06And look, can the AI industry really be trusted to set the rules for a technology it's racing to build?
00:14You know, it's an excellent question and it's really well phrased.
00:18I can understand where you're coming from thinking about these companies and a constant drive to innovation,
00:23creating these new technologies faster than they can really possibly hope to regulate them.
00:28But at the same time, these companies are actually the ones that understand their products best,
00:33sometimes in a way that we aren't able to see because it's happening behind closed doors.
00:37There are AI watermarks being put in place into most, if not all, of these models
00:43that create telltale signs in the way these autonomous agents can compute.
00:49So there really is some indication that these companies are already considering ways of making their LLMs stand out as
00:57unique.
00:58And that helps us to think about ways that we could make them traceable and make those companies accountable.
01:03At the same time, though, we have to be cautious about this because the innovation is currently trending towards agents.
01:10And where there are AI agents, there are AI agent swarms.
01:14And tracking the actions of thousands, if not millions, of autonomous bots sent out on single missions,
01:22it could be difficult, if not impossible.
01:25OK, so if the companies building the most powerful AI are writing their own rulebook,
01:31who makes sure they stick to it?
01:34There are different rules internationally.
01:38And starting in the U.S., really what we would hope is that because these companies are trending towards IPOs,
01:44because they are beholden to their shareholders, we would hope that any bad news about AI going rogue,
01:52any breakouts that are happening, that hopefully that would impact share price
01:56and that these companies would trend towards secrecy and privacy being protected.
02:02Now, we did see that in the social media era, but it depends very strongly on the political climate in
02:09the U.S.
02:10In China, of course, there are great inroads being taken towards creating AI frameworks
02:16that trend towards individual accountability, which could potentially in the future
02:21lead to legal repercussions for these companies should their agents go rogue.
02:26It's interesting you say that, because these companies do change their mind on a whim.
02:31You know, OpenAI started as a non-profit, then changed its mind.
02:34You know, Trump is saying, look, this agreement is morally binding.
02:37But what happens if one of these companies decide, you know what, don't want to change the rules,
02:42I'm moving the goalposts?
02:45Yeah, so far we aren't really seeing a massive indication of the kind of repercussions
02:50that are going to happen in the West anyway.
02:53We are seeing big news stories coming out there.
02:56The repercussions really should be huge, where investors should be pulling back.
03:01And if anything, it's making people more excited, which does trend towards us,
03:06maybe suggesting globally that what needs to happen is that there needs to be
03:11an AI accountability framework that's somewhat shared across the world,
03:15at least so that companies can understand that they are the ones ultimately responsible
03:19for the actions of their LLMs.
03:21Okay, so you're saying the companies are ultimately responsible.
03:24So where does it leave governments?
03:26And what happens when the next generation of AI arrives?
03:31So I think it leaves governments in an extremely difficult position.
03:36And if anything, something that I've been discussing with other research colleagues
03:40is that the actions and safeguards for this really should have been in place 15,
03:45if not 20 years ago, when this was still neural network research being done
03:50in places like Imperial College London.
03:52Where this really leaves us with the next generation of AI, I would hope,
03:57is pushing a much more educational angle, letting people know of the new dangers
04:03that exist out there on the internet.
04:05And remember, this is an internet that's over 50% AI generated now.
04:09We have that figure.
04:11We know that that data is out there.
04:13So we need to be telling people not to necessarily trust what's happening on the internet,
04:17even if it feels a lot like there's a human on the other side.
04:21It could very well be an AI agent.
04:24Yeah, proceed with caution.
04:26Colin McEwan, thank you very much indeed.
