00:00So joining us now is one of those taking part in one of the panels at the conference today,
00:04the panel looking at how to ensure the technology is used for the public good in a resilient and
00:08open AI ecosystem. Beba Behane is founder of the AI Accountability Lab at Trinity College
00:15in Dublin. She joins us now. Thanks very much for being with us on the programme. I mean,
00:19on the surface then, what France is trying to do sounds great, doesn't it? Making AI that's ethical,
00:24that's accessible as well. Yeah, absolutely. And spending a huge amount of time, you know,
00:34looking at public AI, artificial intelligence in public interest. And some of that goes to
00:42ensuring that AI systems that are deployed into the world that are integrated into critical social
00:48infrastructure are actually tested and vetted and can do the job that they claim to do.
00:57This is one of the core and important aspects of spending and investing in
01:05artificial intelligence that is in the public interest. Is it really possible,
01:09though, to make it ethical and accountable? How on earth do you police it?
01:14Well, I tend to believe that we should focus on, you know, ensuring that those that are developing
01:23and deploying and the vendors of AI systems, we should make those bodies as accountable as
01:30possible rather than trying to think around making AI systems accountable itself. Because at the end
01:37of the day, AI is human through and through, from the data that's required to train the AI systems,
01:45from the developers and the scientists, from the companies that are, you know, developing and
01:50selling these products. Accountability squarely lies in ensuring that those that are responsible
01:58and those that are behind AI systems are accountable for the models they choose to
02:03develop and they choose to deploy. So the focus has to be ensuring, you know, relevant bodies are
02:10accountable rather than the artefact itself is accountable. And do you feel everybody's on the
02:16same track with that? I mean, there are signs, aren't there, that perhaps some countries,
02:20notably, I'm thinking, of course, of the US under Donald Trump, also China,
02:24are rather less worried about all of that. Yeah, yeah, you are right. Not everybody's on the same
02:31track. And because, you know, AI systems are tend to be seen, often tend to be seen by, you know,
02:38AI companies and tech CEOs and big tech companies and the government itself, AI systems present,
02:46you know, financial opportunity. So as a result, people tend to focus on AI systems themselves,
02:55forgetting that AI and capitalism are inherently interlinked. So the idea of pushing for
03:02accountability of the bodies that are developing and releasing AI is not something that is shared
03:09by everybody, again, but it's something that we should all focus on. I mean, it is a worry,
03:17isn't it, perhaps, that technology is sort of just running ahead. I mean, initially we had
03:22social media, which at the time, at the beginning, was thought of as a great thing,
03:25and perhaps now a lot of the negativity of social media is being uncovered.
03:30The fear is that the same thing could and will happen with AI.
03:35Yeah, again, like social media, social media has been great in, for example, connecting people
03:41across the world. But again, without the necessary guardrails, without the necessary
03:47accountability mechanisms, we have also seen, for example, Cambridge Analytica is one of the
03:54major examples where we have witnessed democratic processes being manipulated and broken down.
04:04So just like social media can become a force that destroys democracy, AI systems, again,
04:12have to be, those that are developing, there has to be clear regulatory guidelines
04:19that ensures that the AI systems that we are developing are actually likely to benefit
04:26society, not just the handful few tech companies or big tech corporations. And unfortunately,
04:33that's currently what's happening. If you look at Meta, just at the beginning of this year,
04:38they rolled back a lot of the guardrails they had, not just on the social media platform,
04:45but on their AI systems themselves. And if you look at Google, just a few days ago,
04:52they also walked back their promise, their pledge not to use AI for military purposes.
04:58And again, this is why holding these big corporations and AI companies accountable,
05:04that's why it's really important in order to shepherd the AI systems towards societal good.
05:13And what about the overall effect on AI, on the human race, if you like? I mean,
05:17there's been a lot of concern, hasn't there, about the effect on jobs?
05:23Yeah, that's one of the main concerns. And one of the concerns has been that AI is replacing people,
05:31displacing jobs, and so on. And I think something we tend to forget is the fact that
05:39despite what we hear around AI systems being fully autonomous and fully agentic, the reality
05:47is that, again, AI systems, even the most developed state of the art, large language models,
05:55or what's now known as agentic systems, are inherently human through and through. They need
06:03constant human handholding. So even though we tend to hear about AI displacing jobs,
06:11at the end of the day, what's happening is that people are, their jobs have been displaced in
06:17order to kind of, you can call it, to babysit AI systems. So there is still, humans are still
06:25required, for example, to what's known as reinforcement learning via human feedback.
06:32So even after AI systems have been developed and released, there is constant need for humans to
06:38monitor, to assess, and to rate and evaluate these AI systems, so that, you know, again,
06:45they are heading in the right direction. So what I wanted to say is that even though AI might be
06:51displacing some jobs at the end of the day, we still need humans to ensure that AI systems function
06:58the way they are supposed to be. Yeah, I noticed you mentioned humans there. I know you've spent
07:02time as well researching human behaviour as well before you went on to this. Bearing in mind
07:07everything that we've talked about, should we fear AI, do you think, or is it all positive?
07:15That is a complex question. I think the answer is somewhere in between. And again,
07:21I tend to think AI is inherently human through and through, again, from the data that's used
07:28to train it, from the various, you know, gig workers and data workers that ensure that the
07:35data that we are using, the data that we are feeding AI systems is in the right format,
07:40and to ensure that there is no toxic content in the training data set, and the scientists,
07:46and the tech companies themselves. So the idea of, you know, fearing AI itself doesn't really
07:54make sense. What we should fear is, you know, governments, and most importantly, AI companies
08:00and big tech corporations with AI. So that's what we should fear. And as I mentioned, you know,
08:07these companies have become too big to care, too big to regulate it. And, you know, after,
08:14once they have, now that they have this unprecedented power, they can do without
08:20proper guardrails, they are likely to do as they please in a way that makes financial sense to
08:26their corporations. So what we should fear and what we should care about is, again, you know,
08:32powerful people with AI, rather than, you know, AI itself, rather than the artefact itself.
08:38Good to have you on the programme today. Thanks very much for joining us,
08:41Abeba Bahani, founder of the AI Accountability Lab at Trinity College in Dublin. Thanks.
08:46Thank you for having me.
Comments