Artificial Intelligence: 'Most obvious use cases are related to informational retrieval & synthesis'

FRANCE 24 English

Visit our website: http://www.france24.com  Like us on Facebook: https://www.facebook.com/FRANCE24.English  Follow us on Twitter: https://twitter.com/France24_en

Transcript

00:00Hoping to lay down rules for international controls on artificial intelligence while

00:15maintaining individual sovereignty over the technology, leaders of dozens of countries

00:19and hundreds of companies are gathering here in Paris for a two-day summit looking at the

00:24potentials and pitfalls of the tool, promising to make it less energy-hungry and more ethical.

00:29Ahead of the summit, French President Emmanuel Macron was touting the financial windfall

00:33AI could bring and listing investments being made in the field.

00:37Emelie Boyle tells us more.

00:41Ahead of the AI Action Summit starting on Monday in Paris, French President Emmanuel

00:48Macron gave a special interview.

00:50We're living through a technological revolution the likes of which we've rarely seen.

00:56This is a moment of opportunity for humanity.

00:59The summit will be attended by heads of state, industry leaders and experts who will try

01:04to hammer out agreements on guiding the development of the rapidly advancing technology.

01:09This has to be done through global regulation, through partnerships between private and public

01:18players so that the right behaviours emerge.

01:23What we want is for players to say when it's an artificial intelligence.

01:30AI has always been high on Macron's agenda.

01:34The president hopes to boost investment in the sector for France to stay in the game.

01:40The first battle for us Europeans is to invest, invest, invest.

01:45For us in France, tomorrow's summit will announce 109 billion euros of investment in artificial

01:50intelligence in the next few years.

01:52Geopolitics will be the major focus of the conference.

01:55Tensions are running high in the backdrop of DeepSeek's launch and Donald Trump's first

02:00few weeks in office.

02:01The American head of state has spoken of his desire to make the United States the world

02:06capital of AI.

02:09One thing that is sure when it comes to AI, it is big business.

02:12According to Goldman Sachs, within the next decade it could pour some 7 trillion dollars

02:16into the world's GDP and private investment in AI was already over 90 billion two years

02:21ago.

02:22I want to speak now to the director of operations at the company Lighton, which wants to help

02:26businesses seize the opportunities of generative AI.

02:29Mr Milo Regnal, thanks for being with us.

02:31Firstly, explain to us, we've heard so much about how AI will take people's jobs, so explain

02:36to us how it's good for business.

02:39So I think in a first instance, no one thinks that AI will be taking jobs in the next one

02:44to two years and that on the contrary, companies that don't seize the AI opportunity will be

02:50companies that will be having to lay off workers because their competitors will be growing

02:55much faster than them.

02:56So I think that is kind of the key worry for the next two to three years.

03:02After that, who knows?

03:03And things become more uncertain.

03:05Indeed, it goes so rapidly.

03:07When you think back, we've only had that chat GPG app for, I think, two years, and it's

03:11already gained so much in popularity and indeed advanced in what it can do, which is the impressive

03:18thing of AI is it can nearly teach itself things.

03:21And that's the danger we're hearing about more and more.

03:23Could AI become autonomous?

03:25Could it lead to a very grave danger?

03:27Yeah, so I think we're not yet at that point.

03:31And I think also we're not even yet at the point where companies are fully integrating

03:35AI into their base businesses.

03:39So the first step we need to work on in Europe, if we want AI to create jobs, is to increase

03:48the speed at which European companies are adopting AI in their processes.

03:52And on a side note, today, the AI systems that we are using aren't what we call kind

03:57of auto learning.

03:59So they're trained in the first point, and then they're deployed and they kind of help

04:03businesses.

04:04I think one point that we can work on a lot is we've seen AI used on the kind of chatbot

04:09form.

04:10In Europe, we have a lot of industrial safety critical system companies.

04:14And that's something where AI still has a lot of technological barriers related, for

04:20instance, to reliability, safety.

04:22So something that we can work on to kind of increase its opportunity for European industrial

04:27companies.

04:28Can you give me a concrete example, we'll say, of a company that could really profit

04:32from AI that is available today?

04:35Yeah.

04:36So today, to be very clear, the kind of most obvious use cases are related to information

04:44retrieval and information synthesis.

04:46We talk about large language models.

04:48So all of these use cases around language are those that have been proliferating.

04:53And so this means that any company that has large amounts of textual knowledge hanging

04:59around.

05:00So in particular, this can be, for instance, complex industrial companies that onboard

05:07new people coming into the company that have to learn about procedures to build certain

05:13car components, etc.

05:15Simply being able to easily access, thanks to chatbots, thanks to large databases and

05:20what we call RAG to retrieve that information.

05:23You can kind of upskill people faster and give them a faster access to information.

05:27So that's the kind of base use case, but it's the use case that's been deployed the fastest

05:31up until now.

05:32It almost sounds like a highly efficient research tool.

05:35But there's also the question then of the bias in the data being put into those systems.

05:41And you know, I was hearing from a woman on Sunday who said certain types of people are

05:47putting in the information, certain companies, certain countries are putting in information,

05:51and it will lead as a result to a lot of discrimination for a lot of other countries, notably.

05:56Yeah, that's right.

05:57And I think there are two points to kind of take that question.

06:02The first one is obviously we need to be working on building AI systems that A, look into the

06:08extent to which they might be biased and B, that kind of use various data sources to kind

06:16of contradict that bias.

06:17So for instance, up until now, these systems have been trained on much of the Internet,

06:22which happens to be very English language focused, very kind of Western centric.

06:27If you're in a country that is less so, then you want to be building up data sources to

06:32prevent that.

06:33That's the first element.

06:34The second element is that you can use these AI systems more as in quotes, reasoning machines,

06:43and you want to kind of anchor these systems in your own reality.

06:48And so that's what I was describing with companies that have large corpuses of their

06:52own data.

06:54What we do, for instance, at Lighton is we take these systems, we anchor them in a company's

06:59data and it works the same for governments, etc.

07:02And when you ask a question, the language model will be acting as a reasoning machine,

07:09but going to retrieve factual information from your data sources and replying saying,

07:14look, we have the source to your answer.

07:17This is the answer and this is where you can check to make sure that it's correct.

07:21And there are a certain number of other ways in which you can kind of bound the system

07:24for it to retrieve information from the right points.

07:26Okay, you have to have a check and balance on it.

07:29But currently, most AI is being generated by the United States or China.

07:36We're calling for global governance on the issue.

07:39We know in the past, the US has been quite lax when it comes to rules and regulations.

07:43A lot of innovators say, well, if you put too much regulations, we'll stop innovating.

07:46So if the US continues in that stealth and decides to leave the rules to one side and

07:51just to advance on innovation, won't they ultimately have all the control?

07:56So I think that's a valid and tricky point to perhaps to push back on that.

08:02I think that the firstly, the US under the Biden administration actually was very much

08:08aligned with the direction that the EU was taking on kind of AI governance, very much

08:13aligned with the overall global view that was being taken on AI governance, which is

08:18a risk-based approach, saying that these systems, as and when they become more capable and therefore

08:25the potential impact that they have is larger, we need to put in kind of high levels of safeguards

08:31and obligations.

08:33And to be honest, the EU Act has a relatively high bar for this.

08:38When systems become extremely capable, there is a bit more kind of documentation that you need

08:44to be doing to make sure that your systems are safe.

08:47I think the flip side of that is that it kind of accelerates government capacity building.

08:54So it facilitates information sharing between companies and government, which can also have a

09:00good advantage for kind of educating regulators as to what can be relevant.

09:05But you're right in the kind of very immediate fallback is that places where companies can go

09:11as fast as they want will be more competitive.

09:15It's a hard balance, isn't it?

09:16And finally, just on the individual level, are there things I know you work for companies on the

09:21issue, but on the individual level, are there things that we should be aware of as we start to

09:25use AI? Because it does look like it's pretty much unavoidable.

09:29Yeah, that's right. I think so.

09:31I also worked at a think tank called Institut Montaigne.

09:34At Institut Montaigne, we published an online training course called Destination AI to precisely

09:41try and educate as many citizens as possible on how AI functions and how they should be

09:46integrating it into their daily lives.

09:48And I think that the first element is just for people to understand the broad principle of how

09:54these systems work based on data, based on, at the moment, predicting the next token or word in a

09:59sequence of elements, precisely so that they can be aware of, for instance, AI-generated content,

10:07whether it's textual, whether it's visual, also to be able to identify situations in which they

10:13might be confronted with these kinds of AI-generated content.

10:17That's the first element.

10:19Second element, to simply be able to use these tools in their everyday lives.

10:23We were mentioning that companies that would adopt AI faster would be more competitive.

10:27That also applies on the individual citizen level.

10:30So these are kind of two elements that are very important.

10:33I'm going to have to dig my head out of the sand and face up to it and trade myself off, I think.

10:37Milo O'Reilly, thanks so much for coming in and joining us on the channel.

Category

Transcript

Recommended