Tech and AI: 7. Where is AI Working Already?

This BBC podcast is supported by ads outside the UK.

BBC Sounds, music, radio, podcasts.

Welcome to Understand Tech and AI, the podcast that takes you back to basics to explain, explore, unpick, and demystify the technology that's becoming part of our everyday lives.

I'm Spencer Kelly from BBC Click, and you you can find all of these episodes on BBC Sounds.

Artificial intelligence has been with us for longer than you'd think.

Years, decades before chat GPT made us all sit up and pay attention, computers had already been learning how to do complex, difficult to define tasks.

It hasn't just taught robotic bodies how to see and move.

Artificial intelligence has been used in medicine, science, the military, finance, customer service.

Well today, we're going to lift the lid on different industries and see if there's some AI wriggling around inside.

I'm joined once again because we liked him so much in the previous episode by Dr.

Michael Pound, Associate Professor in Computer Vision at the University of Nottingham.

Mike, thanks for coming back again.

Oh, you're very welcome.

Thanks for having me back.

I think counter to what a lot of people would think, AI is already in use and has been in use for quite a long time in many different areas, hasn't it?

Can you give us some examples to demonstrate just how widespread it is?

Sure.

Yeah.

AI is extremely prevalent.

It has been for many years.

It's only recently really that it's received this mainstream attention, but it's been used for years before this.

So a good few examples that people might know about, so self-driving cars, they have a lot of AI in.

Medical imaging, systems that help doctors and radiographers and so on diagnose conditions.

These are being handled with AI.

So this is like looking for tumors in x-rays and things looking for tumors maybe just sorting things out to try and make your life a little bit easier we also have image classification it's very very common so many many images get uploaded onto the internet every single day and these are run-through AI algorithms that classify the content of those things and look for patterns look for things so that they can recommend them to different people and a really good example of this is on your phone if you go into your photos you know the app you use for your photos and you search you can usually find any of the photos you've ever taken with different objects in so for example if you search for pumpkin it'll bring up a lot of pictures of pumpkins if you have them in your back catalogue and that can only have been done with AI AI has gone off it behind the scenes and it's tagged these images automatically for you so that you can find them more easily it's being used in specialist areas as you say but it's also in use in the services that we use every day isn't it like social media companies search engines that kind of thing absolutely so search engines social media companies, media streaming companies that offer you TV shows and films, these are all using AI all of the time to essentially try and best work out what you would like them to do for you so that you stay on that platform.

And I believe that when Netflix or YouTube shows you a video and you watch it to the end, that is counted as a success.

The AI learns that it did a good thing there.

Whereas if it serves you up a video and you skip it after a few seconds, it marks that down as a fail and it makes a note of that and that gradually enhances its learning too.

That's absolutely right and often this benefits you because you might be presented with shows that are more interesting for you.

If you're on a Spotify playlist and you continually skip the same song, the hope is that that system will learn, okay, he doesn't like that song, right?

We're not going to convince him however many times we show it, we'll just not show that next time.

There's a word automation, which is used and has been used throughout history to talk about how machines are now doing jobs done by humans.

AI is an an example of automation, isn't it?

Even though there's nothing really mechanical and physical about it.

Yeah, the vast majority of AI is deployed in quite simple ways to just solve simple-ish problems that we could do ourselves, but we don't want to do.

And it offloads that from us.

Mike, I'm going to pause you for a couple of minutes, if you don't mind, because it is time to go back in time.

Here's our resident tech historian, Dr.

James Sumner, with a catalogue of some of AI's earliest successes.

Chess programs existed before there were even computers to run them on.

In 1948, Alan Turing and a friend, the statistician David Champernau, wrote a series of rules which they tested out using paper and pen, finding it was good enough to beat a beginner.

With better algorithms and ever-faster hardware, the quality improved steadily.

And in 1997, IBM's supercomputer Deep Blue defeated the world chess champion, Garry Kasparov.

Kasparov, after the move C4,

has resigned.

You could question, of course, whether this was solving a real-world problem at all.

Coverage of AI playing a more obviously useful role in the 1970s and 80s focused on what were called expert systems.

What kind of pain is it?

It's a sore pain, but it's not all that bad.

By making a careful clinical examination of the patient and feeding the findings into the expert system, it comes up with a probable diagnosis.

Other researchers were building important computerized replacements for human thought, which don't tend to get called AI.

Mathematical modeling of the behavior of the atmosphere since the 1950s has given us accurate weather forecasts far beyond anything humans could achieve.

One quiet but important success story is finding the shortest or quickest route between points.

As with chess, the key discoveries were made early on, but the possibilities were little known in everyday life until the 2000s, when cheap access to GPS signals and road databases made it an indispensable tool for getting around.

This illustrates something important.

AI that succeeds and becomes part of background reality often doesn't feel like AI anymore.

Perhaps because it's no longer futuristic.

That's why AI has a 70-year history but remains the next big thing.

That was Dr.

James Sumner.

And I tell you, it really is amazing to hear how long artificial intelligence has actually been around.

It still astounds me.

Mike, why do you think most of us haven't known about all this specialist AI that's been invisibly working away in the background?

Well, until sort of 2012, 2014, AI only worked about half the time.

So it might not have been such a good idea to advertise that you were using it because it might not work.

Now it's actually become a badge of of honor for companies rather than something to be avoided.

One reason is that the performance has got much better.

So we can actually offload tasks that before there was no chance that we just have to do that ourselves.

And as technology gets better in other ways, like network infrastructure and computational power, that allows us to deploy these in other places as well.

You mentioned that companies now use AI as a badge of honor.

There's a product that says it's got AI on board.

I'm always dubious about how much AI there really is on board or whether it was used at some point in the development, but it's not really AIing when you're using it.

Yeah, I think that's right.

And there have been examples of companies that use AI, but actually it was a person behind the scenes.

Oh, really?

Yeah.

Surely that's not allowed.

You wouldn't thought so.

I think that AI is used as a badge of honor in companies at the moment.

You know, we have this phrase, AI startup, which is some company using AI for some task.

It's an area that's very exciting, but also there are lots of tasks that we don't need AI to solve, right?

Online shopping, you can use AI for the recommended products, but the checkout process is very much a standard thing that doesn't really benefit from using AI in any way.

And I think that there are a lot of companies that think they're going to have more impact if they're using AI in some way and are maybe not necessarily thinking through whether it was really needed in that case.

Can I just give you an example of how stupid I think AI can be sometimes?

I bought a washing line off of an online shopping site once.

It kept recommending me the same washing line for the next three or four years.

In fact, I think it's still following me around the internet.

How many washing lines does it think a chat needs?

You can probably save 5% if you get regular monthly washing line deliveries.

And more recently, of course, since the explosion of ChatGPT, we're not just seeing products powered by AI, we're seeing products powered by ChatGPT.

What do you think that means?

So suppose you worked for a company that wanted to add tech support for their users, but you wanted to save costs and not hire anyone to actually do this.

What you could do is you could give all the frequently asked question answers to chat GPT and then when someone comes along, hope that it gives those answers back to them in a really interesting and informative way.

Of course, the risk you take if you do that is it will start spouting off Shakespeare or something like this, but it's got nothing to do with it.

And so you can't control that kind of system.

You have to be very careful.

So where we've had chat bots on websites like my bank, it will say, hello, can I help you with something?

And you type in some stuff and maybe it'll understand it and maybe I'll just think, pop me through to a human.

are we talking about the fact that websites are now they don't have to develop their own chatbots they just bring in chat GPT and throw in a load of initial information

In essence, yes, that's exactly how it could work.

You know, you might have had experiences of chatbots over previous years that have not been that great.

They understand limited things and then they quickly degenerate and think, okay, we better put you through to someone because we don't know what you're talking about.

Chat GPT is much better at understanding our sentences and giving us interesting information back.

So you might find that that you ask it a question and it gives you a very plausible answer and it acts exactly how you would want a chatbot to answer in this particular case.

But ultimately, this network has still been trained on lots and lots of internet data, which has got nothing to do with the problem that it's trying to solve.

And that means that if someone asks about those things, it might ignore what I've told it to do and start telling them about other things, you know, global politics and interesting stuff like this.

And you don't want that as a company, right?

That's not a PR discussion you really want to be having.

My bank has just advised me on how to vote.

It's confusing to me.

Okay, let's move back away from the chatbots, which are, as we say, just one type of artificial intelligence, and look more generally at what AIs are doing behind the scenes.

They're gathering lots of information, possibly about me, and they're analyzing it and acting on that.

And I would imagine that's setting off a lot of red flags for listeners.

What?

information are they after and what do you think they do with it and should we take it personally?

They're after almost everything they can get.

It would be would be the slightly inflammatory way I put this.

This is the big data idea again, isn't it?

Yeah, it is.

When you're on a website, they're going to want to capture as much information about how you use that website and what you're interested in as possible because ultimately, if they can understand you as a user, they're going to be able to make more money off you.

And so in some cases, like social media companies, that's to do with learning the kind of person you are so that they can target advertisements at you.

or things that you might interact with.

For a shop, it's going to be trying to target you with products you might buy.

So what they're trying to do is learn about your likes, your dislikes, the way that you interact with these systems, so that they can make better decisions for you, which ultimately keep you on that service.

And AI helps a lot with this because, of course, you could imagine that for the millions of users that some of these companies have, it's totally impractical to do this by hand.

There is an argument, of course, that we shouldn't be given more of what we like, but we should be exposed to new things, new ideas, new programs, new music that we wouldn't ordinarily choose.

That's a really interesting issue.

And it's one I think that is right at the heart of this, which is that if you train an AI to learn all about me and give me what I want, that's the best way to get me to buy products, to get me to spend time on your platform.

Showing me things that conflict with my views might annoy me and make me leave.

And so there is an inherent conflict where Doing what's best for society perhaps isn't what's currently best in terms of user engagement and things like this.

And this is a problem we need to deal with.

But ultimately, some of these services are provided for free, like search engines, and they have to recoup costs somehow.

And I'm not naive, I understand that that's the case.

And I've bought products that have been recommended to me, and I've watched shows and listened to music that's been recommended to me before.

So I am benefiting from these systems as well, as well as many other people.

Would you like a washing line?

You've got some spare.

I've got some spare.

Mike, thanks once again for your time.

And once again, can I ask you to return for our final episodes where we're going to try and work out whether we can control AI?

It would be my pleasure.

So what have we learned in this episode?

Well, AI has been here for some time, working and learning behind the scenes, learning to do simple tasks that we can do, but we just don't want to.

But also, it's learning about you, your tastes, and what you might like to click on next.

And that's what we'll be talking about in the next episode as we try and work out if we have all become slaves to the algorithm.

I'm John Sudworth and this is the story of my quest to ask a question.

No interviewer, parents.

You have no right to tell me not to ask questions.

It's one that's become embroiled in the fractious and fevered politics of our times.

It's very dangerous to stir up suspicion, rumors.

It's not racist at all, no, not at all.

It comes from China.

But it's a question that matters: where did COVID come from?

Fever.

The hunt for COVID's origin from BBC Radio 4.

Listen and subscribe on BBC Sounds.

When it's time to scale your business, it's time for Shopify.

Get everything you need to grow the way you want.

Like all the way.

Stack more sales with the best converting checkout on the planet.

Track your cha-chings from every channel right in one spot.

And turn real-time reporting into big-time opportunities.

Take your business to a whole new level.

Switch to Shopify.

Start your free trial today.

Want to stop engine problems before they start?

Pick up a can of CFOAM Motor Treatment.

Seafoam helps engines start easier, run smoother, and last longer.

Trusted by millions every day, C-Foam is safe and easy to use in any engine.

Just pour it in your fuel tank.

Make the proven choice with C-Foam.

Available everywhere, automotive products are sold.

Seafoam!