AI Diary, Part 1
- This new series of articles will give perspectives about the costs and benefits of AI to technology innovation, and society as a whole;
- There’s been a tidal wave of hype around Generative AI since the launch of services including ChatGPT, and SnapChat’s My AI;
- We wanted to ask the experts – the technologists who have been working with AI algorithms for many years – to give us their take on how AI can help (or detract from) our lives.
Generative AI – most famously, ChatGPT – proved within days of its launch that high-quality writing can be created in seconds using just a few prompts. This writing is not sourced, nor referenced, while the content it outputs is scraped from information on the internet. This has the potential to rapidly change job markets, and potentially spark a wave of joblessness.
No industry is immune to the impacts of Generative AI – not even Hollywood. Tom Hanks recently spoke out about how Deep Fake technology could ensure he continues to ‘star’ in movies “from now until kingdom come.” He commented to the BBC: “Anybody can now recreate themselves at any age by way of AI or deep fake technology. I could be hit by a bus tomorrow and that’s it, but performances can go on and on and on and on. Outside the understanding of AI and deep fake, there’ll be nothing to tell you that it’s not me and me alone.”
Rock musicians have also weighed in on the debate over rights to digital content.
Sting recently commented to the BBC: “The building blocks of music belong to us, to human beings. That’s going to be a battle we all have to fight in the next couple of years: defending our human capital against AI.”
The technologist’s viewpoint on AI
Liopa’s CTO, Fabian Campbell-West, has been using AI and deep learning for decades during his PhD, and then in the commercial world as he launched startup companies, Liopa being his current venture. Liopa uses AI algorithms to “teach” computers to read lips – so people can communicate, using a screen on a mobile device, and be understood based on their lip movements alone.
Fabian started by saying “AI is already all around us, and it has been for a long time.” But he expressed concerns that this en-masse scraping of the internet for information, could result in a reduction in the quality of content. The internet could become a place where content is endlessly consumed and regurgitated. Fabian said, “It’s like looking in a mirror with a mirror behind you and it’s an infinite reflection of the same thing.”
He went on, “These AI models are pulling content that people have contributed, but they don’t pay anything for it.”
“It could suffocate the internet, if content makers stop creating new content because their monetary models – such as ad revenue – are eroded.”
“People who are making money and people who are producing content are not necessarily the same thing – Twitter is sort of the internet town square, but people don’t get paid to Tweet – Twitter is the one making money. If you look at other platforms like YouTube, which is one of the most dominant platforms, Youtubers get an ad split based on how many people are subscribed. The real value lies with the content that people create, and that other people will want to consume.”
Fabian noted that because ChatGPT doesn’t link to a source, “no one is watching adverts” that will ultimately pay the people who create the content.
He says that ChatGPT takes on “an air of authority but ultimately it’s just a computer model that’s relying on information being good.”
But how long will that information remain high quality?
“These trends are not necessarily happening now, but it might be the thing that happens in the future.”
“In summary – I feel that good quality content will decrease and AI-generated content will increase.”
A business leader’s perspective on AI
Liopa’s COO, Richard McConnell, has been involved with building businesses in technology and telecommunications for more than 30 years. He describes his current role as “not really hands-on” but rather he oversees engineers using AI to create a commercial product.
Richard discussed the sheer speed at which AI has moved forwards:
“In terms of what AI can do and the accessibility of it – there’s been a big acceleration. It’s on an exponential curve in terms of its capabilities. AI has been around for a long time, but in the past 5-6 years the models and the ability to generate data, and also the underlying platform, have all improved significantly. There is hype around generative AI – and I do think it’s a game changer because it’s so accessible. It will change the way people do things. It becomes a much more powerful tool for everyday usage.”
“I think it will have a big impact on a lot of industries. As one example, a sales and marketing executive had told me that ChatGPT has revolutionised what he can do because it saves him so much time.”
“My view is that it’s going to have a huge impact on quite a few verticals, such as financial services, legal services, and customer service.”
How do you feel it impacts the quality of the internet?
Richard said, “With respect to quality of content – the amount of content that’s been available on the internet has been huge for many years. ChatGPT is only using info up to 2021 – so it’s not current – it’s only relying on what was already out there. It could combine good and bad information or somewhere in between. I still think it’s a step forward because it does a better job than people collecting sources manually from the internet.”
What if reputable sources – say, the Financial Times or New York Times, disallow ChatGPT to scrape their content?
Fabian said: “Web scraping is a fascinating topic because ultimately if you can read information, you can copy it. It can be made more difficult by having a paywall. I don’t think you can stop scraping, but you can take steps to minimise it.”
“Who owns data and who owns information is a fascinating subject – Facebook and other social media sites make money off the information people provide for free. But when someone goes out and extracts information en-masse that’s a different thing.”
“Fairly recently, a large company was found guilty of scraping people’s data without consent. They had broken the rules in terms of gathering data without permission.”
What about DeepFakes and the legality of computer-generated video?
“What we’re seeing now is large companies gathering huge amounts of data – for example YouTube is owned by Google. They own huge volumes of data that they can use to train models and create state of the art performance.”
“On the topic of actors like Tom Hanks being in films from now into infinity, there are a few examples of actors who have died and been edited back into films – Paul Walker in Fast and Furious – and Carrie Fisher in Star Wars.”
“Similarly, there is concern in the US with deepfakes affecting the presidential election next year, and spreading misinformation.”
“A deepfake is potentially easier to make with politicians, because they might have one way of speaking whereas an actor would have a whole range. You need a lot of video and audio of the person speaking in a certain way, to train a model to create a deepfake.”
So non-famous people are immune to deepfakes?
Fabian said: “A huge volume of available video and audio content is necessary for this to work – so normal people couldn’t be faked as easily as people in the public eye.”
“Ultimately, the public will decide and they’ll vote with their wallets. Why would you watch an AI film of a dead person rather than watch a new actor for a new generation? Just because something is possible doesn’t mean it will happen.”
Watch out for Part 2 of this AI Diary, which will discuss disinformation, social media and how AI has been good and bad for society.
How Liopa uses AI technology
Liopa is one of the world’s only companies focused purely on the area of AI-based lip reading. The award-winning technology has been based upon decades of research, starting at Queen’s Univeristy Belfast. The company sits at the intersection of three popular fields in computer science: AI, Computer Vision and Speech Recognition. For more about our healthcare-related lip reading app, SRAVI, visit www.sravi.ai. For more about our other R&D applications visit www.liopa.ai.