From Google Maps to facial recognition, we’re ever more reliant on artificial intelligence. Expert MIKE WOOLRIDGE gives his reasons to be fearful – and cheerful.
We asked ChatGPT (chat.openai.com) to create a main image for this piece from the following YOU prompts: ‘the future looks bright’, ‘flat illustration’, ‘poster style’, ‘embrace technology to our advantage’ and ‘older woman’
1 AI is not as intelligent as it seems (yet)
The one golden message is that, even with the next generation of conversational software tools like ChatGPT, which is the biggest thing – possibly ever – in AI, you are not dealing with a mind. ChatGPT [an AI-powered large language model that is trained on huge amounts of data] is a computer programme that has been heavily optimised, to use the technical term, to do one thing: tell you what you want to hear.
It’s very good at that: very fluent and very eloquent. That leads us to believe that we are communicating with a mind, like a human being, but we aren’t.
2 It can’t cope with the real world
AI is all about getting computers to do things that can only be done by human brains and nervous systems. AI finds some things tremendously hard that we don’t associate with intelligence, like driving a car. Untold amounts of money have been invested in trying to get cars that can safely drive themselves and we are still not there. We are nowhere near having robot butlers.
3 Google Translate is a game-changer
AI translation tools are one of the most amazing things humanity has achieved. You can translate between pretty much any widely used language in the world. Future AI translation apps will learn about you, the language you use and the way you communicate. You will be able to have simultaneous spoken-word translations.
4 Don’t trust ChatGPT with your credit card
One of the fundamental problems with large language model (LLM) technology is that it doesn’t tell you the truth. You ask it a question; it gives you the most plausible answer. It has no conception of truth. If it doesn’t know the answer to a question it will simply make something up. I asked an early version of ChatGPT what it knew about me. Its two-line summary said I’d studied at Cambridge, which is a typical background for an Oxford professor, but I haven’t. That’s why I’d be reluctant to put one of these LLMs in any situation where it was making decisions with a real consequence. The technology is unreliable and prone to weird errors. I wouldn’t trust ChatGPT with my credit card.
5 Don’t believe everything you read on social media
AI can go to your feed, look at your personal prejudices and produce misinformation on an industrial scale about your favourite, or least favourite, politician based on the language you’ve used online – so you can relate to it if you’re a teenager in Middlesbrough or a Tory voter in the Home Counties. It’s easy to see how that might be used to influence voters to impact political elections in the UK and overseas.
Within a small number of years, it’s highly probable that on social media there will be much more AI-generated content than human-generated. It will be very difficult for us to know what is authentically human and what isn’t. Trusted sources of news will become extremely important to navigate that. Go forward 50 years and there will be vastly more AI-generated books, music and videos than there will be human-generated ones.
6 AI will transform healthcare
Doctors will use AI as a diagnostic tool in the same way that using an X-ray machine is a tool. But I would caution against becoming completely reliant on an AI diagnosis. My big concern is that when a machine tells you something that disagrees with your own personal perceptions, you have to invest some energy to figure out why the machine might be wrong. And often people can’t be bothered to do that if they’re tired. I would rather have a human being than AI. Some people won’t have that luxury – it will be AI or nothing. On the plus side, AI tools will make very expensive healthcare, only accessible to a tiny fragment of the world’s population, accessible to a huge number of people. For example, AI technology will be able to analyse scans from the other side of the world. That will be transformative.
7 It’s biased towards certain sectors of society
AI is dominated by North American, white, college-educated men. That is a widely acknowledged problem: LLMs communicate using North American (rather than British) English. This matters. A recent data set widely employed for voice-recognition technology used recordings done by white, male, college-educated Americans. If you were outside that data set – female, for instance – the voice-recognition systems didn’t do a very good job of understanding you. To build an LLM, you need lots of examples of the language. Icelandic doesn’t at the moment have a digital footprint big enough to build LLMs that can communicate in the language – that is a subtle danger about the way AI can fail to encompass a wider range of human views and experiences.
8 There is a risk of misuse
Most human programmers would decline a request to write a programme to launch a cyber attack on NHS data servers. But if you have an LLM that will do whatever you ask it to do, attacks like that are easier [in the wrong hands]. People using AI to do bad things is something we need to think about. At a recent AI summit in the UK, the discussion had moved on from existential to catastrophic risk whereby a failure of AI leads to something – like a plane crash in a highly populated area – that would be devastating, but isn’t going to lead to the end of humanity. That’s a healthy direction to take the discussion, because it’s more real, and when it’s more real it’s something you can think about mitigating.
9 AI is just a tool
AI isn’t something to be in awe of or fear. It’s going to be used to do wonderful things, like recognising tumours on X-ray scans and spotting early-onset dementia or heart disease, but it’s just a tool like any other tool; don’t talk about it as if it’s a person.
10 It could boost creativity
The goal is for AI to free people from drudgery and enable them to focus on important things that require human understanding, creativity and empathy. One unexpected application of LLMs is brainstorming new ideas: something people don’t find easy or value. Ask ChatGPT for pitches for a banana-flavoured milk drink and it will keep producing, junk or not. And they will get human creative juices flowing.
LLMs will open up opportunities for completely new art forms we can’t even imagine right now, just as whoever invented early synthesisers couldn’t have imagined rave music taking off in the 80s. I think we’re on the cusp of an explosion of people doing really creative things with AI.
Mike Wooldridge is a professor of AI at Oxford University. The Truth About AI, his Royal Institution Christmas Lectures, are on iPlayer.
Illustration: Chloe Sharp/Midjourney
By Daily Mail Online, January 13, 2024