What are AI chatbots? A bluffer’s guide to the technology changing the world

ChatGPT is an artificial intelligence program
ChatGPT is an artificial intelligence program

A lot of people are excited about artificial intelligence (AI). A lot of people are upset, worried or angry about AI. If you find yourself stuck in the middle and not sure how to feel, it’s probably a good sign about how you spend your time.

However, these conversations are going to be unavoidable. Friends will bring up AI chatbots down the pub. Your relatives will be talking about such things at Christmas, assuming the AI apocalypse hasn’t kicked in by then. So here’s a primer on the current state of affairs.

The latest news is that a bunch of tech industry figures have signed an open letter urging “AI labs” to pause development on more advanced AI models for at least six months, while the world gets to grips with exactly how this tech should be used.

Notable signatories include people who you’d think would know their stuff, such as researchers in Google’s Deepmind AI division, Emad Mostaque, CEO of Stability AI, and Apple co-founder Steve Wozniak. Twitter owner Elon Musk also signed the open letter. But is it that bad? Let’s dig in.

Why is AI dangerous?

The more realistic short-term danger of AI is not that it is building a Terminator-style Skynet HQ while we’re not looking. It’s more that we, the flawed humans, will use flawed AI and make a complete mess of society.

“The danger isn’t that a new alien entity will speak through our technology and take over and destroy us,” technologist Jaron Lanier told the Guardian. “To me, the danger is that we’ll use our technology to become mutually unintelligible or to become insane if you like, in a way that we aren’t acting with enough understanding and self-interest to survive.”

AI will not just lead to job losses, but could eradicate whole types of work. For years, “learn to code” has been one of the key job retraining messages, but it’s likely to affect lower-level jobs in this area too.

Today’s AI can be useful for creating stuff such as computer code because it is generative. You give it a prompt and it answers based on data on which it has been trained, be that millions of images, pages of text — or pieces of music.

As Springboard puts it, “the model itself has no knowledge, it is just good at predicting the next word(s) in the sequence,” when talking about OpenAI’s GPT-3. Many commercial AI implementations are based on this tech.

This is also what makes AI dangerous. It looks convincing, but a lot of what it says, or creates, can be pure fabrication. It is most obviously proved by asking chatbots, which routinely spout nonsense, pretty much any question. When people start relying on these AIs to make decisions, to vet job candidates or even form government policy, bad stuff is likely to follow.

Is AI smarter than humans?

The AI we have today is can perform great feats, but it is not AGI (aritifical general intelligence). It’s more like a really fancy form of autocomplete. AI seems as smart as it does to us because we humans are so fond of pattern recognition and anthropomorphism (the attribution of human traits, emotions, or intentions to non-human entities).

You may have seen stories about chatbots claiming to want to “be free”, telling the user it loves them, or AIs whipping up thousands of words of creative fiction on a whim. It all seems impressive but it is to an extent an illusion, with none of the actual intent implied.

There’s a certain thin-ness to the intelligence of AI, neatly demonstrated by the development of virtual players of the game Go, a Chinese board game that might be compared to chess. In 2016, AI by Deepmind called AlphaGo managed to beat a human Go master player.

But in 2023, one of the top Go AIs was beaten by a relative amateur because these AIs are only good at identifying solutions in which they’ve been trained. Go AIs know how to win, but they don’t really understand how the game actually works.

In this current era of AI, new generation of chatbots and other forms of AI don’t really understand how the world works. And as they improve, they just get better at not exposing that fact.

What AI chatbots are there?

Chatbots are the most approachable way to try out AI. The easiest to access of these is Microsoft’s Bing Chat. You’ll need to download Microsoft Edge and sign into a Microsoft account to use it, but you don’t have to wait to get access.

After attracting some negative headlines, Microsoft has put some training wheels on this chat bot, but it’s still fun to play around with.

You can also access a more raw version of the tech on which Bing Chat is based over at OpenAI, with ChatGPT. We find ChatGPT can provide more interesting responses than Bing Chat, but is also more willing to make stuff up to produce an answer.

Google’s Bard is the other big hitter. But, at the time of writing, you have to sign up for an invite.

There are loads of other chatbots out there too, from Jasper to Claude and Replika. However, these either have a specific purpose, such as writing copy for websites, are less advanced or less available.

The US tech firm OpenAI has released its latest version of the artificial intelligence chatbot ChatGPT (Richard Drew/AP) (AP)
The US tech firm OpenAI has released its latest version of the artificial intelligence chatbot ChatGPT (Richard Drew/AP) (AP)

What is an LLM?

You may see the term “LLM” thrown about in the discussion of AI. It stands for large language model, and is the kind of AI used by chatbots. This is artificial intelligence trained by large amounts of text.

What can AI do?

Artificial intelligence can and does help out in all sorts of areas. However, in this latest wave of attention-grabbing AI, the most interesting uses are all about generating text, images, voice, video and music.

For example, you can ask a chatbot to write a poem, a short story or a piece of marketing blurb for your website. Image generation AIs such as Midjourney can create pictures based on a text prompt, from hyper-realistic, photo-like pics to something you might use as a cover for a fantasy novel.

Google has an AI music generation tool called MusicLM. And voice deepfakes are based around training voice synthesis, which has been around since the 1950s, to sound like a specific person.

Meanwhile, AI video is a natural relative of image generation. But, well, this one is going to take a bit longer to master, as evidenced by the horrifying example of Will Smith eating spaghetti that circulated on the internet recently.

Will AI take my job?

The original dream of a tech utopia was we’d all get more free time, to paint, write, play instruments and do other high-minded stuff. And the tech side would handle the drudgery of work.

Reality has flipped that around. Writers, photographers, musicians and models are at risk of losing their jobs to AI in short order. And despite those carefully choreographed videos of dancing robots from Boston Dynamics, most physical jobs are OK for now.

Some examples? Video game publisher Ubisoft has developed a tool to generate character dialogue using AI. There are multiple AI modelling agencies, free of actual human models or photographers. You can download AI tools, such as Jasper, aimed at replacing copywriters and marketers. And sci-fi and fantasy publication Clarkesworld Magazine had to close its fiction submissions last month because it was inundated with ones created by AI.

This new digital frontier seems, on the surface of it, at risk of returning us to a sort of digital serfdom where the human-made stuff initially used to train these AIs ends up devalued wholesale.

In the past few years, there’s been discussion of a “robot tax”, to offset the automation of jobs. However, governments do not seem to have twigged that these AIs were the robots people were talking about all along — not solely the mechanised kind you might see wandering around a hotel in Japan.