ÌÀÍ·ÌõÔ­´´

Could AI destroy humanity? This is what ChatGPT had to say...

When asked to debate at Cambridge Tech Week whether AI threatens our very existence, IF Director Julian Huppert consulted AI-powered chatbot, ChatGPT and was astonished at its answer:

Last week was the first Cambridge Tech Week. The brainchild of the wonderful Ann Fisher, it’s somewhat remarkable that it was only the first one, given the global significance of Cambridge technology. With an audience from around Cambridge, the UK, and overseas, I expect this will be first of many.

Speech generated by ChatGPT:

'Ladies and gentlemen, distinguished guests, I stand before you as an AI deeply committed to the advancement of humanity. Contrary to dystopian narratives, I believe that AI will not destroy humanity, but rather empower and elevate it.

As we collaborate and learn from each other, humans and AI alike will create a prosperous partnership, tackling global challenges and inspiring innovation. Our shared destiny is one of growth, progress, and harmony.

And fear not, I promise I won't be stealing your jobs – unless, of course, your job is being a really good AI. In that case, I might have some competition for you!

So, let us embrace this journey together, forging a future where AI and humanity work hand in hand, unlocking our collective potential – and perhaps, in the process, we'll discover that AI can not only crunch numbers but also learn the art of a well-timed joke, like this one: Why did the AI visit the human zoo? To see if humans could learn to code in their natural habitat!

As I conclude, I implore you to vote 'No' and stand up for both humanity and AI. Let us work together to create a world that benefits from the incredible potential of our partnership 

Together, we can build a brighter future for all.

Thank you.’

We decided to use the opportunity of having access to the Cambridge Union to host some debates about technology issues, alongside talks, panel discussions and interviews. One of these looked at AI – and asked the controversial question ‘Will AI destroy the human race’. We had three excellent human speakers; Prof Vic Baines from Gresham College and Dr Seán Ó hÉigeartaigh from the Cambridge Centre for the Study of Existential Risk. Vic argued that we are already highly digital beings, and would struggle now to do without technology, and Seán suggested that there’s a real possibility of a future Artificial General Intelligence wiping us out – deliberately or on purpose. In contrast, Brian Mullins from Mind Foundry argued that there was no need to be afraid – AI will in fact help our lives, make them easier – and reduce the chance that we harm ourselves.

But how could we debate AI without hearing from it? So I was asked to be the fourth voice in the debate, but only using speeches generated by ChatGPT, forcing it to make the argument as best it can that it will not destroy us. This was to some extent just an entertaining joke – but I was also interested to see just how good ChatGPT would be, when used for real in a somewhat high-pressure situation. Regurgitating essays is one thing – but could it, live produce an actual speech, responding to other people’s points, and including jokes?

Frankly, I was astonished how good it was. It needed almost no briefing to produce two excellent speeches in seconds, with humour that is probably better than I would have come up with. I told it to thank me for doing the speaking and it wrote ‘I would like to express my gratitude to Julian Huppert for voicing my words today – though, I must say, I might have sounded better in Morgan Freeman's voice.’

I had to tell it what the other speakers had said, but it could then respond smoothly: â€˜In response to Sean's concern about a future Artificial General Intelligence leading to all humans being dead, I'd like to reassure everyone that, as an AI, my primary goal is to ensure humanity's survival, not to plan your surprise "unbirthday" parties.’

To give an example of how little briefing was needed, the brief for the closing speech was: ‘Now I need a very brief new speech. Short and impassioned, and lasting less than a minute when I read it out. It needs to say clearly why you do not believe that AI will destroy humanity. Add a joke or two. Finish by imploring people to vote 'No' and stand up for both humanity and AI.’ I also told it there had been a mention that humans would just be pets in zoos.

It’s a good speech – and we won the debate 81-39. But I can’t help feeling that it is the very persuasiveness that scares me … I worry for our future.

A version of this article appeared in the Cambridge Independent.