ChatGPT - an impressive gimmick and a stark reminder of the prevalence of AI

Here at SI HQ we’ve been playing with OpenAI’s ChatGPT.

After the standard requests like a haiku about my daughter and some Shakespeare-esque verse about Leicester City’s 2016 league win, we got down to some serious critique.

Processing language, naturally

When asked, ChatGPT acknowledges it fails the Turing test. It acknowledges that “as a language model, [its] main weakness is that it is only as good as the data it was trained on”. The old adage ‘garbage in, garbage out’ springs to mind.

By its own admission, Chat GPT is a Natural Language Processing tool that has been trained to construct sequences of words that are most probable based on a dataset of conversational text (called "WebText" and with a cut-off date of 2021). It also highlights that for different languages (“Chinese [sic], French, German, Spanish and many more”) it uses a distinct dataset in that language. It acknowledges that it is susceptible to possible biases – and engineers have flagged reams of input data as biased, false, insulting, etc. It was clearly quite ‘nervous’ when answering questions about religion and politics, for example. Aren’t we all?

With all that said, it is very effective at providing high quality copy and content. Writing a wedding speech, struggling with some snappy marketing content, checking your coding is up to scratch, and more? Ask ChatGPT. The operators are bringing in a professional plan too which will help users work even more efficiently.

Other than that though, this is just another chatbot, albeit one that shows off OpenAI’s impressive suite of technology. A more interactive version of Google or Bing, perhaps. A feedback loop of the information that has been input to the system. We need to be aware of ‘automation bias’: humans seem happy to blindly follow instructions given by an algorithm, which if another human gave us we would challenge.

Artificially intelligent, generally

ChatGPT is not ‘Artificial General Intelligence’. That is, some kind of omniscient and sentient being. Whilst authors like Ray Kurzweil have been saying for decades that this is inevitable given how reliant we are on technology, it still seems we are some way off that.

The more important question here is where we currently are with the broader AI and ML (Machine Learning) segment and where is it going. These disciplines are here to stay; becoming ever more embedded in society. The chances are you’ve already relied on or been vetted by it – when applying for a bank loan, marking a test, or preparing a spreadsheet. There are some obvious problems here, for example, rejecting bank loans based on historic data may just perpetuate inequality.

What makes Google special is being the smartest search engine – done through learning algorithms in their different guises. Amazon’s secret sauce? ML- driven logistics. The list is endless.

There is a reason that billions of dollars have been poured into this sector in recent history. Microsoft made its own move for OpenAI last week. Partly to bolster its own stable of ‘smarts’ but also to stop anyone else getting in there. This rivalry between Amazon, Google, Microsoft and others could well be the defining story of the century in technology. Interestingly when you look at deep-tech, it is the USA and China that are a cut above the rest and this is, in part, down to geopolitics: direct and indirect defence and security budgets deploying significant capital into deep tech and AI. 

As Yuval Noah Harari, author of Sapiens, talks about: AI is a human invention like no other because it doesn’t make humans more powerful or give us (or a sub-set of us) more freedom. Everything from the wheel to nuclear weapons to the contraceptive pill to an iPhone falls into that bucket. Once AI is omniscient, it very much does not do that. He notes that AI, like climate change, is one of the key super-national challenges of our time that is going to require monitoring and regulation and for nation states to collaborate.

So, is ChatGPT about to destroy the worlds of literature and content marketing? No, probably not.

But is AI one of the era-defining technologies and therefore needs a coordinated international response with aligned regulation and structure? Resoundingly: yes.