The AI Panic Is Overblown

For a short while, I was of the opinion that the most unhuman-like thing about artificial intelligence was its capacity to ignore context, to forget salient details, and on occasion to brazenly make stuff up. On reflection, I’ve come to the opposite view. Miraculous though they may be, the AI models that are currently available to the general public are frustrating in precisely the same way that human beings can be frustrating. I am not yet sure whether I should be comforted or alarmed by that fact.

In recent weeks, I have been using large language model AI tools (LLMs) such as ChatGPT and Google Gemini rather obsessively, and the key thing that I have learned is that, if one wants to get the most out of them, one has to be willing to argue one’s case. Underpinning much of the coverage of the technology is the fear that, before too long, it will render many human jobs obsolete.

I must confess that I am not remotely convinced that this will come to fruition. Why not? Because, in its current iteration at least, AI delivers the best results when the prompts that it is given are specific, and, for those prompts to be specific, one has to have a solid grasp of the subject one is discussing. Yes, AI can save you a lot of time writing your PHP script, but for it to do so, you have to know enough about PHP scripts already to give it the sort of detailed instructions that you would give a human coder. If you don’t, you’ll get lost in five seconds flat.

In the handful of TV commercials that depict its use in IT, AI merrily prints hundreds of pages of perfect code in response to the sparsest of requests. In reality, it does no such thing. As a checker of syntax, fixer of indentation, and spotter of elementary errors, AI is nonpareil. As an architect of anything complex, it leaves a great deal to be desired. A good analogy might be with foreign languages.

If you speak fluent French, asking AI to write you a translation might save you a lot of time. If you do not speak fluent French — and, as a result, you cannot tell what is right and what is wrong, and why — asking AI to write you a translation is likely to be a disaster. I have found that, as a rule of thumb, you never want AI to be doing anything for you that is more than two or three steps beyond your own abilities. That way lies disaster.

This, in part, is because it lies. Bizarre as it may sound, both ChatGPT and Google Gemini are liable to teach you coding rules that do not exist; to “fix” your mistakes by suggesting precisely the same block that you explained was the problem in the first instance; and to “forget” large parts of the work that you have done together hitherto, and thus to strip out half of your progress each time you introduce a new topic. In a similar vein, both products have a tendency to fall victim to what I have come to think of as “tunnel vision.”

When testing their ability to code or correct foreign languages or do mathematics, I have got the best results by periodically switching between ChatGPT and Google Gemini so that the one that isn’t being used can get a “fresh look” at the problem. Intuitively, one might have thought that the systems would benefit from having a full record of the conversation, but often, the opposite is the case. Experience, not novelty, seems to present the bigger challenge.

And then there is the question of creativity — which, currently, the technology lacks completely. Ask it to write a short account of an event for which you have provided the salient details, and it will excel. Ask it to write a poem in the style of Shakespeare, and it will embarrass itself. This is not merely because the poem will be bad — although, invariably, it will — but because the system clearly has no way of fulfilling the request without direct reference to Shakespeare’s existing work, which it will source verbatim, cut up into chunks, and then sprinkle throughout its parody without care. Try anything distinctive — be it the Bible, Edgar Allen Poe, the Declaration of Independence, or Jane Austen — and you will find the same flaw.

This is all to say that, for now at least, the panic seems rather overblown to me. For basic tasks — rebooking flights and issuing refunds, for example — I can see AI being pretty useful. And as a parser of math and code and basic grammar, it’s invaluable. But for drafting meaningful human communication, comprehending and retaining complicated ideas, or replacing skilled workers, it seems much less so. Worry not, lads; you aren’t going to be replaced by a $30/month subscription and a copy of Google Chrome — yet.

At Spectro Agency, we understand that while AI is an invaluable tool, it requires

 human creativity and expertise to truly harness its potential. Our high-end digital marketing, app creation, AI-powered solutions, chatbots, software development, and website design services are designed to complement and amplify the capabilities of AI technologies, ensuring that our clients not only keep up but stay ahead in a rapidly evolving digital landscape. Discover how we can transform your business at SpectroAgency.com

Source: [National Review](https://www.nationalreview.com/2024/05/the-ai-panic-is-overblown/)