Is that what you want? Actually, what you want may not matter, you’ll get it anyway.
It may come as a surprise, but AI has been predicting the future for a long time – and very successfully. To the extent that we’ve been totally addicted for ages. Oh, we haven’t had AI for ages? Well, we used to call it something else. What’s new is the speed, accuracy, availability and not the least: The areas of application. Would you like to know about your next car accident? Where and when the next cyber attack is coming? Click here …
I’m not joking. Your next car accident may actually be predictable. But before we get into that, let’s take a step back: How is it that AI has been predicting the future for a long time? Like I said, we’ve been calling it something else. It used to be ‘models’. Economists have been using them since the birth of computers. Meteorologists for even longer – and delivered ever more impressive (reliable) weather forecasts.
But that’s not AI? Actually it is, if we use today’s definition (which I don’t agree with BTW, but here we are). What the meteorologists have been doing for more than a hundred years, is creating models. The models take in to account experience, statistics, all kinds of relevant factors, observations, measurements, trends from last week and the month before. Comparing, calculating, iterating etc.
If that sound like algorithms to you it’s because it is. The meteorologist’s models have improved dramatically over the years, in particular since they got computers, and even more when they got satellites and radars and all kinds of new data sources. They have become increasinly good at predicting the future. The near future (the afternoon weather, maybe tomorrow’s weather) and the outlook for the month or the year. Just look at the newsbeat about the upcoming El Niño). Or closer to home, the weather app on your smart-whatever. I’m catching myself using the local weather app at least 4 times daily. And while it may not always be exact, it’s impressively close most of the time – even when looking a week ahead. If that isn’t predicting the future, what is?
Weather models – or meteorological models – are just that, models. Digital models consuming vast amounts of data, generating useful results. Being improved, expanded etc. if not daily at least monthly. This is beginning to sound like chatbots, isn’t it? Because what are chatbots? They are models, more specifically Large Language Models (LLMs). Different from but with a lot in common with weather models. They are modeling something for a purpose with a set of inputs and outputs – which is different from one to the next, but the concept is the same, many of the ingredients are the same.
Thus it shouldn’t surprise anyone that bots of various types (I hate the term AI for the simple reason there is NO intelligence whatsoever in these models, just smart algorithms and lots of data – impressive and useful but not intelligent) can predict the future. And – as the recently acquired knowledge about large models of many types has demonstrated, we’re about to see some extremely interesting results – in new areas and disciplines.
Such as in cyber security, where the company behind bfore.ai claim to be able to predict the next attack on your infrastructure. Seriously. Or british hitnmix.com, which doesn’t predict per some timeline, but analyses music and its timelines to the extent that it can remove single instruments from a piece without affecting the rest. The effectiveness and quality at which the tool is doing this, blew me completely off. A extremely visual example of what digital processing plus machine learning can do when the processing power and the (learning) data are adequate (as in ‘huge’). If you’ve been practicing music at a just about any level above normal listening, maybe even used analog and/or (traditional) digital filtering, this is going to blow your mind. By the way, the instrument removal and remixing are just two of many tricks this tool has up its sleeve. Check it out – you have been warned: it’s addictive.
My third example is – believe it or not, ChatGPT. An unlikely candidate because I consider ChatGPT to be stupid (see Chatbots are lying, now what?). But stupid doesn’t mean useless. Actually, ChatGPT is quite impressive in many areas. The problem with chatbots is that they’re being ‘sold’ (or thought of) as something very different from what they are. They’re models, LLMs to be specific. If you know the data being used and know how to use them for whatever your goal is, they’re great. Even fantastic.
Over the past few months – and most recently in a local newspaper – I’ve read about investors and stock brokers using ChatGPT for hours every day to analyze trends in the stock market. Their job is to predict the future and the new tools are saving time and increasing accuracy. Interesting, isn’t it? These tools (or rather, models) – Bard, Ernie, Bing, ChatGPT etc. – have been proven inaccurate (or worse) for all kinds of purposes. Why do they seem to work well in one setting and not in another?
Of course it’s the data. And the modeI. And the tuning. But even more importantly in these cases, it’s the user. The users have a specific (usually narrow) purpose to which they are applying the tool, they have experience, they have expectations and are able to evaluate the results. They spends the time necessary to hone the questioning (aka ‘learn the tool’), and – eventually – they get the great results. They are literally predicting the future.
Most of us haven’t thought about it that way before – maybe because all these ‘prediction machines’ have seemed to be a combination of guesswork and science or math. They still are, but the balance is shifting. More data, more power, better models, smarter users, and we have real predictive power.
The security scenario I referred to above is possibly the best example of all right now, maybe because it has so much in common with weather forecasting. Think about it (and I’m sure you can relate to this even if you don’t have a clue about cyber security), how different would it be if you could prepare for the next attack with specificity instead og general defences? Like how, where, intensity, tools/weapons etc. Just like we watch the weather forecast in the evening and dress up for the bad weather (or the sunshine) in the morning.
Wow. It it possible? My first reaction – and probably yours – was ‘forget it, it’s a pipe dream’. But the more I think about it, the more sane it gets, eventually ending up as ‘of course, now it makes sense’ – with a few caveats. Like the amount and freshness of the data. But definitely possible. And again, that’s what bfore.ai claims and does.
Fascinating indeed. Can we do the same in say, medicine? Politics? Etc.? Look, researchers are predicting the next El Niño as we speak with incredible specificity (after the fact we’ll call it ‘accuracy’). A phenomenon previously thought of as unpredictable. It is tempting – tongue in cheek – to ponder whether politics (seemingly completely unpredictable these days), can be modeled and predicted…
Back to reality: If you made it this far, I’m sure your head is spinning. What about ….???? Now, here’s a serious challenge: We already have models and data predicting the future of the world with significant accuracy. Using enormous datasets from real time sensors and hundreds of years of observations and recordings. They paint detailed pictures of our world 5-10-20-50 years from now. It’s a grim picture to say the least. We – most of humanity – don’t like it. But – given the established reliability of the models and the data, it would make sense if we were rushing to fix it. Change the future, literally ensure survival.
Yet we don’t. We ignore it, pretend it’s not true, call it speculation. I call it the ‘don’t look up’ syndrome. It’s a choice, just like trusting (and acting on) the weather forecast and the stock broker is a choice.
‘Is that what you want’ I said in the heading about our new and growing ability to predict the future. The fact is – it doesn’t matter what you want. You’ll get it anyway.
Get ready for it. Turn it into an opportunity. Otherwise it will automatically become a threat.