Generative AI, technological progress... an end in itself?
So-called generative Artificial Intelligence (AI) (a sub-category of AI that uses machine learning to generate new data that resembles existing data, to put it simply) continues to dominate the headlines, the most widely publicized being Chat-GPT. Between blissful blindness and panicked fear, all the experts are showering us with their opinions. For ordinary people, however, this technological breakthrough seems remote and abstract, and yet…
Why is Chat-GPT such a hot topic? There are many answers to this question, stemming from Open-AI’s particular approach and the overall progress observed in the general functioning of AIs. The increase in computing capacity of AI chips, and the immense volume of data on the Internet, means that learning is now possible on an unprecedented scale, even if there are risks concerning the quality of the data sets made available for learning.
Although we’ve been talking about AI for several decades now, we have to admit that the first attempts were rather approximate. And, yes, without necessarily realizing it, you’ve all tried your hand at using AI at one time or another if you’re big consumers of smartphones, tablets or computers. What applications? These are message writing aids (sms or messaging) on your smartphones, whether on the Android (Google) or IOS (Apple) side, and also found in cloud mails like gmail (Google) and many others. Admittedly, performance was rather sketchy… Then there were the “Ok Google” or “Ok Siri” assistants, or even “Alexia” (Amazon). Here again, while progress was observable, there was a lot of questioning about the relevance of these assistants and the way in which publishers improved them (using recordings of your conversations, etc.).
How are Chat-gpt and the other AIs developed using OpenAI technology bricks changing the game? It’s a language model known as “Generative Pretraining Transformer”, hence the initials. At the heart of this model’s architecture is a structure called the “Transformer”, a neural network that uses a particular method called “attention” to determine which words in the input are the most relevant (statistically) for predicting each subsequent word. Chat-GPT uses a variation of this model, specifically designed for text generation. Pre-training has been carried out on a database extracted from the Internet (until September 2021 for the free version).
The gas pedal for OpenAI was to open up their tool to all Internet users wishing to test it. This openness led to a drastic improvement in the model, thanks to contributions from all users, and also enabled OpenAI to establish a strong reputation in relation to all its competitors. However, it should be noted that the estimated cost of daily operational maintenance is of the order of $700,000/day, which would mean a cost of 36 cents per request.…
We were at Chat-GPT 3.5 at the end of the year and are now at version 4.5. A number of plug-ins are available, enabling users (with a subscription) to browse the current Internet and thus avoid being confined to training data, which was limited until September 2021. It is expected that the increase in computing capacity combined with the use of the software will enable great progress to be made, although it should not be forgotten that this is word prediction, so the coherence and relevance of the answers provided remain to be verified and should not be taken at face value.
Apart from the fun aspect of Chat-GPT, what upheavals can we expect to see with the arrival of these generative AIs, whether we like it or not? The potential for productivity and efficiency gains in the “intellectual” professions is enormous. There are already sectors, such as the (German) press, where the shift has been made, with a reduction in the number of freelancers. Fewer are needed, as AI replaces the departures, often with impeccable spelling and grammar. Fields such as marketing, finance, healthcare, justice and I’m sure many others will see their professions and resource requirements drastically altered, and these changes are likely to be brutal and happen much faster than we imagine…
Many believe that we are on the cusp of a new revolution, similar to that of the 80s and 90s, which saw robotics completely change the nature of manual, tedious and repetitive jobs, but this time it will affect white-collar workers in particular…
That’s what we’re likely to see in the short term (3 to 5 years). And to those who imagine that many manual jobs will be preserved, I have my doubts. Indeed, while the focus is on generative AI, we mustn’t forget that robotics is also making constant progress, as demonstrated by the Boston Dynamics videos that can be found on the company’s YouTube channel. Imagine an alliance between robotics, which already use AI building blocks, and the power of generative AI such as Chat-GPT? (And let’s not forget all the other sectors in which automation is becoming more and more advanced: autonomous vehicles, drones, delivery robots and so on.
In fact, if we ask Chat-gpt about the impact of the widespread use of generative AI, he suggests several possible scenarios. The first is the advent of a “post-work” society, where the massive automation of jobs would push people to spend most of their time on leisure and creative activities (with Chat GPT? Midjourney?). This underpins the need to introduce a universal income… But this poses several problems: what about self-fulfillment through work? Or what would be the unspoken corollary of this situation? Given the current economic model, if it is to survive, wouldn’t the majority of the population be confined to the restricted and unique role of super-consumer… This consumption is necessary for companies to generate sales, grow and continue to innovate, leading to an obligation to consume for the beneficiary population in order to avoid the collapse of the model…
The other vision put forward by Chat gpt isn’t necessarily any brighter: we’d end up with a two-tier society between those who have the skills to work with AIs through their ability to interact skilfully with them to get the best out of them, and the others, left behind with an exponential increase in inequalities.
The last scenario would involve continuous education and training for the entire population. This would be the norm, enabling people to maintain their skills and remain relevant in the job market, but do we all have the capacity and energy to train continuously? Is there not a risk of a multi-speed society here too?
Let’s not talk about the consequences for our ability to learn, when knowledge will be available to us simply by asking. Will we still feel like learning? Can we be satisfied with knowing, without doing?
As you can see, this technological revolution, which is coming at us at the speed of a galloping horse, should inspire us with healthy concern, so that we can discern all the risks and make the most of them for our common good. Unfortunately, it seems that thinking on these subjects is only left to specialists and engineers, perhaps intoxicated by innovation as innovation, but unable or unwilling to see the dangers that lie ahead. What’s more, in our country, we can only regret the lack of interest shown by our political forces in tackling these issues head-on, and in particular the necessary transformation of our education system in the face of these new challenges…
A final remark on this subject: given the various elements discussed in this text, how can we think about this technological revolution in a world of finite resources, where energy will eventually no longer be abundant and cheap? Can a hyper-technological civilization last? What are the implications for mankind, faced with ever-increasing dependence on technology? Is it our destiny to become a mere interchangeable cog in a large technical machine where nature will no longer have a say?