If I tell you that about 2 years ago, GenAI arrived with ChatGPT and that now we almost only hear about Artificial Intelligence, in every way, I probably don't tell you much.

I would therefore like, in this article, to bring clarity to the subject, by summarizing the implications of AI today, what are the current issues that we are facing, but also, the concrete and useful applications from which we can benefit.

An abstract definition

AI (Articial Intelligence) is a field of computer science that focuses on creating systems capable of performing tasks that typically require human intelligence. These tasks include learning from experience, recognizing patterns, solving problems, understanding natural language and making decisions.

Pay attention to one thing, it is important to realize that this is a very generic expression and that it is nothing more or less about computer, algorithmic techniques which are actually used, far from replacing the functioning of a human brain.

In other words, what we call “AI” depends entirely on our perception of it, and should rather be called “algorithmic techniques”. So much so that Luc Julia, who created Siri, came to say that “There is no such thing as AI”.

A technical definition

Systems commonly associated to AI are designed to learn from data, adapt to new inputs, and improve over time. This key components and subfields include :

• Machine Learning (ML): A subset of AI that involves training algorithms to recognize patterns and make decisions based on data. ML models are built using statistical techniques to enable systems to improve their performance on a given task with more data over time.
• Deep Learning (DL): A specialized branch of ML that employs neural networks with many layers (hence "deep") to model complex patterns in large datasets. DL is particularly effective in tasks such as image and speech recognition.
• Large Language Models (LLMs): These are deep learning models that are trained on vast amounts of text data to understand and generate human language. Examples include GPT (Generative Pre-trained Transformer) and BERT (Bidirectional Encoder Representations from Transformers).
• Generative AI (GenAI): A type of AI focused on creating new content, such as text, images, or music, that is indistinguishable from human-generated content. This includes techniques like Generative Adversarial Networks (GANs) and autoregressive models like GPT.
• Retrieval-Augmented Generation (RAG): A hybrid approach combining retrieval-based and generative techniques. RAG systems first retrieve relevant documents or information from a large corpus and then generate responses based on the retrieved content. This approach enhances the accuracy and relevance of generated outputs.

In essence, AI leverages these advanced techniques to build systems that can perform sophisticated tasks, ranging from data analysis and prediction to natural language understanding and content generation.

Two things should be noted: on the one hand, these are only statistical and analysis tools, which go through large sets of data, very far from the functioning of a human brain. The other thing is that, for the end user, what is important is still the perception, to qualify as AI, so these techniques will not necessarily be used in the product that is sold as such.

Just another buzzword ?

It has become the latest hype in IT, GenAI has propelled AI onto the podium, as have blockchain, VR headsets, or even 3D TVs, which, as we know, left the front of the stage as they came.

It is also good to remember that AI is a subject as old as computing itself, periods of craze followed by disinterest, called "AI winters", have been regularly observed, the first dating from 1966.

The hype is such that in recent times, the term "AI" has been attached to most computer products, whether it has any real meaning or not, sometimes in a truly ridiculous or inappropriate way. Each application wants its "AI feature".

The same goes for computer conferences that were twisted into being on the topic of AI, even though the real topics were completely different.

One truth remains, this enthusiasm unlocks research funding, which allows us to believe that we will be able to see real progress emerge in the future.

A threat ?

I don't know if you've noticed, but our thoughts often move much faster than our ability to act.

We are indeed witnessing progress currently, as well as an amplifying marketing and journalistic frenzy, of which we must not be fooled.

I'm not going to dwell too much on the subject, but we can discern worrying scenarios, which are the loss of jobs and loss of control.

Loss of jobs

A widely talked about study from the IMF indicates that 40% of jobs would be affected by AI. These predictions should be taken with a grain of salt for several reasons.

On the one hand, certain jobs were assigned impressively quickly, such as graphic designers or developers. These professions generally know how to adapt to new technological developments and take advantage of them to gain productivity, we cannot therefore talk about job losses. Besides, in the next section, I will talk about the tools that I use daily as a developer.

On the other hand, it must be taken into account that the development of new working methods, as all industrial revolutions have demonstrated, have not led all of humanity to idleness, but have contributed to the creation of new jobs, whether qualified or not. Even if unemployment issues are very real, they are also nuanced and contextual.

We must not hide our faces either, we will observe changes due to paradigm shifts induced by AI, such as recently the elimination of jobs in Duolingo's translator teams, the deterioration of the DeviantArt website, AI Crawlers, or many other examples.

What is certain is that jobs as we know them will evolve and the skills needed will be different, which concerns current and especially future generations.

Loss of control

AI may take control of humans, betray them or even destroy them, that's what Skynet, HAL 9000, I, Robot, or globally Hollywood and Pop Culture told us. We can also call it the "Hollywood AI".

These are only extremely unlikely scenarios. For this, the machine would have to truly emancipate itself from human command, have sufficient power of action, have a motive for such actions and above all work completely differently than the tools we develop today.

For the moment, all that's happening are clickbait publications, such as, for example, that Google AIs are plotting among themselves in an unknown language, or that a tool is working to destroy humanity.

It is important to be alerted to the threats that these new tools may have (we will talk about them again, particularly on privacy and ecology), but we must also put things into perspective and carefully check the sources of the informations we consume, to demystify things and understand them.

Individual usage

The opening of access to the general public to ChatGPT at the beginning of 2023 was successful and its popularization was impressive.

The general public very quickly understood the services that it would provide them:

• Help with content generation (homework, articles, creations of all kinds)
• Search assistance, which replaces the Google engine
• Help with problem solving

This use has become a new common model, for example, for this search, it is simpler to ask ChatGPT to give me the solution, rather than searching from page to page to find my answer (and that's very complete) :

Google, which has long been at the forefront in the field of AI, suddenly found itself shaken up by the acquisition of ChatGPT by Microsoft. Google therefore finds itself in the situation of being confronted on the front line by these GenAI tools which aim to replace the classic Google search page, and it is not in a strong position.

Now, Google offers Gemini (its ChatGPT competitor) on its home page and Bing offers Copilot (which uses the ChatGPT engine):

Many tools on GenAI have emerged recently and are referenced on the futuretools site.

The next developments revolve around improving the engine: the completeness of its dataset, the speed of execution, as well as the recent update of the engine training and its internet access.

The other development is to be multimodal: this affects images, music and even video.

Sora, OpenAI's tool for generating video, unveiled a fairly impressive video in Japan, even if some imperfections were detected.

These tools are very practical and impressive, but require human validation (or even editing) and require new skills to generate the most effective prompt possible.

Like any tool, it can also be misused, which can lead to the deterioration of some content available online. It is up to us to also be vigilant about the quality of the content we consume.

Industry usage

For developers, Microsoft has been offering Github Copilot for several years now, which allows assistance with writing code (whether in autocomplete or via a prompt).

GenAI is also very useful for debugging phases, because it gives food for thought, hence the reaction of Stack Overflow), which also offered an AI service.

For Cloud architecture usage, today the challenge is no longer to build your own Machine Learning or LLM models, given that they are available off the shelf by Cloud providers.

Each provider therefore offers a range of services to support the use of these technologies.

For the moment Google and Microsoft are in the lead, AWS also offers a service called Bedrock that can use different Foundation Models, including Mistral AI.

AWS also offers Amazon Q, a tool more focused on the AIOps approach, which aims to integrate AI into the operational part.

It’s interesting to see that AWS, which has always been at the forefront of the market, seems to be behind its competitors on GenAI. We don't know how it will evolve, but they could surprise us, they are investing actively in this domain.

At a time when we are talking about sovereign cloud and GDPR in Europe, GenAI represents an additional constraint in terms of data protection and will probably not be validated in all contexts (because usage means giving your data to the cloud provider that provides the service).

Despite everything, GAFAM are at the forefront but today a lot of brands are trying to associate their image as being at the cutting edge of AI, this extends to automobile or even cycling industry.

Limitations

The technology has developed well, but creates new problems inherent in the way it works and requires increased supervision:

• Hallucinations: To answer correctly, the model can very well invent informations from scratch with breathtaking confidence. This is why it is essential to scrupulously recheck everything before validating and ensuring that you have sufficient expertise for the area being addressed.
• Bias: For example in image generation, you type: “Show me a CEO”: the generator will only output images of tall brown men with white skin. These are clichés that can become even more ingrained in our way of thinking, instead of encouraging diversity.
• Reasoning problems: The model does not have our way of thinking and is capable of making grotesque errors, certain examples demonstrate this: “Making a sentence without the letter e", “Give a positive prime number after 2”, etc. The engine returns logical errors with the greatest confidence.

These problems are however corrected with each version, but will never be corrected completely.

We must also not forget the problems already inherent in computing, which are greatly amplified by AI models and constitute, more than anything else, the real problems caused by AI:

• Privacy: Data models recover a large amount of information, of which we do not necessarily have the capacity to verify whether they are respectful of privacy or not, we have to trust the software provider. There is a particular topic with Meta about the use of personal information to train its models.
• Ecology: Climate issues are more than ever at the forefront, at a time when we could instead advocate a form of energy saving, AI models consume a lot of resources: even though this information is kept private, we can clearly imagine (confirmed among others by a Sam Altman declaration) that this represents astronomical quantities of GPUs running constantly (which makes Nvidia happy) and that it’s only growing, we can unfortunately no longer count on Moore's law to compensate.

Regulation

Failing to be at the forefront in the field of creation, Europe has taken the step of being at the forefront in the field of regulation with the AI Act. This is nevertheless a major advantage, because the problems posed are very real and complex in the regulation of such tools.

Related topics

AI alone is not enough in itself, and makes it possible to develop other IT sectors such as the cloud, data, chatbots or even robotics, which are very related subjects. For the years to come we can imagine that combined progress in these different sectors could once again change the situation in the technological field.

To conclude

The IT market has found itself extremely shaken up recently: all software wants to release its AI feature, IT specialists suddenly have to be specialists in AI and the big players are investing colossal sums to corner the market.

In this context, it is interesting to adapt to take advantage of these technologies, to always marvel at the progress and the help it brings us, but also to be aware of the issues (particularly on the privacy and ecology side), hence the importance of regulation.

Keep in mind that the use of these technologies requires reinforced supervision in order to always produce quality work.

The big development that makes many people dream, but which is probably not for tomorrow, contrary to what Elon Musk said, is the AGI (Artificial General Intelligence), maybe that will be the next breakthrough, or maybe we will just get tired of AI, saturated from having talked about it too much, until the next real update.

In any case, it is important to follow the news with full knowledge of the facts, by selecting your sources carefully and by following people who really practice AI and can talk about it without mystification, as do very well Yann Le Cun, Aurélie Jean and Luc Julia.

More progress is still to come and it is very likely that I will write other articles on these algorithms which work on data sets, which some have decided to call “AI”.