Undoubtedly, generative AI , i.e. LLM models , transformers and tools based on them that allow you to talk to AI using natural language, are sexier today than all the rest of the AI algorithms put together. We can certainly talk about a breakthrough, even a revolution - and not only in the quality of algorithms, but also in the perception of the role of AI in society.

While so far, for the average citizen, AI was a curiosity from the IT world, today AI and ChatGPT are widely talked about all over the world, among others in the context of possible job loss .

But can we equate ChatGPT, or even the LLM concept, with AI as a whole?

Of course not. The various classes of algorithms that we collectively call AI have been in development for decades. For example:

1. OCR (Optical Character Recognition) programs have been using neural network-based algorithms to improve the recognition of blurry and distorted characters since the 1990s.
2. Since at least 2016, Roomba vacuum cleaners have been learning the layout of rooms in the home as they work to make subsequent cleanings more effective.
3. In 2017, the first deepfake videos based on GANs appeared.
4. Starting from iOS 10 (released in 2016), Apple has been offering the FaceID service, used to unlock the phone using your face, based on the Viola-Jones algorithm (i.e. so-called deep learning ).
5. Various banks and financial institutions use big data technologies and various classes of AI algorithms to better and better profile customers – in Poland, Alior Bank was the first to admit to this at the beginning of 2013, but the author of these words personally implemented customer credit scoring for one of the well-known auction portals in 2012.
6. The Nobel Prize in Chemistry in 2013 was awarded precisely for experiments with machine learning for the development of new drugs – the experiments themselves had of course been conducted for many years earlier and had already given promising scientific results in 2008. Currently, in China alone, at least several thousand scientists are involved in the development of AI for medical purposes only.

Could ChatGPT help or hurt AI development more?

There is no doubt that ChatGPT is a groundbreaking service from at least two points of view:

1. The average user, who finally does not have to be a scientist or a programmer to communicate with an algorithm using appropriately prepared data sets, but can simply use natural language – and not necessarily even English.
2. Ease of implementing this form of AI in dozens or even hundreds of additional industries – where there is no need to hire a data science team to analyze the company's data and produce appropriate interfaces – because data can be pasted into ChatGPT (in the simplest case) directly from Excel.

Unfortunately, ChatGPT also has a major flaw that is rarely forgiven when a given topic becomes politically controversial: immaturity , which in its case manifests itself in:

• incomplete knowledge ( until 2021 )
• very limited possibilities of replenishing it on an ongoing basis
• hallucinations
• censorship of many topics considered potentially dangerous or inconvenient by OpenAI
• the possibility of accidental disclosure of confidential data ( personal and otherwise )
• doubts regarding copyright of generated content

As a result, there is already growing talk about the low maturity of such solutions and the need to "do something with them" - similarly to how 15-20 years ago people talked about the immaturity of the mechanisms of cloning living organisms in the context of Dolly the sheep and concerns about human cloning. Paradoxically, the solutions that were introduced as a result make it impossible to continue legal work on improving this maturity.

In the case of ChatGPT, it may be even worse: while the entire concept of LLM models is indeed immature and simply expensive to maintain, other classes of algorithms, also commonly referred to as AI (in particular algorithms for the analysis and classification of various types of content) have already gained a relatively high degree of maturity, reliability and security of operation.

And the possible delegalization (called “regulation”) of AI, which is being talked about more and more openly, may affect not only ChatGPT or OpenAI, but also all other AI algorithms. And thus return many industries to the level of development of the 1990s.

What can and should we do as citizens?

First of all, avoid lumping different classes of algorithms into one bag called AI or artificial intelligence. What do algorithms for generating deepfake videos have in common with algorithms analyzing the credibility of borrowers, apart from the fact that both "underneath" perform very large numbers of simple operations on matrices and use Nvidia graphics systems for this?

After all, the exact same thing can be said about computer games. And we're not afraid of those anymore, are we?

Work

There has been a lot of talk in recent months about ChatGPT replacing some positions – and that is true, but unfortunately greatly exaggerated. Mostly out of fear.

The truth is that regardless of the industry and profession, various AI algorithms will be increasingly able to replace workers performing elementary tasks – but humans will still be needed :

• to verify and possibly correct the results of AI operations
• to manually paste these results into the production systems used by the company (and therefore have access wherever we are afraid to pass them directly to AI)

And above all, there will still be a need for specialists and managers who have the “big picture” in their heads of:

• what the company does, why and for whom
• for how much – and how it relates to cash flow, profitability and other financial indicators
• how individual departments and processes work – from general to specific
• what and how can be improved in the future
• how does this relate to the legal, technological and market environment
• what the competition is doing

All this means that in the next 10-15 years, in the most optimistic scenario, AI will take 75-85% of positions in some industries - but in many others only 30-50% . And the only people who should be afraid are those who are just entering the job market, or are simply bad at what they do.

benefits

Despite some specific threats, we should not forget about the benefits we are already gaining thanks to AI, for example:

• thanks to learning scoring algorithms, various financial services are more transparent and honest – it is much more difficult to, for example, obtain a loan by fraud
• dozens of new cancer drugs created solely thanks to AI are already in clinical trials
• hundreds of different services that we use every day work much more efficiently thanks to AI – for example, we can unlock our iPhone with our face just by looking at it, the Roomba vacuum cleaner efficiently cleans our house, etc.

What can we do as employees?

First of all, realize that personal high performance and good fit for a given job may be enough for the next few years – but in the long run, everyone may become an underperformer and simply a candidate for replacement by AI. Efficiency is no longer an option.

As an employee, you should focus on identifying areas that cannot be delegated to AI in your industry , e.g. due to primitive fear, lack of trust , ethical issues , or simply legal regulations (GDPR or industry regulations). These will primarily be areas related to:

• having key and possibly extensive knowledge about the company's operations
• having access to various systems used by the company – primarily production systems (remember, however, that the systems themselves can and will change, so the key is not so much having specific accesses, but fulfilling the organizational roles that these accesses are associated with )

Simply put: the larger the competence silos you create, the harder and longer it will be for AI to replace you , and other employees as well. On the other hand, if you create such a silo from the bottom up, as an ordinary employee, there is a big risk that managers will want to make it difficult for you. Silos are a solution that is beneficial for individual employees (especially the weaker ones), but disadvantageous for the company as a whole.

That is why you should be discreet in this respect – any form of boasting that you have an “AI idea” will most often be counterproductive, because you may become a candidate for quick replacement not so much by AI, but by employees more loyal to the company. Instead, you should:

• present the attitude of a helpful colleague on a daily basis
• on various occasions, suggest places, processes and positions where work can be more formalized, and then introduce ChatGPT there – it is important that they are selected in such a way that it does not threaten your own work
• try to be invited to various company knowledge exchange events – and try to actually exchange this knowledge to understand what is happening in the company
• where possible, offer to conduct presentations about your department and its unique role in the company – focus not on operational work, but on showing those aspects of work that “accidentally” are the most difficult for AI to take over

What else can we do as politicians?

The most important thing is not to get carried away by emotions and the temptations of seemingly easy solutions proposed by the left and populists.

3 levels

More specifically, when talking about AI-related topics, you should clearly distinguish between 3 further levels of generality:

1. AI as a whole – i.e. the general concept that algorithms can learn in some way.
2. Generative AI – i.e. LLM models, transformers and other inventions from the field of deep learning, the aim of which is to be able to talk to AI in natural language.
3. Specific companies and their services – e.g. OpenAI and ChatGPT.

In particular, if we are talking about the potentially harmful impact of ChatGPT on employment, let's remember that the potential problem is most often related to specific companies and services , and in the worst case, to generative AI - and it should not be extended more broadly than necessary. Because "all AI" also includes dozens of other directions, which, acting without thinking, we can inadvertently "throw out with the bathwater".

job security or development?

China has been at the forefront of the development of various AI algorithms for many years, followed by the US:

For Polish politicians, this should mean exactly that they can help Polish companies specializing in AI technologies – or they can not help them. But in both cases, this has absolutely no impact on global trends and will not in any way stop, for example, the replacement of jobs by ChatGPT.

At most, companies from outside Poland will earn from this.

Therefore, as a politician, you should first of all find out what hinders the development of Polish AI companies and what is worth changing in Polish law to help them build services that can compete on an equal footing with companies and services from the US.