Is ChatGPT blocked or not by Italy? – 06/04/2023 – Tech

The Italian Data Protection Authority (Guarantee) inaugurates the interest of regulators in ChatGPT. Incredibly, no newspaper reported the news correctly, taking the version released by OpenAI, that the tool was blocked by the country, as the official version.

A careful look at what was decided by the Guarantor reveals that there was no ChatGPT blocking order, the decision being taken unilaterally by OpenAI.

The Italian regulator asked OpenAI to stop processing data from residents of Italy and gave the company time to demonstrate that data collection during the training of its AI models and during the use of the tool are in compliance with the European Regulation. on Data Protection.

This first case tells us a lot about what some of the main problems raised by generative Artificial Intelligence (AI) will be and what we can expect from the companies that are at the forefront of its development.

In the general excitement for the arrival of ChatGPT, it ended up neglecting that the models that support it, called large language models (LLMs), were developed outside the existing data protection laws in several countries, including Brazil.

These LLMs are trained from data collected on the internet, including personal data, as if it were a resource freely available for any use. In this sense, the Guarantor highlights several important points that should be read very carefully.

First, there was no transparency about the data usage on which ChatGPT is based. When personal data is involved, it is mandatory to transparently inform what data is processed and why, what is the legal basis for the treatment, etc. OpenAI has not sufficiently clarified these basic elements.

The Guarantor rightly observes that the processing of personal data is not a free activity. It is possible only if there is a legal basis. It would be unreasonable to assume that someone has expressed consent to the use of their data to train ChatGPT, as this is a new and, until recently, unimaginable use. There is no law providing for the obligation of such treatment, nor does OpenAI have a contractual agreement or a legitimate interest that justifies it.

Furthermore, the treatment is highly imprecise, despite appearing convincing. When you ask ChatGPT who Luca Belli is, the answer is a mixture of truths and falsehoods very similar to the truth – the so-called “hallucinations” of the AI ​​system. That is, data quality is neglected, despite presenting the answers persuasively.

The reason is simple. ChatGPT, like any AI based on linguistic models, acts as a statistical parrot, identifying which word should follow the precedents, but it does not take care to find out if such a result is true.

Considering that the Guarantor’s decision is based on normative foundations common to all data protection laws, it is natural to anticipate that this is only the first of a series of similar decisions.
OpenAI’s unilateral decision to block access to its service reinforces the service’s appearance of illegality. If it had been developed with minimum data governance standards, it would have been very simple to provide the information requested by the Guarantor.

Furthermore, the choice to block the service has the effect of arousing the indignation of Italian users, who will blame “the government” for ordering the blocking of ChatGPT – even though the Italian government never ordered such blocking.

ChatGPT, like most generative AI, has enormous potential and can bring enormous benefits, but it cannot be developed and distributed in total disregard of the law. While parliaments study the development of regulatory frameworks on AI, the law in force must be applied.

The only consensus that exists among experts is that new AIs do not need to be developed as quickly as possible, but in the most ethical and sustainable way possible.

OpenAI’s strategy is very reminiscent of the original motto of Mark Zuckerberg, founder of Facebook: “move fast and break things” [movimente-se rápido e quebre coisas]. It took breaking democratic processes in several countries to understand that this approach is anything but ethical or sustainable.