AI ACT e going dark: secondo troppi un modo per aggirare i principi della privacy

Premessa. Ovvia. Ma va fatta. Nel 1989, anno piu' anno meno, dicevo che qualcosiasi cosa si fa a computer, resta per sempre a beneficio delle indagini serie. Il computer e' una opportunità per indagare.

Negli anni successivi ho visto incapacità di indagare (distruggendo materiali) e crescere (migliorando di professionalità). Oggi invece la direzione e' quella della raccolta di massa di tutto per poi dimostrare la colpevolezza di qualcuno anche a distanza di anni, anche fuori dal contesto, indebolendo ogni strumento di indagini difensive, e presupponendo dati corretti pur dopo manipolazioni, estrazioni e altro.

Le indagini devono essere fatte. Si devono poter fare. Con la supervisione di un magistrato. Quello che non e' consentito e' la sorveglianza di massa che precede e segue il reato. Non e' costituzionale. Punto.

Le indagini sul territorio (non sui singoli) sono un esempio virtuosodi indagini di polizia, con metodo tutto italiano. Ho già parlato dell'ebook "sicurezza 4p" che spiega come in 30 anni si e' arrivati a prevenire i reati senza trattare dati personali.

Peccato che nessuno voglia leggerselo. L'intelligenza artificiale e la conoscenza del territorio e' un tema. Sorvegliare le persone per sapere cosa fanno singolarmente, no. Le posizioni sull'anonimizzazione dei dati tuttavia non aiutano, preferendo un approccio ideologico ad uno concreto. Tutto questo porta l'Europol a pretendere di non avere limiti, e le DPA ancora in attesa,

Going dark e' lo slogan usato dalle polizie per indicare la cecità ai reati.

Sembra di essere finiti in pieno complottisimo, ma con fatti elencati e documentati fin nel dettaglio e denunciati da Amnesty e da tanti altri autorevoli da tempo.

La sorveglianza di massa e' ampiamente sponsorizzata da Governi e forze di polizia che dimenticano che il fatto stesso di usare i computer e' un aiuto incredibile alle indagini caso per caso.

Quello che non si dice e' che si vuole usare i dati delle attività online (ritenuti affidabili acriticamente e a prioristicamente) per le indagini. Voglio sapere i cellulari dove si trovano per indagare chi era sul luogo del reato.

Tutto in nome di proteggere la sicurezza e solo per i reati minori. Naturalmente questo prima che vengano commessi, e non e' un dettaglio come vorrebbero far passare.

Vi cito le fonti e vi imploro di dirmi che mi sto sbagliando e si stanno sbagliando tutti.

C'e' qualcosa di marcio che porta ad usare i social per evidenziare contenuti a favore della sorveglianza, dopo avere favorito gli estremisti.

Buona lettura, lenta. Fa male allo stomaco.

• Algorithmwatch sui retroscena dell'ai act e i "consiglieri"
• netzpolitik.org La presidenza svedese fortemente determinata a sorvegliare di massa gli europei
• Amnesty e il suo comunicato
• Joint Declaration of the European Police Chiefs - Europol
• 2022 Joint Declaration of the European Police Chiefs on the AI Act
• Consilium Artificial intelligence act: Council and Parliament strike a deal...
• Europarl Artificial Intelligence Act - European Parliament
• La Commissione: Commission welcomes political agreement on Artificial...
• Proposte di Europol per ChatGPT - Europol
• thread su linkedin
• Queste le linee guida sulla sicurezza nello sviluppo di AI

Questo il comunicato di Europol del maggio 2022

Joint Declaration of the European Police Chiefs
as approved by the European Police Chiefs
during their informal meeting in Berlin on 24 May 2022
On 21 April 2021, the European Commission published a draft regulation designed to govern artificial
intelligence (Artificial Intelligence Act - "AI Act").

We, the European Police Chiefs, generally welcome the Commission's initiative for regulation, because
rarely has a new technology been associated with as many opportunities and risks as artificial
intelligence. However, in line with the objectives of the draft regulation, we would like to emphasise
that the rules formulated could seriously affect police work and address the following particularly
critical aspects:

1) AI is absolutely essential for police work
Given the fact that the number of crimes committed in digital space is increasing and that these
offences already involve significant quantities of data - in the multi-digit terabyte range in individual
investigations - the use of AI-supported tools and machine learning is indispensable for the successful
fight against crime and the enforcement of the law. AI is an essential part of a high number of digital
forensics and analysis tools, including commercial ones, which are currently used by law enforcement
authorities to analyse ever-increasing volumes of data and extract digital evidence.

Only through the
use of AI will it be possible to respond to the ever-increasing data quantities to be analysed in an
effective and efficient manner. Without the help of advanced technologies, we will not be able to stay
abreast of the progressing digital transformation, to adequately analyse the rapidly increasing data
volumes - frequently under particular time pressure - and to initiate necessary police measures to
enforce the law and avert danger. For law enforcement to bring criminals to justice, mitigate crimes
and adequately protect victims in the digital age, the use of AI-supported tools is not a choice but a
necessity.

2) No disclosure of particularly sensitive data
The discharge of a significant proportion of police tasks involves the processing of particularly sensitive
data, the disclosure of which is subject to certain restrictions and cannot be authorised on the basis of
the decision-making powers of the police authorities alone.
Therefore, when rules regarding access to test and training data by third parties are laid down, it must
be ensured that adequate consideration is given to the sensitivity of police data, for instance by excluding
unrestricted remote access.

3) No blanket classification as high-risk
The risk-based approach advocated by the European Commission, which aims at counteracting the
unregulated and uncontrolled development and application of this technology, is plausible. This must
not, however, lead to a situation where entire areas of application affecting the security authorities are
generally classified as high-risk and thus subjected to significant restrictions by default.

The use of a
fingerprint quality check algorithm created by machine learning, for instance, must not result in a
blanket high-risk classification of the entire process. In fact, most of the AI applications used for law
enforcement work in the European Union do not pose a high risk of harm to the health and safety and
the fundamental rights and freedom of persons.

This is because, AI applications such as speech and text recognition (text classification, machine
translation), information extraction (named entity recognition), object detection or image classification, are
aimed at automating data pre-processing and processing tasks, so as to relieve human analysts from
repetitive tasks and from exposure to gruesome materials, and allow them to focus on more cognitive
tasks. These AI systems are used to support the work of criminal analysts faced with a dramatic increase
in the number and size of structured and unstructured datasets, but they do not involve automatic decision-making.

The rules need to enable the specific evaluation on a case-by-case basis and the Assessment of risk in the
concrete individual case. In this context, it has to be borne in mind that the systems employed by security
authorities are exclusively used in line with the relevant police powers and legal provisions and that they
are already subject to rigorous control. These systems do not replace but only support investigative
activities. The results achieved by the use of AI are always evaluated and checked by humans, i.e.
specially trained and skilled police officers. Moreover, many AI-supported tools are applied to data that has
been seized in the context of a criminal investigation, meaning there is no real-time dimension or
indiscriminate analysis of data.

For this reason, we wish to formulate the following expectations of the future regulation of AI:

•  The regulation must provide for exceptions applying to law enforcement authorities, which take sufficient account of the peculiarities and specific requirements of police tasks within the already existing legal framework as well as the realities of investigations in the digital age.
•  The regulation must not lead to a situation where the police cannot use AI at all or only with considerable effort or delay. Therefore, suitable expedited procedures are needed to ensure that police tasks are performed effectively and without delay while ensuring the necessary checks and balances.
•  The regulation must include relevant terms, definitions and phrases (e.g. mandatory, transparent, plausible, explicable), clearly define them in a legally reliable manner and must not be in conflict with existing tasks of the security authorities.
•  The regulation and the definition of AI systems it contains must not lead to a situation where already established IT procedures, which thus far have not been generally considered as AI, fall within the scope of the regulation and may therefore no longer be used.
•  The regulation must enable the concrete evaluation and Assessment of risks in specific cases and must not make general provisions for a blanket classification of police systems involving AI as high-risk.
•  The regulation must not contradict already established legislation with relevance for the police, particularly the existing data protection and data processing provisions, or result in a collision of pertinent legal norms.
•  The regulation should also recognise the criminal use of AI and the need to empower law enforcement to investigate those “cyber-enabled” crimes.

We as the police welcome the European Commission's initiative to regulate AI.

However, in order for
the police to continue to effectively carry out the tasks assigned to it by law in the areas of crime control
and law enforcement to offer protection to and provide security for all citizens living in the EU, it is
essential that the rules to be adopted take into account the specific objectives and existing legal control
mechanisms regarding the use of AI by security authorities.

At the same time, separate case-by-case
assessments or exceptions should be provided for when it comes to these systems and cases of
application. This is the only way to maintain the security authorities' capacity to act in the future in a
digital environment which is characterised by the ever-increasing volumes of information including
personal data.

Berlin, May 24th 2022