The UN warns that AI is advancing towards more autonomous systems with a “profound impact” | human rights | economy | Nacones Unidas

The document, a preliminary report that establishes a baseline for the current state of AI, warns that this technology has gone from “systems focused on predicting patterns” to “others capable of reasoning, planning and acting in different contexts,” which opens the door to “a new generation of digital agents with operational autonomy.”

LOOK: 66% of children perceive more cyberbullying: how to detect the warning signs

In this sense, the panel, made up of 40 experts and co-chaired by researcher Yoshua Bengio, considered one of the fathers of ‘deep learning’ for his advances in the field of artificial neural networks, maintains that the evolution of AI points towards “systems that will coordinate tasks, interact with real and virtual environments and be integrated into economic and productive processes.”

This change, the text says, will mean the emergence of a “digital workforce” with “implications that are still difficult to measure.”

The report highlights that investment in computing infrastructure has reached “levels comparable to large industrial projects,” with a strong concentration in the United States and in a small number of technology conglomerates.

This trend can reinforce “structural asymmetries” in access to artificial intelligence, the study says.

Evaluation and control challenges

One of the main risks identified by the panel, which is also co-chaired by Filipino journalist and Nobel Peace Prize winner Maria Ressa, is “the gap between capacity development and the capacity for evaluation and control.”

In that sense, it indicates that AI is advancing faster than the tools used to measure it and check whether it works well or safely: “Current systems are finding it increasingly difficult to evaluate models that evolve faster than the tools available to measure them, which complicates their supervision,” he indicates.

LOOK: Seven out of ten Peruvian advisors see it as risky to use AI to recommend investments

Furthermore, some models can already “learn to optimize their performance in the evaluation systems themselves”, which “reduces the reliability of the tests and forces us to rethink the control and audit systems.”

Uncertainty about productivity

On an economic level, the panel considers AI “a general-purpose technology”, with “transformative potential”, although it emphasizes that positive effects are not guaranteed.

“The benefits will depend on its adoption, the reorganization of work in companies and the investment in complementary capabilities,” says the report, which recognizes a “wide uncertainty” about its impact on productivity.

The estimates collected range from limited effects to economic “deep system transformation” scenarios, and emphasize that the key question is “who will capture the benefits of artificial intelligence and how they will be distributed.”

In terms of security, the report warns of the increase in capabilities in offensive cybersecurity, including the “automated identification of vulnerabilities in computer systems.”

LOOK: Mexico vs. England are in the round of 16: who will win according to the AI?

These capabilities can be used to reinforce defenses but also to “facilitate the exploitation of critical flaws in digital infrastructures,” he asserts.

Greater risk of misinformation

Likewise, the panel warns of the impact of AI on the information ecosystem, particularly due to the expansion of ‘deepfakes’ and other content created with that technology.

According to the document, these systems contribute to a “progressive erosion of the ability to distinguish between authentic and false information,” especially when combined with “personalized persuasion and recommendation algorithms.”

In the social sphere, the report highlights linguistic and cultural inequalities, since most models operate mainly in English and in a limited number of languages.

This generates a “systematic exclusion of thousands of languages ​​and communities.”

The document also warns of the growing use of AI in mental health, a field in which “adoption exceeds scientific evidence”, with risks of “emotional dependence, inadequate responses in critical contexts and lack of adequate clinical supervision” for users.

The panel concludes that global governance of AI remains “fragmented” and that current mechanisms are “insufficient,” calling for the implementation of “international standards for evaluation, auditing and supervision.”