From Prompt to Reality: The Controversy Surrounding AI

Reporter: Adda Avendaño / Photographers: Israel Vera, Javier González, and Jorge Aguilar

IPN specialists address growing public concerns over the evolution of Artificial Intelligence, agreeing that faced with a technology capable of responding in seconds, the ability to question remains a uniquely human trait.

KEY TAKEAWAY

One of AI's primary pitfalls occurs when a system delivers incorrect information with flawless grammar and a persuasive tone—a phenomenon known as an "AI hallucination," notes Dr. Magdalena Saldaña.

A few years ago, Artificial Intelligence (AI) belonged primarily to the realm of science fiction. Today, it powers our smartphones, curates playlist recommendations, assists in medical diagnoses, optimizes industrial manufacturing, generates imagery and text, analyzes massive datasets, and drives autonomous vehicles.

Yet its everyday presence has also fueled widespread anxieties that do not always align with technical reality: Does AI truly think? Can it achieve consciousness? Will it eliminate millions of jobs? Is it taking control of our lives? And can we fully trust its answers?

AI Is Far from Dominating Humanity

For Dr. Juan Humberto Sossa Azuela, Director of IPN's Computing Research Center (Centro de Investigación en Computación, CIC), AI can be defined as the ability of machines to mimic intelligent behaviors observed in living organisms—particularly humans—in order to execute routine tasks.

However, he cautions against confusing performance with comprehension: "We cannot claim that the machine understands us. It is simply capable of predicting with high precision what sequence comes next; in short, it is a high-performance prediction machine and nothing more."

According to the Emeritus Scientist of Mexico's National System of Researchers (SNII), claiming machines will develop consciousness in the near future is scientifically unfounded.

"Earth is approximately 4.5 billion years old; life emerged around 600 million years ago, human consciousness developed roughly 200,000 years ago, and modern humans appeared 10,000 years ago. We should not expect a machine to experience consciousness in less than a century," noted Dr. Sossa Azuela, who holds a doctorate in AI and Data Science.

He emphasized that society should not fear an AI system acting out of malevolent intent on its own, as machines lack free will. "The underlying mathematics behind current AI models is relatively straightforward and incapable of modeling complex human phenomena such as subjective thought, emotion, empathy, or self-awareness."

Will AI Leave Us Jobless?

Dr. Sossa Azuela acknowledges valid societal fears regarding job displacement, particularly in repetitive tasks and increasingly in areas involving automated creativity. However, Dr. Ana María Magdalena Saldaña Pérez, head of the Geospatial Intelligent Information Processing Laboratory at CIC, offers a more nuanced view: while specific tasks will be automated and job roles will evolve, total professional obsolescence is unlikely. Key human-centered occupations—including trades, manual labor, medicine, teaching, and psychotherapy—remain irreplaceable.

This transformation is already reshaping highly skilled professions. Software developers, for example, routinely use AI to draft code but must manually review, test, and debug the output. In this way, AI alters workflows without eliminating human oversight.

"Intelligent agents require constant human adaptation because their objective is task automation and efficiency—not replacement," explained Dr. Saldaña, who holds a PhD in Computer Science. "Many professions depend fundamentally on interpersonal relationships where synthetic systems simply cannot substitute human presence."

The real challenge, experts argue, lies in upskilling—understanding which operational tasks can be delegated and which require human judgment, ethics, and experience.

Addressing algorithmic bias, Dr. Saldaña noted that AI agents are inherently non-neutral because they are engineered by humans and trained on historical datasets reflecting societal inequalities. "If two women and five men apply for a job position, an AI agent evaluates past hiring trends. If historically the role was held predominantly by men, the algorithm inherits that systemic bias and is statistically more likely to select a male candidate," she illustrated.

"Hallucinations" and the Illusion of Accuracy

A central risk highlighted by Dr. Saldaña is when AI models output factually wrong answers formatted with persuasive rhetoric—a phenomenon known as "hallucination."

"AI models synthesize logically connected language, but they lack genuine contextual understanding. When deprived of adequate context, they can deliver incorrect answers with absolute confidence," she noted. "I often tell my students that algorithms act like young children: they believe everything we feed them. If I invent an arbitrary shape and tell the model it represents a letter with a specific sound, it will accept it as fact and attempt to link it to known alphabets."

Dr. Miriam Pescador Rojas, academic coordinator for the Master's and PhD programs in AI and Data Science at the School of Computer Science (Escuela Superior de Cómputo, ESCOM), reiterated that linguistic coherence does not equal comprehension or truthfulness.

"AI models are trained on billions of parameters, making their responses sound convincingly real," Dr. Pescador observed. "A well-known legal case in New York involved an attorney who submitted a legal brief generated by AI containing cited court precedents that sounded entirely authentic—yet upon verification, the cases turned out to be completely fabricated hallucinations."

Consequently, delegating autonomous decision-making to AI in critical fields poses severe operational and legal risks.

The Essential Need for Human Oversight

Dr. Pescador, an SNII researcher candidate, compares AI to a sophisticated calculator capable of executing complex computations in fractions of a second: if fed incorrect input data, it inevitably yields flawed results. Human-in-the-loop supervision remains imperative.

Illustrating this point, Dr. Sossa Azuela cited a recent milestone where an AI model derived a prospective 166-page "solution" to the millennium-prize Navier-Stokes equations after 88 compute-hours. However, proving the mathematical validity of that output will now require at least two years of rigorous review by professional mathematicians.

Addressing sci-fi tropes of autonomous superintelligences rogue against humanity, Dr. Saldaña reassured that current architectures cannot independently take control. "The actual risk is not a machine developing its own will, but malicious actors exploiting these technologies to generate malware, disseminate misinformation, or engineer deepfakes to deceive the public."

To counter synthetic manipulation, Dr. Pescador emphasized the necessity of digital provenance standards: "We must implement cryptographic watermarking and embedded metadata within digital files, images, and documents to verify authentic human authorship and source integrity."

The Real Risk: Ceasing to Think

Perhaps the greatest hazard posed by AI is not that machines will begin thinking like humans, but that humans will cease thinking critically for themselves—bypassing literature research, cross-referencing sources, or working through analytical problems due to the convenience of instant automated answers.

Dr. Sossa Azuela warned against growing "AI dependency," citing informal tests during his lectures where audience members frequently struggle to recall five familiar phone numbers from memory.

"AI should serve as an empowering instrument that augments human capability, making us more precise and better informed," Dr. Sossa concluded. "The framework we instill in our students is simple: Natural Intelligence + Artificial Intelligence = Augmented Intelligence. We must dispel myths of AI omniscience. It is an extraordinary tool, but it remains just that—a tool—and must never replace human judgment."

Ultimately, the goal is not merely building more powerful computational engines, but cultivating individuals capable of questioning, analyzing, cross-referencing, and deciding. In an era where technology answers in seconds, the faculty to question the answer remains an exclusively human attribute.