27/05/2026
Co-Pilot, Not Autopilot: Beakal Gizachew (PhD) on the Perils of AI Over-Reliance
============
At a recent Science Lecture hosted by the Ethiopian Academy of Sciences, titled "Co-Pilot, Not Autopilot: Thinking Critically in an AI-Driven World," Beakal Gizachew (Ph.D.) (AI), School of Information, Technology and Engineering, CTBE, Addis Ababa University, Principal Data Scientist, Kifiya Financial Technology.
Beakal leveraged his dual academic and industry expertise to address the critical boundaries of human-AI collaboration. Central to his lecture was the argument that while AI serves as an unprecedented catalyst for pedagogical innovation, research advancement, and social good, its efficacy relies entirely on robust ethical frameworks and vigilant human oversight.
He issued a stark warning against over-reliance, using an aviation metaphor to articulate the risk, noting that delegating ultimate autonomy to these systems effectively moving AI from a co-pilot to an autopilot role threatens to atrophy human intellectual curiosity and critical learning capacities, eventually leaving society intellectually dependent and compromised.
The presentation culminated in a rigorous discussion with industry practitioners and academics, focusing heavily on whether the current AI trajectory presents a net societal benefit or detriment, and whether large-scale models possess any form of a conscious mind.
Addressing these inquiries, Beakal (PhD) demystified the current state of the technology by characterizing AI strictly as an advanced data-processing facilitator. Rather than a sentient entity capable of independent thought, he described it as an engine engineered to aggregate, structure, and optimize vast datasets with unprecedented efficiency to assist human decision-making.
Whether AI manifests as a societal asset or a liability is fundamentally an engineering, behavioral, and policy challenge, as the technology's impact is determined by how it is utilized. If subverted through poor guardrails, malicious prompting, or deceptive applications that bypass operational guidelines, its capacity for systemic harm is profound. Furthermore, Beakal highlighted the technical reality that AI models are highly susceptible to hallucinations and errors, particularly when self-referencing.
Consequently, he strongly cautioned against feeding sensitive data, such as proprietary financial records or personal banking credentials, into public or unverified AI pipelines, advocating instead for strict data governance and zero-trust user protocols.
