Top

Human oversight key for AI’s success, says study

Hyderabad: Artificial intelligence has become a buzzword in recent times. While it is proposed as a panacea for industry to improve efficiency, critics point out some questionable outcomes or uninterrupted consequences from the AI in some cases.

An improper functioning of AI-based Facial Recognition Technology (FRT) — used by police for surveillance and criminal investigation — resulted in the wrongful detention of innocent people.

According to a study on AI by Delhi-based Indian Institute of Social Science Research and Development, the Delhi Police’s predictive policing has been criticised for perpetuating biases in policing practices, often leading to over-policing in marginalised communities.

Critics argue that the algorithms can reinforce stereotypes and lead to unfair targeting without necessarily reducing crime rates.

Similarly, artificial intelligence used by some Indian fintech companies and banks led to inaccurate credit scoring of people because of discrimination based on location, income, or educational background.

In financial services, the use of AI fraud detection by identifying unusual transaction patterns also resulted in grave blunders.

While the AI speeds up the process, experts said human intervention was found crucial to verify the flagged transactions. Blind reliance on AI has led to wrongful account freezes or disruptions for legitimate users.

The AI-powered diagnostics and decision-making tools, especially for screening diseases like tuberculosis, led to misdiagnosed cases or missed crucial signs, leading to improper treatment recommendations — raising questions about the reliability of AI in high-stakes, life-or-death situations.

While AI has transformative potential, the study said it needs careful design, regulation, and human oversight to ensure fair and accurate decision-making, particularly in diverse and complex environments like India. “Over-reliance on AI without critical human oversight can lead to unintended consequences. AI decisions can go wrong due to bias in training data, inadequate testing for edge cases, over-reliance on limited data, and lack of transparency,“ the study said.

A mixed approach in usage of Artificial Intelligence (AI), the study said, strikes a balance between leveraging its potential and maintaining a cautious, human-centred perspective in decision-making.

AI excels in processing vast amounts of data, identifying patterns, and automating repetitive tasks, making it indispensable in several areas like finance, and manufacturing. However, it cannot be considered to make an error-free decision.


( Source : Deccan Chronicle )
Next Story