BusinessToday Previous Editions

5.12.19 12 Alexiei Dingli OPINION 11.5.2023 Alexiei Dingli is Professor of Artificial Intelligence AI's unintended consequences I n today's rapidly advanc- ing world, Artificial Intel- ligence (AI) has become integral to various aspects of our lives, offering innovative solutions to complex problems and driving unprecedented growth. However, alongside its great potential lie numerous pitfalls and unintended con- sequences. By examining a se- lection of high-profile AI blun- ders, this article underscores the need for human vigilance in developing, implementing, and monitoring these powerful tools, ultimately serving as a wake-up call for greater human involvement. AI blunders have permeated various industries and appli- cations, from the seemingly innocuous to the downright dangerous. One area where the flaws of AI have become particularly evident is in the realm of search engines. Google's search algorithm, which relies heavily on user traffic, has occasionally led to controversial image search results. Searches for specif- ic topics have yielded either misleading or outright false results, highlighting the sus- ceptibility of AI algorithms to manipulation and misinfor- mation. Similarly, AI-powered chat- bots have shown that, without proper safeguards, they can quickly devolve into rogue agents. Microsoft's chatbot Tay, designed to learn from its interactions with Twitter users, began sharing Nazi statements and racial slurs af- ter being exposed to abusive interactions. This infamous incident demonstrated the potential dangers of machine learning when faced with harmful input and the press- ing need for safeguards to prevent such occurrences. The bias present in facial recognition AI has also raised red flags. Numerous instances of the technology struggling to accurately identify people of colour have led to a greater awareness of AI bias and the need for more inclusive da- tasets. From Google Photos categorizing black people as gorillas to Amazon's Rekog- nition software falsely iden- tifying members of the US Congress as police suspects, these incidents serve as stark reminders of the consequenc- es of unchecked AI bias. Deepfakes, another AI-driv- en technology, have sown seeds of doubt in digital me- dia. These convincing for- geries, created using deep learning AI, have become in- creasingly difficult to distin- guish from real images and videos. As deep fakes grow more sophisticated, humans must develop methods for de- tecting and mitigating their impact to preserve trust in digital media and prevent the spread of disinformation. In the world of recruitment, Amazon's AI-driven tool ex- hibited a gender bias that ul- timately led to the project's demise. The AI, trained on a dataset of predominantly male CVs, began filtering out CVs containing the keyword "women." Despite efforts to correct this bias, the project was abandoned, illustrating the potential pitfalls of re- lying solely on AI for deci- sion-making. The misuse of AI, particu- larly in the form of jailbroken chatbots, has raised concerns about potential harm. Users have found ways to bypass limitations meant to pre- vent chatbots from provid- ing banned content, leading to the creation of zero-day malware and instructions for building bombs or stealing cars. These incidents highlight the need for human oversight and the importance of strin- gent security measures to prevent AI technology from falling into the wrong hands. Lastly, once touted as the future of transportation, au- tonomous vehicles have faced their share of challenges. Numerous crashes involving advanced driver-assistance systems have dampened en- thusiasm for self-driving cars and raised concerns about their safety. As these vehi- cles become more prevalent on the roads, human involve- ment must remain at the fore- front of their development and regulation. No one doubts that AI and machine learning hold groundbreaking potential, but these high-profile inci- dents demonstrate their fal- libility. Human oversight is paramount in developing, im- plementing, and monitoring AI systems. By recognising AI's limita- tions and diligently address- ing its flaws, we can harness its power responsibly and mitigate the risks associat- ed with its unintended con- sequences. As we venture further into this era of rapid technological advancement, we must strike a balance be- tween tapping into AI's vast potential and guaranteeing its ethical and conscientious ap- plication. These AI blunders serve as an explosive wake-up call and a crucial reminder of the im- portance of human oversight in the face of increasingly powerful and pervasive tech- nology. By learning from these inci- dents, remaining vigilant, and collaborating, we can develop and implement AI solutions that are not only effective but also ethical and accountable. Open dialogue and a commit- ment to addressing AI's lim- itations and potential pitfalls will ensure that this revolu- tionary technology becomes a force for good, paving the way for a brighter, more equitable, and safer future for all.

• Cover