TECH4GOOD
Artificial Intelligence (AI) has made significant strides in recent years but still grapples with fundamental challenges when generating responses. Imagine asking your handy AI assistant for the best burger joint in town, and with the usual robotic flourish, it suggests a “charming, family-run establishment… located on the moon!”
Welcome to the fascinating, entertaining, and occasionally frustrating world of AI responses. While artificial intelligence holds vast potential for making our lives more convenient and has made significant strides in recent years, it's crucial to remember – that these digital wizards are still learning.
A colleague tried asking a well-known chatbot to check on its association skills. His simple prompt was, “How many “t”s are there in rappler?” The response he got was, “two “t”s. He then tried asking another chatbot from a popular search engine with the same prompt. The answer he got this time was, “I am still learning how to answer the question. In the meantime, try searching for it.” He tried the same prompt with the other chatbots and got either one of those two answers.
I tried asking AI the classic chicken joke, “Why did the chicken cross the road?” It gave me several responses. One is the witty “To get to the other side.” The other two were creative answers: “The chicken saw a KFC on the other side” and “Because it was free range.”
While artificial intelligence can now generate jokes, a recent study done by Neurosciencenews.com suggests it lacks a proper understanding of what makes them funny. In an intriguing experiment, AI models and humans were pitted against each other using New Yorker magazine’s Cartoon Caption Contest entries. Tasks included matching jokes to cartoons, identifying winning captions, and explaining their humor. Humans significantly outdid the AI models in all tasks, showing that the models’ understanding of humor still has room to improve.
The above study revealed a significant gap in AI and human understanding of why a cartoon is funny. Comedy is a uniquely human art form, relying on context, timing, and cultural nuances. Generative AI can mimic patterns but lacks the nuanced understanding that makes jokes truly funny. While it can access vast knowledge, it can only approximate life experiences. The authors, however, concluded that AI can be a powerful collaborative tool humorists could use to brainstorm ideas.
Some of you may have read the recent Tech.co news feature about Grok, the chatbot of X, accusing Golden State Warriors guard Klay Thompson of a vandalism spree after it misinterpreted tweets about a recent game last month. The story was generated after Grok took social media posts that said Thompson was “shooting bricks” (Basketball slang for “missing his shots”) a little bit too literally.
Generative AI occasionally “hallucinates.” These ridiculous outputs stem from insufficient training data or model biases. Just as humans perceive things that are not really there, AI can produce content that reflects flawed information that it can gather. It can sometimes fabricate data that appears authentic. Trusting AI blindly is like believing a conspiracy theorist—it might sound convincing but lacks a solid foundation. Imagine an AI historian – it might create historical events that never really happened.
AI systems are trained on massive data sets of text and code. This allows them to identify patterns and make connections. However, the quality of these data sets directly affects the quality of the AI's output. If the data contains errors, the AI will learn and perpetuate them. The most concerning aspect of unreliable AI responses is their potential for bias. If the training data is skewed toward a particular viewpoint, the AI will develop and reflect that bias in its responses, perpetuating real-world societal inequalities.
At the current stage of AI development, we should approach it with guarded skepticism. We should not take everything our favorite AI assistant says at face value. We should do our research and verify the information. Let us check how it was trained and what data were used. I usually do not rely on a single AI source because I want to see multiple perspectives. Remember that AI is meant to enhance our lives, not replace our judgment. AI bloopers? I look at them as sources of amusement but will not let them fool me.
The good news is that AI is constantly evolving. Developers are working on addressing bias and improving AI's ability to understand context. By understanding AI's limitations today and approaching its responses with a critical eye, we can navigate this exciting new frontier and ensure that AI serves humanity, not the other way around. For now, we can laugh at the occasional AI blooper, but remember—do not take everything AI tells you at face value. ([email protected])
The author is an executive member of the National Innovation Council, lead convener of the Alliance for Technology Innovators for the Nation (ATIN), vice president of the Analytics and AI Association of the Philippines, and vice president of UP System Information Technology Foundation.