AI Models Human-like Bias Discloses Intriguing Insights
Artificial Intelligence (AI) models frequently astonish us with their abilities and limitations. A recent exposure is that these models pick random numbers as if they were human beings. This behavior, while artificial, provides deep insights into these systems.
Humans have a well-known limitation: we overthink and misread randomness. When asked to predict heads or tails for 100-coin flips, humans’ predictions often lack the streaks of heads or tails that appear in actual coin flips. Likewise, when asked to pick a number between 0 and 100, humans rarely pick 1, 100, multiples of 5, or numbers with repeating digits like 66 and 99.
Amusingly, AI models exhibit the same behavior. Engineers at Gramener conducted an experiment where they asked several major LLM chatbots to pick a random number between 0 and 100. The results were far from casual. For instance, OpenAI’s GPT-3.5 Turbo favored the number 47.
These models don’t understand uncertainty. They answer questions by looking at their training data and repeating what was most often written after a similar question. This occurrence is a reminder that while AI models may imitate human behavior, they do not think or understand concepts in the way humans do.
The effects of this discovery are insightful. It suggests that AI models, regardless of their advanced capabilities, still have a long way to go in terms of truly understanding and replicating human cognition. It also raises questions about the extent to which AI models can be truly random, and whether this lack of randomness could have effects for their use in fields where randomness is crucial, such as cryptography.
This discovery also highlights the importance of transparency and understanding in AI. As AI models become increasingly integrated into our daily lives, it’s crucial that we understand their limitations as well as their capabilities. Only then can we make informed decisions about how and where to use these powerful tools.
Read More:Â OpenAI ChatGPT GDPR Complaint in EU Over Misinformation Issue
Read More: OpenAI Reveals Enhanced GPT-4 Turbo, Making ChatGPT More Direct and Less Verbal
