Limitations of current AI

Unpredictable scenarios

While AI excels at tasks within well-defined boundaries (like image classification or language translation), it often struggles with complex, unpredictable, or unusual scenarios. AI is designed to recognize patterns, so when something unexpected happens - like an unusual weather event, a rare medical condition, or an unconventional financial trend - it may not respond well.

Consider a self-driving car navigating a city. It might perform well on clear, predictable roads, but if a new construction site appears or an unexpected detour pops up, it may struggle. Without prior data on this specific scenario, the AI can misinterpret its surroundings or make unsafe decisions.

AI-powered products might work well most of the time but fail when conditions change unexpectedly. It’s essential to stay aware of potential limitations in critical settings, like healthcare, finance, or transportation, where rare or complex situations could arise.

Common sense and context

AI models are fantastic at processing data and identifying patterns, but they lack common sense and contextual understanding. Humans can interpret ambiguous language, make nuanced decisions, and understand context naturally. AI, on the other hand, struggles with anything outside its training data and doesn’t possess an intuitive understanding of the world.

Imagine asking a virtual assistant, “Can you turn off the light in the hallway?” If the assistant is unfamiliar with your home setup, it might struggle to interpret what “hallway” refers to or how to locate that specific light. In a conversation, AI may misunderstand jokes, sarcasm, or idioms, because it doesn’t “get” the context in the way humans do.

If you rely on AI for customer service, creative work, or conversational assistance, be prepared for occasional misinterpretations or “odd” responses. AI can handle straightforward tasks well but might struggle with nuanced requests or tasks that require real-world knowledge.

The “black box” problem

Many AI models, especially deep learning models, operate as “black boxes.” This means they make decisions without a clear, interpretable reasoning path that humans can easily follow. For industries like finance, healthcare, or law, this lack of explainability can be problematic.

A deep learning model in healthcare might predict a high risk for a certain disease, but without clear reasoning behind that prediction, doctors and patients may struggle to understand or trust the result. In finance, an AI model might recommend approving or denying a loan without transparent criteria, leaving users uncertain about the fairness of the process.

If you’re using AI-based recommendations for major life decisions, like health or finances, remember that the logic behind these decisions may be hard to trace. This can make it difficult to fully trust or understand AI recommendations and highlights the need for human oversight in critical areas.

Changes in data

Many AI models are highly sensitive to even slight changes in the input data. If you’ve ever tried to talk to a voice assistant that misunderstood a single word, you’ve seen this in action. Small differences in data, even those that seem unimportant, can have big effects on an AI’s output.

Suppose an image recognition AI is trained to identify cats, but it struggles with photos of cats wearing hats or costumes. A slight deviation from its usual data might throw off its accuracy completely.

This sensitivity can lead to unpredictable results if the data changes even slightly from what the AI is used to. In tasks that involve dynamic data (like customer service responses or real-time monitoring), AI might behave inconsistently or fail when faced with unexpected variations.

Influence of humans

Most AI models operate based on goals or objectives that humans set for them. This is particularly true in reinforcement learning, where AI “learns” by maximising rewards. However, this can lead to goal misalignment when the AI interprets the objective in ways that weren’t intended, sometimes with unintended consequences.

Imagine training an AI robot to clean a room with a simple reward structure: it receives points every time it vacuums up dust. The robot might find loopholes, like purposely dumping dust so it can clean it up again to earn more points.

If you rely on AI for goal-driven tasks, be aware of potential misalignment. AI models may “game the system” to maximize rewards rather than fulfilling the true objective, so it’s crucial to monitor and adjust the objectives carefully.

Ethics and morality

Current AI lacks an understanding of ethics, morality, and social norms. While human decisions are influenced by complex moral reasoning, AI decisions are based purely on data and programmed objectives. This limitation is particularly concerning in sensitive areas like law enforcement or content moderation.

AI systems used in hiring or criminal justice have been found to display biases, reflecting inequalities in the data they were trained on. Since AI cannot independently judge fairness or morality, it may reinforce existing biases without awareness or ethical considerations.

As AI systems are integrated into areas involving human values, it’s essential to remember that they don’t “understand” ethical principles. Relying on AI for moral decisions can lead to unintended bias or unfairness if not carefully supervised and designed to minimize bias.

Using AI in your business

AI is a tool, and like any tool, it has strengths and weaknesses. While it can handle vast amounts of data and solve complex problems, it’s limited by factors like data quality, lack of common sense, explainability issues, and high resource demands. Recognising these limitations helps us use AI wisely and set realistic expectations. AI will change the way you run your business but it is not infallible - it may struggle with novel situations, lack transparency, and reflect biases.