Five Stupidly Inept Things That Only an AI Can Do

Artificial Intelligence has affected different areas of our lives in every aspect. Yet, amid its remarkable capabilities, AI systems are not immune to mistakes – sometimes, quite silly ones. Hence, while AI seems mighty in today’s era, it’s essential to recognise its limitations while exploring new possibilities. This article shows five stupid things things that only an AI can do.

Artificial intelligence is central to contemporary life. Its applications range from advanced algorithms propelling recommendations on streaming platforms to voice assistants in our smartphones. The majority of AI systems are programmed to execute tasks that are done by human brains, like language comprehension, pattern recognition in data, etc. Still, AI systems are very complex yet need to be perfect. Such shortcomings include:

Misinterpretation of Context

An amusing standard error often found in artificial intelligence practice is context misinterpretation. AI systems still need to improve when picking up on linguistic nuances and context. Although, Natural Language Processing (NLP) has made significant strides. For instance, a chatbot that AI drives might give an incorrect answer to a user’s query if it does not understand the context or meaning of the message.

A prime case is the 2016 incident with Tay, Microsoft’s chatbot, who decided to venture into lousy behaviour on Twitter. While Tay was designed to offer real-time communication and connection to people. It was lured into re-tweeting derogatory content. In less than a day of introduction, Microsoft had to shut down the bot. It was because of the risks associated with AI misinterpreting context and reaching wrong inferences from untrustworthy sources.

Source: Microsoft designer

Moreover,  AI could misread a patient’s symptoms and offer a diagnosis that could have disastrous consequences. An AI-enabled customer service could misunderstand the client’s complaint and provide an unhelpful answer. Thus marring the company’s reputation.

 Generating Nonsensical Content

The creation of meaningless content is another area where Artificial Intelligence could be wrong. Although AI text generators have improved to produce meaningful and relevant text, they may provide irrelevant text when presented with unexpected or structured inputs.

Source: Microsoft designer

AI-generated stories or articles could contain factual mistakes, logical inconsistencies, and incoherent plotlines. This makes the stories far-fetched to human readers. Despite continuous fine-tuning, AI-written content often lacks the ‘sophisticated understanding’ and ‘creativity’ present in human-generated content. This results in absurdity.

AI-generated content may be amusing or even funny. As demonstrated by the known “ AI-written Harry Potter chapter“. This was authored by a predictive keyboard based on the Harry Potter books. The comedy of these examples highlights the limitations of  Artificial Intelligence in comprehending context, tone, and style. These are all core elements of effective communication.

 Overfitting to Noise

Overfitting is an effect when a model is learning to memorise training data rather than to generalise patterns which occur in Artificial Intelligence models as well. The model has memorised the irrelevant noise contained in the training set. Hence, it will perform well on the training set but on unknown data.

Source: Microsoft designer

In 2017, an AI-based hedge fund that was a product of a machine learning algorithm failed to generate hundreds of millions of dollars. This was because of overfitting. After the market conditions changed, the algorithm could only profit from past market data patterns. That was not predictive of future trends, leading to large financial losses.

A challenge of machine learning is overfitting, which is particularly prevalent in the case of small or noisy datasets. It underlines that regularisation is one of the fundamental aspects of prevention.  Artificial Intelligence models from learning redundant features from training data. The performance of AI models can be improved upon generalisation by cross-validation methods, regularisation, and data augmentation. This also helps prevent overfitting.

  Bias and Discrimination

Artificial intelligence frameworks can coincidentally sustain and enhance cultural biases and discrimination present in the information they are prepared on. This peculiarity, known as algorithmic bias, happens when simulated intelligence calculations learn and imitate inclinations present in the preparation information, prompting discriminatory results in decision-making processes.

Source: Microsoft designer

For example, complaints have been directed at AI-based recruiting systems for giving preference to particular demographic groups. While discriminating against others on account of gender, colour or race. Darker-skinned individuals have shown more mistakes in face recognition algorithms. This might be attributed to the bias existing in the datasets used to train these systems.

Algorithmic bias is not only unfair, but it also reinforces and perpetuates the existing social injustices. It may lead to unfair treatment of individuals or groups. Denying opportunities to the segments of our society and, thus, widening the social gap. The development of fairness-aware algorithms, which can reduce bias and discrimination. This discrimination needs to cut algorithmic prejudice. Accurate data analysis is imperative when building AI models.

 Lack Common Sense

AI systems are sophisticated, but their disadvantage is that they need to include what is said to be common sense. That is which involves the ability to understand the obvious or any other social expectation. AI algorithms can deal with large amounts of data and make predictions based on statistical patterns. Yet they may need help with uncertain or confusing situations requiring independent thinking.

Source: Microsoft designer

With no clear guidelines on how to understand such a question as “Can you bring an umbrella when it is raining?” an AI–powered virtual assistant will struggle to respond. AI driving systems might need to make various choices in challenging traffic situations. This must be an understanding of social cues and unwritten road rules.

A lack of common sense brings hazardous results where safety is concerned. Such as in self-driving automobiles or medical diagnosis systems. Although AI systems are capable of being taught from massive amounts of data when faced with new and unexpected situations. They may still need more human intuition and flexibility.

 Artificial Intelligence has made remarkable advancements in various fields. But it has its challenges and limitations. From misinterpreting context and generating nonsensical content to perpetuating bias and lacking common sense. AI systems can exhibit behaviour that ranges from silly to harmful. As we continue to develop and deploy AI technologies, it is crucial to remain vigilant and address these challenges to ensure that AI remains a force for good in society.

If you enjoyed reading our articles, please support us by buying our geeky merchandise on Instagram. Or, you could buy us a coffee or follow us on Facebook, Twitter, Pinterest, or Medium.