If there is a billion-dollar industry behind any claim or belief you should question and investigate those claims. Artificial intelligence (AI is now a billion dollar industry, built on top of AI models that are claimed to have intelligent conversation with users, diagnose medical issues, code, entertain and the list goes on. Pushed by Hollywood and AI companies, most people have forgotten the “Artificial” in AI and don’t understand how massive amounts of data can be processed by a computer to produce what we now think are intelligent entities.
We are going to explore how:
- AI models lack imagination, context understanding, emotional intelligence, and creativity.
- Humans possess unique abilities that make them irreplaceable in certain roles.
- Collaboration between humans and AI is the ideal situation, combining the strengths of both to achieve remarkable outcomes.
Artificial Intelligence (AI) has made tremendous progress in recent years, revolutionizing industries and transforming the way we live and work. However, despite these advancements, AI models, in their current form, cannot replace humans. While AI excels in processing vast amounts of data, recognizing patterns, and performing repetitive tasks with precision, it lacks the creativity, empathy, and complex decision-making abilities that humans take for granted.
Intelligent conversations
AI Models do not “understand” in the human sense but rather mimics human communication by generating text that aligns with the patterns it has learned. It can adopt personas, express what appears to be empathy (via learned phrases), and provide information in a way that simulates intelligence and awareness.
Essentially, the appearance of an intelligent conversation is a result of advanced computational power applied to a massive amount of linguistic data, creating a compelling imitation of human conversation.
Models are trained on enormous datasets of text from the internet, books, and other sources. This training and learning is different from how humans learn how to communicate from childhood to adulthood.
AI model’s underlying neural networks excel at identifying statistical patterns in language, enabling them to predict the most appropriate next word or phrase in a sequence. This predictive capability results in fluid and topically relevant dialogue.
Modern AI uses mechanisms like attention heads to track and refer back to previous parts of a conversation, creating a sense of continuity and understanding of the ongoing context.
Advanced Natural Language Processing (NLP) techniques allow AI models to interpret the nuances of human language, including syntax, semantics, and “implied meaning”, allowing for more natural interaction.
While communicating with AI seems magical and interesting, remember AI models lack emotional intelligence and empathy, essential for building trust and fostering meaningful relationships. While AI can simulate conversations, it doesn’t truly understand the emotional nuances of human interactions. Humans, on the other hand, possess a natural ability to empathize, making them irreplaceable in roles that require emotional support, conflict resolution, and complex social interactions.
If you are using AI to communicate directly with your customers, example, as your customer service reps, you are on a slippery slope. Real people prefer to interact with real people. If there is interaction between an AI agent and actual people be careful. Use AI to gather training materials for your customer-facing staff or use AI in a complimentary role. If your customer-facing people are not giving you a competitive advantage, you have a bigger problem.
Medical Diagnosis
This is similar to the techniques used above for intelligent conversations but it combines the former with another dataset. AI models analyze vast amounts of medical data and try to recognize patterns, mimicking the diagnostic process used by human experts. AI models, particularly deep learning systems, are trained on extensive datasets including medical images (like X-rays, MRIs, and CT scans), patient records, lab results, and scientific literature. They can identify patterns and anomalies associated with specific conditions that might be difficult for humans to spot quickly.
The models often use statistical and probabilistic methods to determine the likelihood of a certain diagnosis given the input data. They present these findings with confidence scores, which can sound like a definitive diagnosis.
The models can predict the risk of disease progression or the likelihood of a patient responding to a specific therapy based on their data profile, further giving the impression of making a complete diagnosis.
AI models process and synthesize information from multiple sources simultaneously, weighing various symptoms and test results to suggest potential diagnoses or treatment options. This ability to integrate diverse data quickly can emulate a clinician’s comprehensive assessment.
You must always remember that these outputs are not actual medical diagnoses in a clinical sense. AI tools are designed to support systems for qualified healthcare professionals. They lack the clinical judgment, empathy, and legal accountability of a licensed physician. Medical professionals use these tools as one component of a broader clinical decision-making process, incorporating their own expertise and direct patient interaction to make the final determination and provide care.
Coding or Computer Programming
You should already noticed the patterns from the two previous examples – it is all about the data being used by these AI models. AI models can scour sources like GitHub,Stack Overflow and Stack Exchange to gather info, suggest solutions, and even auto-complete code snippets but AI models cannot code. AI models are as good as the code available to be copied from the internet, books or other text data.
Models from ChatGPT and Gemini are pre-trained on billions of lines of code from these open-source, learning various languages and coding conventions.
Using Tokenization, code and text prompts are broken down into smaller units (tokens) that the model can process. An AI model interprets your request using NLP, example, “write a C# function to sort a list,” then you get the impression of understanding of your intention. Using the transformer architecture, (common in Large Language Models), it recognizes patterns from its training data that match your request. It predicts the most probable sequence of tokens (code) that fulfills the requirement, offering suggestions that fit the context.
This approach can lead to:
- Deprecated code: AI might suggest outdated methods or libraries because they were common in the training data (e.g., code from years ago).
- Lack of context: AI might not fully understand the project’s specific requirements, leading to suboptimal or insecure code.
- Inflexibility: AI struggles with novel problems or highly customized solutions.
AI models cannot replace your experienced coders but your coders can use AI to work faster. Understand this and you will not fall into the trap of thinking that AI coding is a silver bullet. You can, in fact, lose money money and customers if you fail in understanding how to incorporate AI in your technology infrastructure.
AI generating your company code cannot give you a competitive advantage. AI code and competitive advantage is an oxymoron, if you think about it. The generated code is available to everyone. Think about what makes your company unique and attractive, then determine how AI can help you achieve those strategic goals.
Suggestions for human-coders:
- Review and refine: Vet AI-generated code for security, efficiency, and best practices.
- Contextual understanding: Provide the “why” and “what-for” that AI often misses.
- Creativity and problem-solving: Handle complex logic, edge cases, and innovative solutions
The Sweet Spot:
- AI handles repetitive tasks (e.g., boilerplate code, routine optimizations).
- Experienced coders focus on high-level logic, architecture, and ensuring the code is maintainable and secure.
A note to the wise: never use code generated by an AI model in your production systems before it it is thoroughly checked by your experienced coders. If you don’t have experienced coders, you have a bigger problem. Experienced coders are needed to formulate the “correct” prompts for the AI agent.
AI in entertainment
Again, AI models use data that it has been given access, to mimic real entertainers. This is where the limitations of AI model can be seen the most. AI models cannot tell a good joke or be even come close to being as good as a Dave Chappelle.
Creativity and innovation in the entertainment industry are areas where humans excel. AI models can generate content, recognize patterns, and optimize solutions, but they rely on existing data and algorithms. Humans, however, have the capacity for imagination, intuition, and outside-the-box thinking, allowing them to create novel entertainment solutions and push boundaries.
Go your AI prompt and ask it to generate a joke based on something that happened yesterday. Most AI models will not be able to do it effectively, at the time of this writing. You will get similar results in music and acting.
Let’s stop here with the examples but please share your experiences with the limitations of AI.
Returning to a general discussion of AI Models: One of the primary limitations of AI is its inability to understand context and nuance. AI models are trained on historical data, which, no matter how extensive, is limited to the scope of human experience and bias (Please see our post on AI Bias). AI models struggle to grasp the subtleties of human language, idioms, and cultural references, often leading to misinterpretations or awkward responses. In contrast, humans possess a unique ability to understand context, infer meaning, and adapt to unexpected situations.
Despite these limitations, AI is a powerful tool that can augment human capabilities. By combining the strengths of AI (data processing, speed, and accuracy) with human strengths (creativity, empathy, and complex decision-making), we can achieve remarkable outcomes.
Collaboration is the ideal situation, where AI handles repetitive tasks, provides data-driven insights, and frees humans to focus on high-level thinking, innovation, and strategy.
A computer is always faster in generating output but the quality of that output is directly related to the input data. Fast doesn’t mean smart or secure, data is not information and AI is not intelligent, you are.
Use AI in school and work for efficiency gains but but make sure you become an expert in your domain. AI can help you achieve your goals because you no longer have to search and read a number of websites to get an idea of what exist already – AI models do that for you with ease – providing a summary on any topic. Your brain is still needed to determine if your AI is feeding you rubbish, outdated information or putting you at a security risk.
In conclusion, AI models, in their current form, are not replacements for humans, but rather complementary tools designed to enhance human capabilities. By embracing this collaborative approach, we can unlock new possibilities, drive innovation, and tackle complex challenges that neither humans nor AI could solve alone.
Always be skeptical of an any company selling an AI idea to you. The way to counteract an “Hustler” is to “get in the know”.
Note: The above information should also be taken into consideration for companies that claim they can replace your pets with AI robots.
