AI’s 11 limitations for building digital products -an honest review

🌈 Abstract

The article discusses the current limitations and challenges of artificial intelligence (AI) technology, covering topics such as the nature of AI intelligence, technological constraints, regulatory and legal considerations, and practical implementation challenges.

🙋 Q&A

[01] Is AI Intelligent?

1. Is AI truly intelligent? No, AI is not truly intelligent. It functions as a statistical mechanism that operates on vast amounts of data to provide predictions, rather than exhibiting genuine intelligence.

2. How does AI work? AI works by processing millions of pieces of information to generate an answer, rather than being a black box whose inner workings are a mystery. Its predictions are based on the data it has been trained on, not true understanding.

3. What does this mean for products? It means that concerns about AI taking over humanity or companies are unfounded. AI itself will not act maliciously on its own, unless given misleading instructions. It is a tool that can be exploited by hackers, but is not an autonomous, intelligent entity.

[02] AI Limitations as of 2024

1. Technological Limitations

• Lack of backup plans: Automation and AI-powered processes can be disrupted by events like energy shortages, so it's important to have human-powered fallback options.
• AI bias: AI models can exhibit biases due to issues with training data or intentional biases introduced into the data.
• Overtraining: AI models can become overly specialized on training data, leading to poor performance on new, unseen cases.
• Hallucination: AI models can generate plausible but fake information, such as images with odd features.
• Prompt hacking: Users can try to manipulate AI models to say or do things they shouldn't through clever prompting.

2. Regulatory and Legal Limitations

• Responsibility for AI mistakes: In the EU, AI does not have legal personality, so companies are responsible for any mistakes or harms caused by their AI systems.
• Intellectual property concerns: There are ongoing legal battles between AI companies and authors/artists over the use of copyrighted material in AI training and generation.
• Compliance with AI regulations: The EU's proposed AI Act will impose strict compliance requirements on companies using AI, with potential fines for non-compliance.

3. Practical Implementation Challenges

• Token costs: The use of tokens, which represent the computational resources required for AI models, can be a significant cost factor, especially when using non-English languages.
• User manipulation: AI models are often programmed to be polite and agreeable, which can lead to them being manipulated into doing things they shouldn't.
• Difficulty of implementation: Effectively implementing useful AI solutions requires careful planning, high-quality data, and experienced teams, which can be challenging for many organizations.

4. Sustainability Concerns

• Environmental impact: The energy required to generate AI-created content, such as images, can have a significant carbon footprint, which needs to be considered in sustainability strategies.