Limitations of AI

1. Limitations of AI

Knowing the limitations of AI can help you make decisions on which use cases to invest in.

2. Limitations of AI

We can think about these limitations in several categories - understanding,

3. Limitations of AI

empathy,

4. Limitations of AI

ethics,

5. Limitations of AI

creativity,

6. Limitations of AI

transparency,

7. Limitations of AI

and dexterity.

8. Limitations of AI

Starting with "Understanding". AI systems may appear to understand inputs and context by their response. However, it is just probabilities. For example, it may seem like an AI solution has an opinion on the best business strategy for an e-commerce website. But, its response is simply based on the probabilities of these words being related to the topic of business strategy. AI systems don’t currently have the capability to understand all direct and indirect context which are important for comprehension. Likewise, there isn’t a concept of common sense, just what is “common” based on probabilities. This can lead to undesired responses from the AI system.

9. Limitations of AI

AI systems lack empathy. It doesn't genuinely comprehend the human experience, such as emotions and relationships. Therefore, it may be good at providing information to customer service questions, but it won't be great at handling customer complaints from angry customers.

10. Limitations of AI

It may be able to recognize words associated with emotions, but it doesn’t understand complex emotional cues in written or verbal communication. At the moment, it can't identify tone of voice as angry versus excited. Again, making it apparent that AI can't fully replace customer service teams.

11. Limitations of AI

Subjective judgment, ethics, and balancing trade-offs in the face of constraints are challenging for an AI system. It may be able to calculate the possibility of an outcome, but it can’t make a decision in the face of a complex, conflicting moral situation.

12. Limitations of AI

AI systems lack creativity and the ability to build novel strategies.

13. Limitations of AI

Generative AI may be able to produce images, sounds, and prose, but this content is only created through prompting by humans. With the right prompts, it can unlock new possibilities and creativity such as this award-winning art by Jason Allen using the Midjourney generative AI tool. Similarly, creatives in your business - marketers, UX folks - can do the same, using it as a tool to come up with fresh content.

14. Limitations of AI

Some AI models can be complex and operate as "black boxes". It becomes challenging to understand and explain the decision-making process, which hinders transparency, accountability, and trust. It’s important to build capabilities within the AI that enables this. One example is explanations paired with AI results to describe why a prediction, classification, or other type of result was given.

15. Limitations of AI

Finally, AI solutions are unable to be dexterous in the same way that humans can. Part of this is a limitation of robot technology.

16. Causes for limitations

One reason for these limitations, specifically adaptability, creativity, and understanding are rooted in not having enough data. A lot is required to build an effective AI model. More is required to build one that can be used for many use cases and situations. An AI solution based on incomplete or biased data will in turn exhibit these biases and shortfalls. Another reason is the inability to encode these complex situations in a scalable way in which an AI model can learn from.

17. Human in the loop

With time, more advancements will address these limitations. Until then - even with those advancements - most AI systems will require a “human-in-the-loop”. This is a design where involvement from a human is integrated into the AI system.

18. Human in the loop

For example, in a content moderation system an AI solution won't flag inappropriate content 100% of the time. A human can evaluate content that has a low confidence score

19. Human in the loop

to determine it's true classification.

20. Let's practice!

There are many use cases some of which we'll cover in the next lesson.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.