The Limitations of Machine Learning
Author: Pinkesh Patel ( BSc. Pharmacology, MBA)
Date: 20 April 2021
Many AI Professionals get excited while exploring the Application of AI in their respected projects. People assumed that AI could solve all kinds of the problems but that is not true. There are many tasks which AI can execute efficiently and there are tasks which AI simply cannot execute it. Also, it is quite difficult and time-consuming process to know if the AI can do the proposed task or not. It requires to see few iterations to see success or the failure of the AI tasks. In this paper, we can explore few examples to get the basic idea about the limitation of AI and hopefully that will help us to understand the feasibility of AI tasks. There are tasks where application of AI is not necessary and there are times where AI application can create difficulties in the execution of the tasks. So, we can select the valuable projects as per the limitation and feasibility of AI tools to have successful AI projects.
Self-driving Car: Prediction of gesture
As we know the AI has been phenomenally successful when it comes to self-driving car. But there is certain things AI cannot do in the Self Driving car through at the movement. For example, AI cannot figure the gesture of picture by the Human now.
For the picture of the different human gesture, it would be exceedingly difficult for AI to predict what person is trying to do. In addition, Different culture have different gesture when it deals with the driving. So, one AI model developed for the one culture would not be suitable for another culture. This is also critical safety issue if the person wants car to stop or would want to make turn, it would be very major safety issue if the AI can predict the wrong gesture.
Self-driving Car: Full Self Driving car
Recently, we seen the sad news of semi-Automated tesla Car accident in Texas where 2 people died. In the case fully automated car, this seems to be AI deliverable tasks and there seems to some unfinished work now.
Testa CEO Elon Musk have mentioned that he is “highly confident the car will be able to drive itself with reliability in excess of human this year.”
So, I believe there is needs to be extra cautious, training and extra care required when it comes to deal with AI application in critical tasks.
Diagnose pneumonia through AI
Science and Medical industry have been phenomenally successful for the use of AI Application in the industry. Researchers at Stanford university developed the AI system which can diagnose the pneumonia in 10 seconds. However, AI application has its limitation in the industry. For example, Doctor or medical practitioner can learn quite well through reading about the pneumonia from the textbook or see the few images of pneumonia. However, AI system is not able do that through book explanation.
Robust Data
AI require robust data to work efficiently on the project Execution. AI tends to have difficulty when you are trying to learn a complex concept from small amounts of data. AI also depends on the quality of the data. For Example, if you train the AI based on the data from the highly qualitative and modern laboratory. This same algorithm of AI will do poorly if it is tried in the lab from village where quality of data is not same as modern laboratory. Also, AI tends to do poorly for the new geography, new population. For Example, if the AI model is trained in Asian population. The same AI model will do poorly when it has to deals with western population.
There has been huge demand and huge acceptance of AI in the various industry. This is field which is changing very rapidly every moment. There is certain task AI can execute very easily and efficiently. There are certain tasks which are not feasible by AI today will be likely to feasible in future with the help of new technology and strategy and then there is task which is simply not executable by AI. So, Author suggestion to the professionals exploring AI, keep learning and experiments with the AI tools with keep in mind the limitations to have efficient and successful execution of AI projects.
References:
Andrew Ng (2021) AI for Everyone, Available at: https://www.deeplearning.ai/program/ai-for-everyone/ (Accessed: 27 April 2021).
Danielle Kirsh (2019) AI-powered radiology model diagnoses pneumonia in 10 seconds, Available at: https://www.medicaldesignandoutsourcing.com/ai-powered-radiology-model-diagnoses-pneumonia-in-10-seconds/ (Accessed: 27 April 2021).
Matthew Stewart (2019) The Limitations of Machine Learning, Available at: https://towardsdatascience.com/the-limitations-of-machine-learning-a00e0c3040c6?gi=e14e2ab7ed91 (Accessed: 27 April 2021).
Reuters (2021) ‘Two dead in Tesla crash in Texas that was believed to be driverless’[Online]. Available at: https://news.yahoo.com/two-dead-tesla-crash-texas-185858536.html (Accessed: 27 April 2021).
About Author: Pinkesh Patel
Pinkesh have Over 16 years of experience in R&D, portfolio management, and business development in life science & retail Industry. He is mentor and investor at gold and diamond Jewelry firm ‘Proyasha Diamonds’. Pinkesh has Received B.A. Honors in Pharmacology from London Metropolitan university and MBA from Anglia Ruskin University.
