When to use ML or AI in your operations?
Welcome back to the blog, today we’re answering one of the most basic questions you might have about automation and ML in your work.
When do I need ML or AI?
While the world is seeing a paradigm shift towards automation, we are seeing more of Machine Learning and Artificial Intelligence in action. The way this works is pretty simple, machines supersede the speed of the human brain and are catching up to the complexities pretty fast. With the advent of Artificial Neural Networks or ANNs, machines are being taught to replicate the biological neural networks showing how the world of ML and AI has been growing exponentially.
The answer to the question of “When ML or AI?” is pretty simple and understandable. Complete credits go to the brilliant brains at MIT who offer this simple definition through their MOC classes on YouTube. Robust AI, as it is referred to, can be understood and used to answer the question at hand with the help of 2 very basic comparisons relating to the task at hand & the decision-making involved which are as follows;
1. Application Complexity v/s Scalability
2. Consequences of the Action v/s Confidence on the Machine Making the decision
Application Complexity v/s Scalability
Easily comprehensible from the graph here, sourced from the AI & ML class from MIT OpenCourseWare, the use of machines in comparison to humans depends on how complex a task is and how many times does it reoccur. The high repetitiveness of a certain task that has low complexity can call for a machine to be used instead of a human where the machine can do the same thing over and over again precisely every single time eliminating the scope of any human error and also helps eradicate the underutilization of time.
An example for this can be assembly line processes in any factory where minute tasks such as placing bottle lids or a label etc. are tasks with extremely low complexity and very high scale. Using a machine in these cases leads to efficiency.
On the other hand, tasks with very high complexities and low redundancy can be performed only by subject experts of the field and cannot be provided to a machine. These tasks can be anything from driving tests, medical operations to designing someone’s home, etc. The fact that machines cannot comprehend why something is to be done, it leaves certain tasks to be done only by humans.
An algorithm can train the machine that anything with 4 wooden legs with a flat top is a table. But the fact that the table is used to put something on it, is not understandable by the machine.
Tasks that can be conducted with the help of machines to reduce manual labor come under the category of medium complexity and scalability. This can range from the use of heavy tools in construction to using software to keep details of transactions where a machine augments a human.
To summarize this comparison:
High Complexity, Low Scale = Humans are a better choice
Medium Complexity, Medium Scale = Machines augmenting humans
Low Complexity, High Scale = Machines are a better choice
Consequences of the Action v/s Confidence on the Machine Making the decision
When tasks have mild consequences and the confidence that the machine holds for decision making in the particular task is high, the task is best matched to machines. These tasks are not very scalable if you take our last scenario and their consequences are reversible in most cases. For example, the refrigerator will choose what the room temperature is and set the right temperature of the air to flow inside it, which (if) in this case is not right, can be adjusted accordingly and the consequences are not very dire.
On the other hand, machines can be used to levy fines on anyone parking in a spot for longer than intended or for overspeeding on roads but cannot be used in giving the verdict in a court case or to perform a surgery autonomously. The mere fact that humans are emotional beings leads us to understand that in multiple cases it comes down to the experience one holds in a certain field to decide what is the best way forward. Thus, in a case where the consequences of a certain task are pretty high and the confidence in the machine in decision-making is low, it’s a task better suited for humans.
Machines augment humans in this case too where the machines only are used to assist the humans in conducting a task. For example, multiple software and technologies are used today in audit as the world progresses towards a digital audit scenario where the tools in use are under the control of the auditor but can highly help the auditor ascertain any factors involving risk for a client.
To summarize this comparison:
High Consequences of Action, Low Confidence on the Machine for decision making = Humans are a better choice
Medium Consequences of Action, Medium Confidence on the Machine for decision making= Machines augmenting humans
Low Consequences of Action, High Confidence on the Machine for decision making= Machines are a better choice
Hopefully, you see this comparison and use all the tools at your disposal to solve real-world problems with their help. AI & ML are the future but their capabilities are only limited by the limits we impose. Used ethically, they can redefine our way of living and can help us have a better functioning world altogether, because of one simple fact: Data is everywhere.
For the MIT OpenCourseware YouTube Video:
For more such articles, stay tuned with us as we chart out paths on understanding data and coding and demystify other concepts related to Data Science. Please leave a review down in the comments.
If you have another comparison to put forth in this debate, please do leave a comment or connect with me. Thank you for reading it all the way.
