AI ML product challenges
Intro
Most corporations today are in the phase where they have started with AI/ML or have some offering in the AI ML space in their products. I would say they are in a space where they have made up their mind that the work is getting complicated and we would need to use AI/ML in the products to make it easier to handle the work. This change is big enough in the world to start looking at all means and ways to adopt the new technology. Though AI is not a magic silver bullet to solve all our current problems, it can help us see the problem in a very different dimension from the data we have collected till now.
Challenges
The Challenge we face in AI ML projects is firstly the Trust factor. Traditional software worked in a deterministic way compared to how machine learning work. Traditionally, we would know what are the inputs, then we would define how these inputs are taken and used ( Algorithm ), we would call it the program logic, these program logic would give the output the same way all the time when you provide the same inputs. The ML way is a bit different. we would present the inputs and the outputs in the past and then ask the Machine learning models to determine which logic, and then evaluate the logic to test with more data to claim if the logic formed is in the right direction ( less error ). A person with logical thinking would be able to relate to the traditional software but would be difficult to understand what logic Machine learning ( ML ) is using because ML can determine similarities in multi-dimensional space. So many times Data scientists are asked to explain how the results were achieved. Some have even created a new term explainable AI.
Explainable AI is a set of tools and frameworks to help you understand and interpret predictions made by your machine learning models, natively integrated.
Secondly, the ML logic depends on the quality of the data it is fed on, the more the Data it, it would be able to derive better logic, As Product Managers or domain experts, it's our role to feed in the relevant data and then clean up unwanted data points, so that the outcome is being taken in a way we would do as humans.
Let me relate you to a quote from parents during childhood. My parents would scold me, for just following the crowd and giving an excuse that the entire class or friends did and hence I followed. My Parents would say
"if all your friends jump into the Well (noun - meaning a deep hole in the ground from which water is obtained ), would you do that as well? "
If today we apply the same logic to the decision tree in AI/ML, then ML would recommend one to jump into the well, if 9 others have done it. But this is not practical for humans. Each individual is different and has different capabilities. A lot depends on the quality of data that is fed into the ML logic, for example, if the data which did not include whether they knew swimming or not was fed, then it becomes a totally bad prediction.
Today the challenge is to feed the required and relevant data to the models which would do predictions of the viable outcome. We call this domain expertise. Today lot of this domain expertise is missing among the Data scientists, who are thrown a dump of available data and based on the different graphs they are predicting and claiming higher accuracy. This put's a simple question in mind, Are we ready to use AI ML in all fields. As an individual, I don't think so. I would drive a car myself than to let it drop me at the Airport. I would not mind assistance, where it's not life-threatening, and may assist me in picking up Stocks for investment.
A lot of people also think that AI/ML would solve all the problems which they are not solving today. My view is it can help analyze in a new dimension, but AI/ML can't solve the problems which we don't know how to solve.
Solutions
Today AI/ML can assist well, thanks to the early adopters of the data collecting corporations. My view is that corporates have to invest still on the collection, and extraction of quality data to drive the organization to scale to the new challenges. There is a saying Data is the new Oil, my view good quality data is a diamond mine.
Data is also segregated among different people in the organization, some are structured and some are unstructured. Corporations must bring it together as a brain map way that all are accessible at any point. Let the experts decide which are required and which are not.
AI/ML would be a good tool to assist an expert User, but today it can't be a leader to solve the issue. There is a lot more need for the Domain expert who can tell or decide which data to feed into the ML models to see that the results are coming near perfect. I am emphasizing the term "Near Perfect" because if your life depends on it and you fall into the 0.01 % error, that is one in every 10,000 times if the machine kills, would you still take the risk of being that one. I would probably not.
It can be indulged or used in cases that are not mission-critical technologies where it can help the expert see the options / or help achieve the tasks. It may produce magical results for a few, but would not be understood by many.
Comments
Post a Comment