“Jan 2019: Gartner says 80% of analytics insights will not deliver business outcomes through 2022 and 80% of AI projects will “remain alchemy, run by wizards” through 2020”
Why is this? At one side we are talking about Driverless cars and AI bots beating best players in the world, but when it comes to business outcomes, AI is failing.
Disclaimer: Below given is completely my point of view and you all are smart enough to accept and reject based on your natural intelligence.
Let’s start with design challenges of non-IT industries. This is where I was born and brought up in terms of my professional life, so I am always inclined to start with those examples.
How do we define the best material in the world. Either which lasts forever (too much to ask for) or say which can heal itself and last longer. Human bone can heal itself, but no other artificial material can heal itself on its own.
You may be wondering, what I am getting into. Let me give few more examples before I come to the point
How do we design a best Submarine. WE learned from sharks to get to OUR best design. How do we design a best aerofoil shape for aeroplanes. WE learnt from birds to get to OUR best aerofoils. Even in my first job we developed profile tubes for enhancing heat transfer without much pressure penalty and the profile shape adopted was the shape of leaves, which has the least resistance when air passes over it.
Let me come to the point now.
So what we learnt is that the naturally evolved shapes are the best for their purpose and when we learn from them, we cannot surpass them in terms of performance, but can surpass them in terms of efficiency by continuously refuelling them with fuel or energy.
So how does it relates to AI
AI is learning from humans or data generated by human led processes. So it can be better in terms of efficiency and performance, on the basis of what is learned from the historical data.
Now AI folks may jump on this.
AI is better because it is learning from data generated by multiple humans and it learns the collective intelligence of those humans through the “ generated data”. The phrase “generated data” is very important and also obvious for a data scientist who builds the model. What goes wrong is expecting ML model to behave like an experienced professional. ML model when developed is actually like a fresher recruited from campus with the knowledge of fundamentals and historical case studies, also knowledge of latest technologies and applications but with ZERO practical experience. So what we do is, put freshers on job and on-job training is provided. Even the workflow/processes are designed to review the output from fresher at the right stages and further train them with feedback. This output is further fine tuned by experienced professionals to make it suitable for business.
So does these freshers add value? YES! they do reduce the load on these professionals and also improves themselves over a period of time. ML models needs to be treated like these freshers and they will slowly start working for you. Once they are trained enough, feed them with required energy/fuel/data and they will work tirelessly.
Augment/Enable/Complement natural intelligence but don’t try to replace it and your percentage of success for AI models will go up. More importantly you will further appreciate human intelligence to do better things!
There are lot of open questions here. I leave them to your natural intelligence.
This is also fascinating! With the same article, humans will have different perspective but all AI models will have same/similar weight functions (finally a jargon) for the same article/data.
Done for the day
“Tell me I will forget, Teach me I will remember. Involve me and I will learn” – Benjamin Franklin