You may have read about ‘Dark AI’, ‘Light AI’ or even ‘Strong AI’ and ‘Weak AI’. There is another way to lay down the complexities of AI use cases. Table 1 shows categorization of Analytics & AI use cases in terms of three levels of human intelligence:
Let us look at one use case, say a self-driving car along the three levels mentioned above:
Iconic: The odometer displays the speed. This is an accepted normal.
Indexic: The car is driving itself in traffic on a freeway. The new normal, but will work only in a given set of circumstances (for instance, the car cannot drive on dirt roads or spaces without clear demarcations).
Symbolic: On a snowy day, if there is an old person shivering at the side of the freeway, waiving his hands frantically, what would / should the self-driving car do? This scenario would open up so many questions around morality / ethics, civility, safety and so on, which in my opinion, will take a very long time for the machines to learn and implement.
Synergy of man and machine
On May 11, 1997, IBM Deep Blue beat chess grandmaster Garry Kasparov. In 2016, Alphago beat Lee Sedol, the world champion of Go, a game many times more complex than Chess.
There has been considerable improvement in use cases ranging from facial / voice / image recognition. One reason could be the abundance of labeled input data sets. We have as many photographs of human beings, cats, or dogs as we can. The more the machine can play and learn, the better it gets. The same as what happened with chess or Go. The more players signed up to play with the machine, the more practice it got, the better it became. AI is becoming stronger and better at learning and decision making, a process known as reinforcement learning.
The keyword is “perfect information”. Rules, boundaries, inputs and expected outputs are given. Let us say, a player cheats in Chess. How would a machine realize and respond to that? These are the complex and unanticipated scenarios where it becomes extremely complicated to train machines.
To appreciate the wide chasm between man and machine, one would need to appreciate the jaw dropping evolution of the human brain over hundreds of thousands of years. Most of us cannot even fathom the way we think, learn, communicate, improve and adapt. Language is but one big differentiator, for it enables human beings to relate to one another and understand emotions, and creates infinite possibilities of thought and expression.
So, what does all this mean for organizations today?
Read related article - https://www.wipro.com/blogs/k-r-sanjiv/that-awesome-hyphen-between-man-and-machine/
Getting AI right
By combining the machine’s superior “Indexic” skills with a human’s “Symbolic” skills, organizations could create significant value. This is the “Man-Machine” paradigm.
How to identify these scenarios? Instead of academic debates around whether one should use deep learning or neural networks, addressing the two fundamental issues could help:
If there are lot of documents with different formats but with the same or similar information or there are high volume of transactions / complex processes / incomplete or incorrect information, these are good candidates for the “Man-Machine” paradigm. The machine can do the grunt work of going through hundreds and thousands of documents, images, audio files, identifying patterns, outliers while the Human can focus on the seemingly unconventional issues, making sense of the machine output and connecting the dots.
Some examples could be:
Solving these use cases could create value in terms of improving customer experience, saving costs, building new revenue models or creating new products and services.