Two examples of my recent interactions with the world of artificial intelligence (AI) were instructive to me in more ways than one.
I was walking down the impressive London Science Museum when I stopped, surprised to see an entire section dedicated to big data. And the star attraction of that was a machine that children and adults seemed to have great fun trying. It was a happiness and age detector.
You had to stand in a designated space while it analysed your face and then, based on data from faces it had studied before — machine learning at work — it told you your happiness quotient and your age. For the two children before me, the happiness was 100% and the detected age was well beyond their years. Both of which made them happy. For me, it turned out the opposite. Happiness was at 68% and age a decade less than my real one.
Obviously, no one had any complaints.
The second instance was different. I was looking for a rare book — a book of plays by Somerset Maugham that I had been unable to find online and in many bookshops. Here’s a gist of my conversation with Siri, the intelligent personal assistant that I turned to for help.
Siri was stuck on finding me a red book and my rare book search ended there, at least for that moment.
My two interactions with the AI world and the outcomes were very different. My expectations were different. So were the results.
In his book I, Robot, Isaac Asimov wrote: “… you just can’t differentiate between a robot and the very best of humans.” Humans are afraid of the unknown and the line can be chilling. But, the truth is the world that we imagined through the eyes of authors such as Asimov, Margaret Atwood, George Orwell, Neil Gaiman and other writers of science-fiction, fantasy and dystopian fiction, is now closer to reality than ever. And building the bridge between imagination and reality is AI.
Researcher IDC in its Worldwide Semi-annual Cognitive/Artificial Intelligence Systems Spending Guide predicts that the adoption of cognitive systems and AI across industries will help drive AI-based revenue growth from $8 billion in 2016 to over $47 billion in 2020; a compound growth of over 55%. Reports from other analyst houses might differ on the exact numbers but are no less optimistic about how spending on AI is set to explode in the years to come.
Which means, we should have people ready to design and take on such AI roles, right? Ever wondered what skillsets go into getting code to perform such AI roles and creating such applications?
According to Prithvijit Roy, CEO of BRIDGEi2i Analytics Solutions, a data science firm, AI represents a three-step process and each step lends itself to a different skill-set. “I see the process of AI as a way of helping machines (artificial) sense the data around them to learn from it (intelligence) to drive actions,” he says. “It’s changing the way we sense, learn, act.”
Step 1 involves making sense of data, a lot of which could be unconventional, such as sensor data. The skill sets needed at this stage are data engineering and an understanding of the entire big data stack.
Step 2 entails going beyond the traditional statistical modelling techniques to apply machine learning. The skill sets needed here are advanced machine learning algorithmic capability.
Step 3 goes beyond the data to enable the machines to make decisions on its own through applications such as robo-advisors, recommender systems or chatbots. The level of such decision making also varies from machines tasked with specific jobs to machines doing everything.
The skill sets needed at this stage are application development skills and, conversely, storytelling skills to create conversations that are real.
Let’s put the sense-learn-act model to test my recent AI interactions. In the first instance, the model sensed and learnt from past data and the results were not 100% accurate but the way it acted was not crucial for my decision making. Hence, I enjoyed the interaction.
In the second instance, my need for the application to act was much higher. I needed an answer. And that’s where the application got stuck. Beyond the traditional skills, the lack of storytelling skills that are essential to making conversations real, probably locked the conversation into a point of no return. So, are AI analysts learning storytelling skills then?
Let’s look to the supply side. How do colleges and institutes need to prepare students for change pre and post AI? Let’s see how the sense-learn-act model changes with two examples.
Imagine there is a fraudulent transaction on your bank account. How does the bank figure this out in the pre and the post AI world?
Akshay Mehrotra, CEO of Early Salary, a company that provides loans to young employees through a machine learning-based lending model, explains the post-AI scenario in fraud: given that fraud is a rare occurrence, the sample data may not be enough to learn from. More importantly, the methods of fraud could keep changing. In such a situation, AI becomes a big change in the way we can assess problems and take more real time decisions in tackling fraud. And the skill sets needed to develop this are not just superior data engineering skills but a deep understanding of human behaviour that could be modelled through alternative data.
Let’s take another very different example to see how companies improve customer loyalty and, in turn, provide a better customer experience.
In the second case, for the chatbot to speak to me in the way a service representative would and to really not just answer but anticipate my queries, the missing skill required is cognitive learning and storytelling. And that’s not a skill set easily found in the traditional data science training programs.
I spoke to Charanpreet Singh, Director of Praxis School, one of the top 10 analytics training institutes in India, to understand the viewpoint of the institutes who train students to be AI- and industry-ready. Most institutes divide their course material into the following broad buckets:
Tools: SAS, R, Tableau etc.
Techniques: Data mining, text mining, modelling
Applications: Functional applications and domain specific applications
“There are two types of students we see enrolling in our course. The ones who come with a background in technology and the ones who have some industry/domain experience and want to shift to a career in data science,” says Singh. Praxis’s curriculum is perennially in beta mode, he says, adding the institute plans to work more with industry practitioners.
So, is it only a question of time then that the institutes can adapt their courses to the new industry needs and the students can truly become AI-ready? Will behavioural sciences and storytelling become increasingly important to train AI actors (chatbots, recommenders etc.) to mimic real conversations?
Roy of BRIDGEi2i says analytics some years ago was unknown and untested as AI is now. “How companies like GE solved the problem is by bringing in people from different disciplines such as market research, academics, consulting, investment banking and literally building an entrepreneurial culture where the smart minds learnt to collaborate and crack the code.”
The answer is clear: instead of hiring people with backgrounds just in data science or application development, companies and training institutes will need to invest in skill sets such as storytelling that will help AI specialists to write code to better understand and react to human behaviour.
“It’s your fiction that interests me. Your studies of the interplay of human motives and emotion,” Asimov wrote in I, Robot.
Given the close link of AI with the world of fiction, it’s probably just sweet irony that storytelling skills are as important as machine learning skills to get the world AI-ready.