During the last several years, the great scientific minds have researched to understand human’s cognitive abilities to understand how humans store, process, and access information. How is knowledge organized? How do we constantly learn from our experiences and apply as needed? This has resulted in a deeper understanding of human learning and to possibly answer the most enigmatic question of all as to what intelligence really means. There have been significant efforts to model that into machines to pave way for Artificial Intelligence revolution.
There is still a long way to go to teach machines to empathise and be creative and truly intelligent. But creativity is hard to explain. An idea originating in the mind is transpired from the seeds of knowledge acquired over time but we can’t explain how it got manifested. Greater knowledge gives more capacity to think of more ideas but we can’t explain the causal inferences. We can’t explain how we understand a language or recognise faces instantly even though we are very good at doing all that. In order to teach a machine to be as creative as humans, it would take huge amounts of training data as humans acquire over a long period of time. Ideas in the areas of storytelling, reciting poems, generating visual art, human-like conversations with understanding of emotions, generating music or even humour. The question arises, does it take a scientific mind or an artistic mind to build a great AI system that is creative, intuitive, and intelligent?
The science behind AI
Developing an AI model requires good quality data and an iterative process to derive insights from it by understating complex relationship between the input and output variables encapsulated into a mathematical model. This requires constant monitoring and evaluation to come up with a good AI model that generalises to the existing input data so well that it gives meaningful predictions corresponding to the unseen input data. AI algorithms provide a way for the machines to learn and derive insights from data on their own. There is a feedback loop in the process to seek confirmation just like humans do when they learn something. Some of the more advanced AI techniques model a machine based on how human brain works to process information through millions of neurons. This science is the basis for deep learning approach which is very popular amongst the AI community due to very satisfactory results.
There are a lot of scientific underpinnings to developing AI in research papers but it is mostly experimental in nature and not yet formalised. A rigorous engineering process is usually put in place to combine various models and make them work at scale making predictions more accurate, more interpretable and more acceptable. Multiple operational procedures are required to train the models on fewer resources and validate fast enough to reduce time to production.
This makes a scientific concept applied on data to make it a piece of technology that solves a specific business problem. But is there more to it than just that?
The art of developing AI
Data science is not just about the Mathematics and algorithms. Data only provides information as facts. Machines use algorithms to derive these insights and they keep on stirring the pile of input data until they churn the right or acceptable output. But an expert human judgment is often needed to make sure machines are fed with "right data" and that they learn what we want them to learn. It also requires a deeper understanding of how humans solve the specific problem. This perspective is crucial to provide a certain amount of intuitive sense to the output of an AI model that is not provided inherently by the data or the algorithm used. More often than not a human will look at the output of an AI and hastily judge it. The output of an AI doesn’t just need to be mathematically correct but "humanly correct" too. Simply put, the output is often expected to be intelligent, intuitive and conform to human expectations. This makes it necessary to involve domain SMEs in the process.
For example, in a face recognition system, a deeper understanding is needed as to how humans recognise other humans through their faces. Considering there are changes in the facial structure over time, do we need to look at a face multiple times to remember it well? Or do we need to look at the face frequently over time to develop an understanding of the facial features to be able to recognise someone from a certain distance? Do we build a system that looks at a face from multiple angles before arriving at a decision in classifying the face? All these aspects are important and are directly co-related to the business problem that one has to solve. Or maybe a system needs to be highly confident while recognising a face as it is a critical business need. All these considerations need to be factored in while building such a system.
Similarly, while developing a chatbot it is one thing to develop an intent classifier to understand or classify the user’s ask but it is an entirely different thing to understand and serve the actual need. It is crucial to make the chatbot intuitive and make right assumptions just like a human would do while responding to a query. This requires a good understanding of human psychology and behaviour, as to how we communicate rather than just plugging in a huge AI model that is mathematically trained to understand a language. A chatbot is automatically perceived as great if it responds more like a human rather than like a machine.
Putting it all together
Building great AI is not just about the right choice of algorithm or technique but requires a greater understanding of how humans solve that problem. It requires the best of both worlds, a scientific mind and an artistic mind to induce the right level of intelligence and intuition in the AI model for it to predict well. This is why mathematicians and scientists collaborate with domain experts in the areas of languages, behavioural psychology, and liberal arts in order to create sophisticated, intelligent and intuitive machines.
It is humans’ expert judgement that lays the seed of greatness in an AI model. A good AI model may be the one that uses the right algorithm, the right technique, and the right data, but what makes it great is imbibing it with intelligence, empathy, and intuitiveness. There is something required beyond the science, mathematics and engineering that the name itself suggests. It’s an art to build a great artificially intelligent system. That is what makes it so intriguing. We may call it machine intelligence but there is an element of human intelligence to it as well. So it takes an explorer, observer, analyst, scientist, mathematician but more importantly a thinker to build great AI.
