AI and Humans

By Meera Vinod

Every era in the technology industry has had its own set of defining buzzwords that defined the spirit of the times. As a millennial, ‘Web development’ is one of the earliest I can remember, then for a while it was big data, then in the late 2010s every other technology article had creative headlines like “10 reasons AI will take away your job”. Today, the hype has only got bigger and words like AI, Machine learning, Deep learning have become the most marketable words in the tech industry. And yet, many (both inside and outside the industry) still struggle to develop a conceptual image of capabilities of AI systems as they exist today, or differentiate between real AI solutions and hyped AI clickbait.

This article attempts to provide some conceptual clarity around explaining the capabilities and shortcomings of AI systems of today and in that process we try to answer questions like:

How is AI different from human intelligence? Can AI take away human jobs? Can AI and human beings co-exist?

In simple terms, the ultimate goal of artificially intelligent systems is to replicate human intelligence. Human cognition evaluates and makes sense of its surroundings by seeking patterns, learning from past experiences, and storing them in the memory so that in the future, it can take the most appropriate action for the given situation. Modern AI systems try to mimic this process. There are many methods employed to ‘teach’ machines to do this. In supervised machine learning, the machine is fed a set of test items with clearly labelled inputs and outputs and the machine learns to identify patterns so that it can identify similar items in the real world correctly. In unsupervised learning the machine is only fed inputs. It does not have a human defined reference point to identify patterns. Rather the learning algorithm is left to itself to inspect the input and identify patterns.

At a low level, this is an attempt to imitate human intelligence. But this does not mean that modern AI systems have become capable of ‘learning their way’ to everything humans do. Human beings are capable of stitching together and making sense out of a wide variety of stimuli, storing them in our memory and then applying them in different contexts. We have 5 different sensory inputs that inform us continuously and simultaneously, and our brain stores and mixes inputs from these signals. This makes us very good at seeing the big picture and making decisions based on cross-thematic inferences.

At present, AI does not have the ability to make decisions that require awareness of multiple seemingly unrelated contexts. For example, we learn traffic rules so that we can safely commute outdoors. If we come across a child wandering on the road, we become alert because our brain makes a cross contextual inference that

1. We must follow traffic rules to commute safely

2. Young children can be careless

3. Hence we need to watch out for a kid playing on the road

This decision process requires recollecting experiences from different contexts which modern day AI has not achieved.

However, AI systems are very good at making minute and detailed inferences from data that are confined within a closed context. The larger the dataset, the more accurate the prediction gets. A successful example being deployed in the healthcare field is in diagnostics. Machine learning algorithms scan through hundreds of images of medical scans and learn to identify patterns of say, cancerous from non-cancerous scans. This involves careful scrutiny of minute details in the image which may feel ‘tiring’ or ‘boring’ to the human mind over a stretch of time, but the machine ploughs through to the last image with the same level of objectivity, accuracy and consistency as it did when it scanned the first image.

With this information, we now have a framework to differentiate current AI solutions with potential from impractical marketing hype. Humans thrive on being stimulated by varied types of inputs - visual, olfactory, auditory and touch. We are good at being creative, taking inspiration from cross-disciplinary observations, and handling tasks with highly uncertain components. We also ‘feel’ which means we can incorporate empathy into our decision making process. On the other hand, AI is very much confined to making steady logical inferences within a closed system (with limited tolerance to perform under external influences).

The core strengths of AI and humans are mutually exclusive and this gives us ample opportunity to combine them to make real world processes more efficient. We can offload tasks that are repetitive and unimaginative to AI systems while we focus on activities that energise our creativity. In the healthcare field for example, this can mean that, in the presence of AI based transcription software, the doctor can be more attentive towards their patient, while the AI focuses on the task of recording the conversation. By allocating more time to empathise with the patient, and letting AI take over the writing task, the doctor actually improves the outcome of the task at hand.

In the immediate future, we can expect a lot of success from this codependent dynamic between AI and human beings. But it will be a long time or never before AI can actually remove humans out of the loop.