AI and its impact on work

Preamble

There is considerable debate in professional and public media about the impact of AI, and assertions from both groups- those who welcome and indicate that there would be no loss of jobs and those who hold just the opposite- would have some consequences and would remove jobs. There is not much other than the advent of computers and digital technologies that are referenced to hold either viewpoint. This paper attempts to generate a structured discussion to arrive at an informed view.

Basis

AI, as the name suggests, is a synthetically created intelligence. The very basis is to support or aid the intelligence of humankind. Let us first see the construct of our brain, not in medical terms, but in a functional way, and see where emerging technologies have come from and what the result has been.

We have memory and knowledge/intelligence as two components in our brain. The memory holds the historical experiences and, in some cases, is supplemented by the perceptions of our knowledge/intelligence. The memory holds visual, aural, and textual experiences. Until a few years ago, we generally had the technology capability to hold only texts and static visuals as images, but now they are segmented to hold aural and visuals in video format.

One of the important qualities of memory is the retrieval based on need. We now have search engines capable of performing that on a much more extensive library collated from public and private sources. While it would have certainly diminished the role of those who held good memory and recall, there are still gaps in the search- either because the searcher is not fully conversant or the engine is not fully equipped to do so. With AI coming in through large language models, it is getting better. Its potential is apparent, but it is not at a scale that we can notice to what extent it has bridged the gap. Even after it gets to scale, there would still be gaps. This is because, like how those engines evolve, people would also do.

Now, let us turn our attention to knowledge/intelligence. Knowledge is always cultivated. Intelligence is partly native and partly cultivated. They rely heavily on memory to feed, support, and build the base. The demarcation between conscious and subconscious segments in them varies among individuals. While it is so, in the majority of them, the subconscious part is relatively high. They come to the surface when they encounter a usage.

Machine learning and AI complement each other. They refine this by looking at relationships among the variables defined in the allocated data sets, termed supervised learning. Though there are also technologies that support unsupervised learning using non-labeled data, they are yet to mature to warrant attention right now. AI/ML can outperform humans on the cultivated knowledge part because of their learning and analytical ability and the reach to a broader information base. On the cultivated part of intelligence, they can, over time, come closer to humans since that would depend on the algorithms and the labeling. It will be interesting to see if AI can do something similar on the native part. There are a few claims that it does, but it is yet to be proven.

A comparison is often held between conscious human intelligence and AI/ML. Not all people's knowledge is in the public or private domain. Not all that is available on those domains are accessible by these engines. Even when they gear towards accomplishing that, will they outperform people?

Their ability to crunch a large amount of data quickly is a given. Yet their knowledge domain is limited by the labeled data and the algorithms. Humans are intuitive by nature and evolve rapidly depending on their circumstances and encounters.

New learning in humans may have seen a slide but will catch up. These engines are focused on bettering the existing. We would be better off if we understood this and harnessed the AI.

Conclusion

If one goes around asking people working in any field, there would be some repetitive tasks that do not involve judgment or the application of new rules. The extent of the need for judgment or new rules can vary depending on the level and importance of the task. Repetitive tasks would have been automated using some technologies in the past, and if there were limitations, they could be augmented by AI/ML. So long as a judgment or new rules count for nearly 30% or more of the task one handles, one can be sure these engines cannot replace them. Can these engines enhance the quality of people's judgments? Most certainly, they can in many occupations.

One better test: how many governments would rely substantially on this for security, and how many people will dramatically trust it for their health care?