Reasoning Between Humans & Machines
Since the rise of artificial intelligence (AI) and machine learning (ML), rethinking our skills has become extremely exciting, especially when it comes to reasoning. The latter is linked directly to whether, someday, machines can fully replace humans as we often see in science fiction movies or not.
“Without the influence of custom, we would be entirely ignorant of every matter of fact beyong what is immediately present to the memory and senses.” — David Hume
Being extremely fascinated by the current progress, I discuss in this article reasoning through its different forms. The aim is to highlight that, although machines evolved significantly, humans still have the lead and will probably maintain it for at least many decades, if not forever.
History
The ability to think may infer that we humans have used reasoning since our early days on earth. At that time, it may have been used to make decisions. Still, some researchers think that they might have been a point back in history when we moved from using myths to explain everything to trying to reason differently about the world. As far as many acknowledge, Aristotle was the first one to provide written insights into human reasoning. He introduced two approaches: analysis and synthesis. In the former, we understand an entity through observation. In the latter, we understand a cluster of entities through similar patterns of each entity within the cluster.
Reasoning Forms
As of today, many agree that there are three forms of reasoning: deduction, induction, and abduction. Although was not clear on induction, deduction and induction seem to be what Aristotle referred to as analysis and synthesis, respectively.
Deduction can be defined as the logical progression from general ideas to specific conclusions. Induction can be defined as an approach allowing to dram conclusions by going from specific observations to general rules. Abduction can be defined as making a probable conclusion from what we know. For instance, If you see an abandoned bowl of hot soup on the table, you can use abduction to conclude the owner of the soup is likely returning soon.
It is worth mentioning that some researchers consider that there are other forms of reasoning like analogical reasoning, cause-and-effect reasoning, critical thinking, decompositional reasoning, etc.
Humans versus Machines
When observing humans and machines, we may compare our capabilities when looking at the three forms of reasoning defined above.
When it comes to deduction, it allows machines to beat humans in several games like chess and Go. The movie AlphaGo can be seen in this context. In this splendid story, when Lee Sedol started the game against AlphaGo, he lost the first three games. In the fourth game, he made a legendary move 78 and won against the machine. The loss was added to the history of the machine and it learned a lot from it. Then, in the fifth game, the machine made some mistakes that looked like errors to many professional players. Still, in the end, it won the game with a 1.5 points difference, i.e. the machine didn’t seek to acquire the largest area possible but just to win. For it, as long as it wins, it doesn’t matter whether it is with a large or small points difference.
For induction, the emergence of new methods such as neural nets opened the door to inductive reasoning (Big Data). Still, in situations where there are no “rules” (like in Go for instance), machines cannot decide efficiently what portion of the accumulated data and history is information. A great example is healthcare where there are still many challenges to overcome (See the case of IBM Watson in Medicine).
Things become even more complicated in abductive reasoning where we make an educated guess. In such a case, we choose the simplest explanation (among many) that accounts for all the facts while keeping in mind the possibility of changing our view if new evidence emerges. This makes humans largely superior to machines since such reasoning requires creativity, innovation, and critical thinking.
As one can see, on the three forms of reasoning, machines seem to be surpassing humans in only one of them, which is deduction.
An Interesting Perspective
I read recently an interesting paper in the context of induction in which Amitabh Basu argues that the key to humans' ability to learn with far more limited data compared to the most developed computers could be the phenomenon of evolution.
Formally, from an instance space X (for instance, images of apples and oranges) and “true” labeling of all instances with labels 0 or 1 (“apples” vs “oranges”), the aim is to find a “true” labeling function f* from the observation of a finite subset S in X. The challenge comes from the fact that X is generally must larger than S, and possibly infinite. The author asks then how can one hope to figure out what f* is by observing its values on S? The class of all possible functions f as H.
The author argues that the process of evolution has distilled a small class H that includes the function f*. Since evolution has taken millions of years to distill out a manageable class H that includes the f* we look for and if such a hypothesis is correct, AI systems may need to go through millions of years afforded to us to reach our level.
While evolution is the simplest explanation we have reached so far, such a perspective may encourage thinking of many other perspectives that may also explain (or give hints) why we humans are quite powerful when it comes to induction and abduction as well. The beauty of perspectives lies in the fact that one has to go to different fields than mathematics or computing to try to explain and understand learning and generalization complexity. This highlights that in the future, we may have to study many more topics (than simply maths and computing) at the same time in order to grasp a little bit of what is going on in AI and ML.
This is an extremely stimulating topic where all of us are either directly or indirectly involved. Right? I am looking forward to hearing your ideas and opinions :-).
