We don’t have to dramatically alter our educational system for AI as is being argued by many because there is no dearth of educational materials on subjects such as Linear Algebra, Calculus, Probability and Statistics, which, however, can be and are studied even without any reference to AI. And the same goes for language too. We need to reorient the study of these subjects from the perspective of AI and build also a new generation of teachers. It is vital that we apply our minds to this if we do not wish this endeavour to go the way of so many Edtech companies
Abstraction and Analogy are the twin poles of any refined intelligence.
Although I have written about it earlier, this point needs to emphasised regularly, at least in the interest of students who wish to study AI. These two poles point to mathematics and language, understandably with some overlaps. It is worth remembering that these two are the most researched subjects with volumes and volumes of writing.
From the few to many
Pause for a moment and think how, with just 26 alphabets in English and rules of grammar, we have the building block to comprehend the universe, an extraordinary feat. Noam Chomsky has argued, through his concept of deep generative grammar, that human beings are born with a capacity for language. It is a pity that Geoffrey Hinton recently dismissed this idea as silly. Hinton’s obsession with AI has probably blinded him to what is quintessentially human.
In mathematics, with just ten numbers from 0 -10 and with many varied rules, we can express relationships among various elements through equations, albeit of different levels of difficulty. Not just express but condense multiple factors or variables into a single equation. One equation can inform us about many things as does the most famous equation: E=MC2.
It is not just simplification but abstraction, which, in simple words, means not being bound by one thing but moving away from specific things precisely to capture them better!
This orientation is central to training neural networks; else, the training will be based on brute force of volumes and volumes of data. Abstraction’s shadow is analogy. As Melonie Mitchell of Santa Fe Institute says, while discussing training neural networks, “You’ve already lost the battle if you’re having to train it on thousands and thousands of examples. That’s not what abstraction is all about.”
New syllabus – Purposeful orientation
Let me quote Claude Shannon to emphasise this point. In his 1948 paper ‘A Mathematical Theory of Communication’, he makes a counterintuitive observation: “The fundamental problem of communication is that of reproducing at one point either exactly or approximately a message selected at another point. Frequently the messages have meaning; that is, they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem”. As a logical corollary, it does not matter what the ‘content’ is – it could be text (prose or poetry), sounds or visual images and so on. The communication system, as Shannon saw it will simply translate into most efficient codes/bits before they are ready for transmission.
The new generation of students must start with this orientation. Else, they will just learn tools and techniques without being adequately understanding the purpose. This the challenge facing curriculum development as it needs a different perspective than just putting together all the relevant mathematical topics together. There is no shortage of free resources on any of these. Experiment with a search for Linear Algebra and discover what a treasure trove of readings you get and for free. Someone can curate all available content to create a ‘new’ syllabus.
Analogy, Language, Social behaviour
Equally important is to examine analogy as a way of simplifying the many into few, as it makes links among similar ideas and reduce the time and effort taken to understand something new, which is a key challenge in any AI. Analogical thinking can take place across disciplines (Inter-Disciplinary Studies) or even within a single subject, the key being making connections based on similarities and thus condensing the process of learning.
Professor Melanie Mitchell of the Santa Fe Institute has devoted her intellectual energies into using analogy in building better algorithms. In an article titled ‘The Computer Scientist Training AI to Think With Analogies’ in the Quanta Magazine talking about her work, Mitchell maintains that analogy can go much deeper than exam-style pattern matching. “It’s understanding the essence of a situation by mapping it to another situation that is already understood”. It’s something (mapping) that we humans do all the time without even realizing we’re doing it. We’re swimming in this sea of analogies constantly.”
Language enables us to experience the world and not just to communicate and makes us social (even when retreat into ourselves) and this too helps in analogical thinking. As Mitchell says “One of the theories of why humans have this particular kind of intelligence is that it’s because we’re so social. One of the most important things for you to do is to model what other people are thinking, understand their goals and predict what they’re going to do. And that’s something you do by analogy to yourself. You can put yourself in the other person’s position and kind of map your own mind onto theirs. This “theory of mind” is something that people in AI talk about all the time. It’s essentially a way of making an analogy”.
Looking ahead
Rather than fall a prey to manipulated and orchestrated screaming about reskilling and building for the future, as a country, we need to take stock of our current intellectual foundations to address the challenges of creating AI. That calls for honesty and humility to encounter and experience the power of reorientation. Think of it as Archimedes’ lever.
