Decisions, Predictions, and AI
Chennai is cold. Can you believe it?
It is cold in Chennai. I mean, anything below 30 degrees Celsius is winter in Chennai and it is ~25 degrees now- especially early in the mornings and post sunset. There is still odd drizzles here and there and the breeze from the sea has been forgiving. It is pleasant, more so when you are driving a bike/scooter and there is no traffic.
However, that wasn't the case for me as I waited for my friend outside her house. Mind you, it was dusk and her house is near the Marina beach where the breeze was supposed to be pleasant. It might have been. I don’t know. My mind was fully occupied with something else. I was wondering what to write about this week. Nothing came up but also everything was running around in my mind. Ugh, it is difficult.
After she came, we went shopping to buy clothes for me. The mid-season sale was on. Offers abound. So many options to choose but I couldn’t pick something that I liked better than others. My mind had had it by then. I gave up. I was not able to decide on anything. And my mind was full and empty at the same time.
Leaving her house and heading back home, I pondered on how tiring decision-making can be. However trivial the decision might be, it is still tiring.
Decision fatigue is real. It is not for nothing that and Zuck and all those successful cool people stick to wearing same/similar clothes everyday. Our brains have limited store of energy with which we take decisions. And the more decisions one takes in a day the worse the decisions become. That is, assuming that we take good rational decisions when we are fresh. Behavioural economics gangs might dispute with that assumption as well.
Danny Kahneman’s system 1 and system 2 framework shows that well. Essentially, there are two people in your brain - system 1 and system 2. 1 is intuitive, automatic and does not need any rational reason. 2 is slow but systematic and deliberative. The interesting part is that most of our decisions, even the ones that you think that you have patiently analysed and decided are taken by your number 1 guy. Our brain does some shady stuff- it post facto rationalises your random decisions making you believe that you actually deliberated and made a decision using your analytical mind.
Related to this- on how bizarre our brain is. Read about the split-brain experiments by by Michael Gazzaniga and Roger Sperry. In one such experiment, participants were presented with stimuli (visual or tactile) in a way that only one hemisphere received the information. They were instructed to pick up an object with their left hand (controlled by the right hemisphere). When asked why they picked up the object, the left hemisphere, which typically handles language and verbal explanations in right-handed individuals, would provide an answer.
What made this particularly fascinating was that the left hemisphere didn't have direct access to the information processed by the right hemisphere. As a result, the explanations offered by the left hemisphere were often confabulated or inaccurate. The left hemisphere, responsible for language, seemed to create coherent narratives to rationalize actions initiated by the non-verbal right hemisphere.
Let's suppose, for a moment, that our decisions are rational. How do we go about making them? Decision-making, simply put, is the act of choosing something based on the belief that the outcome aligns most closely with our desires. Basically, it involves predicting that a particular choice or option is the most fitting for achieving the outcome we want. It means that all day every day, one has to predict to survive.
Lesser animals like humans can afford to rely on their ‘gut’ feeling (but actually it is your system 1 guy living in your brain making random (but evolutionarily consistent) decisions) and be happy/proud about it. But when you are an entity trying to create value in this world (euphemism for making money), better predictions mean better decisions mean better outcomes.
That is the subject of the book that I’m currently reading called Prediction Machines: The Simple Economics of Artificial Intelligence. I’ve just started reading it but has been interesting to say the least.
I know you must be hearing all about AI this AI that. Some of you must have already gotten ‘AI fatigue’ from the endless AI bros raving about how cool all of it is. But the authors have a neat clean framework to look at AI.
When people talk of AI in today’s world, they are usually referring to Machine Learning. Machine learning is basically teaching a computer to recognise patterns and make decisions based on examples. For instance, if you show a computer lots of pictures of idly and dosai, it can learn to identify them on its own. So, when you show a picture of a white, round, moon surface like thing, the computer will predict that the thing is idly and not the actor Kushboo (crass right? click here).
The improvements in the last few years, the authors feel, has made predictions made by these machine learning models reliable and more importantly cheaper.
Take a moment to think how impactful cheap reliable predictions are. You are writing an email, the machine predicts what you are going to write and auto completes it. You are ordering something on Amazon, the machine predicts that you might also need something else and suggests it. Predictions are everywhere. The authors suggest that everything can be looked at from a prediction framework - diagnosing a disease is a prediction problem, self driving cars is a prediction problem even translation is a prediction problem.
Take the self driving cars example,
Autonomous vehicles have existed in controlled environments for over two decades. They were limited, however, to places with detailed floor plans such as factories. The floor plans meant engineers could design their robots to manoeuver with basic “if-then“ logical intelligence: if a person walks in front of the vehicle, then stop. If the shelf is empty, then move to the next one. However, no one could use those vehicles on a regular city street….. - to many “ifs” to possibly code.
Instead of telling the machine what to do,…… the engineers recognized they could instead focus on a single prediction problem : what would a human do?
Now companies are investing billions of dollars in training machines to drive autonomously in uncontrolled environments, even on city streets
As time goes on, corporations will collect more and more information (data) from you and these prediction models will become far more reliable. Do you know Amazon has a patent on “anticipatory shipping”- a system of delivering products to customers before they even place an order. Amazon will be able to predict what your system 1 thinks is an impulse buy and ship it to you before you know you want it.
I’m definitely not an AI alarmist, mind you. This is all very exciting. Exciting interesting stuff ahead. I will share more stuff from the book in the coming weeks.
In the mean time, if you are interested in understanding ML concepts or even learning ML, hit me up. The best form of learning happens peer to peer. Remember during college when you learn with your friends in one night what the professors were trying to teach you one whole semester? Peer to peer.
Reccos:
I have recently been watching Silicon Valley. It is amazingly funny. Highly recommend.
Lex Fridman talks to Jeff Bezos
Had a great time watching Sumukhi Suresh’s Hoemonal. Hilarious. Highly recommend. Also watch Pushpavalli.
That’s all folks. Drink water. Sleep well.
Until next time!
👌🏼👌🏼