AI Revolution

Hal (Heuristically Programmed Algorithmic) Computer 9000 was the name of supercomputer AI in legendary 1968 movie ‘2001: A Space Odyssey’ by Stanley Kubrick, depicting the story of AI going wrong and making a mistake and falling into a belief to eliminate human crew of the space mission. I am not going to summarize this wellknown movie here, but this Kubrick work made it possible to conceive risks caused by superintelligence, not necessarily by technological domination but failed identity and narrative.

When discussing AI revolution, although it is very tempting and interesting to discuss about automation of many modern jobs making humans almost irrelevant, in this episode, I would like to focus on the risks and opportunities in decision making with the AI revolution. And when I say decision making, to be specific, I will be focusing on the ability of automated algorithms to make decision choices by looking at sets of data versus ability of Humans for similar tasks.

Although we humans are very proud of our achievements and our abilities for decision making, studies by Daniel Kahneman, some of which are summarized in his book ‘Thinking Fast & Slow’ demonstrate our cognitive biases and heuristics we utilize in our decision making. Interested readers would be surprised by some of these findings. If this does not raise a concern for our abilities, For the supporters of Human mind and impeccable deduction abilities, I would like to remind the story of geocentric vs. heliocentric model for description of our solar system.

For thousands of years, up until very recently (in relative terms of course) observing the sun rise from the east and sun set on the west, humans concluded that Sun was orbiting earth. Earth was standing still at the center of the Universe in our understanding. This understanding of course was challenged by very special, bright and curious individuals, Copernicus, Galileo to name a few, by observing different alternative data like motion of moons of Jupiter, Galileo proved heliocentric model of the universe, but we all know the story of Galileo, being sentenced to death by the Inquisition, had to refuse his own finding at the end. But the “truth” was out. It took many years and a Renaissance, now we have a better understanding of our universe and our nature. Leaving the historical and philosophical questions aside, I want to reiterate that, although same observations of sun rise and sun set are plausible and observable in geocentric and heliocentric models, we humans thought geocentric model was the “truth”, in which we believed very firmly and ferociously. Scary right? Well, we all know that some still believe in that geocentric view and most of humans actually live with egocentric perspective through their lives, this is scarier, but a different topic ??

So coming back to today, times of AI revolution, we humans are training our new babies, large language models, LLM, with stories and data of our own, full of biases and heuristics we have accumulated so far. Can we expect AI to perform better than we do?

Well, I believe this is a very hard problem to crack, and if you don’t believe me, I suggest going to any of AI tools, either Chat GPT or Google AI chat bot or any other and ask for plausible models, explanations for an observation you are seeing every day. A round object (please don’t tell it is the sun) appears from the east sky and disappears on the west sky every day. ?? Well I did, and after struggling a little with the AI Chat bot, and going through factors, such as time of appearance and disappearance in the sky, path of travel, external influences raised by the AI bot during our conversation, which by the way, all seemingly relevant, but not really at the heart of the issue,? pushed AI chat bot to model the object as an object in space rather than a human made object near earth, and i got the following conclusion:

“While it's essential to consider alternative explanations and conduct further analysis to confirm this conclusion definitively, the consistent curved path observed by multiple observers strongly suggests that the object is indeed orbiting the Earth”

So at least now, we have some tools suggesting it is essential to consider alternative explanations…Promising right?

As someone who researched system identification and modeling during my Ph.D., I do respect the challenges of this problem of finding the true nature of events in general. Especially now approaching this in the area of digital underwriting where understanding the “true” nature of individuals and firms for their credibility by looking at sets of data is the task we seek to excel, the challenge is big but advance is promising.

For the reasons elaborated above, as researchers and practitioners, while we use our experiences, being aware of our biases, we are focusing on data, alternative data and dare to seek further as there is a real chance to do better removing those biases…Depending on the severity of the decisions, we recommend architecture of ranging from fully automated systems to Humans augmented by AI systems as opposed to full automation in decision making. Careful design of system architecture in which Humans augmented by AI systems offer a promising future where biases are identified and alternative explanations emerge.

If you have any questions, please contact Alper Eker

[email protected]