Warning: Long blog ahead…
I am pleasantly pleased with myself for finally managing to complete 2 tasks this week. They may not look like much to an outsider, but personally, it had enabled me to connect the dots in more granular detail with regards to my AI learning journey. It has also helped me understand where we are now and what lies ahead.
Firstly, I completed the AI in Finance MOOC (Massive Open Online Course) which I signed up for in Mar. https://cfte.education/aifinance/ It consisted of 18 modules and took me about 3 months to complete. It re-enforced AI concepts which I had studied about previously. It also helped me to conceptualize the AI big picture, on the direction it is likely to move in the immediate future.
While the course was also a good introduction to AI, the market professionals presented a roadmap as to where they believe the areas of finance will most likely see the greatest impact. The underlying theme is clear. Current AI is narrowly focused – it does very well when directed to a specific purpose. This is due to its ability to review unlimited unstructured data, to be able to detect trends that no humans can.
We had supervised learning where AI is given a set of data and told of what the final outputs must be. What is more interesting is unsupervised learning, where through deep learning techniques, the program seeks to find meaning and trends by itself. For example, the machine learning program can be given millions of images to review and over time, it could correctly identify pictures of cats and dogs on its own. Effectively, it learns on its own.
With reinforced learning, we can move to the next level. This is now possible as technology has finally caught up. The programmer basically provides the AI machine with positive and negative rewards (incentives and disincentives) to drive its behaviour in order to optimize its performance. The AI is free to make its decisions towards a final goal. Over time, it will become better as it gathers useful experience going through the processes again and again.
The Alpha Go phenomenon https://www.alphagomovie.com/ that happened in 2016 was created by Google DeepMind based on this concept. It started to analyze Go games played by humans and slowly, it developed strategies which no human in history had ever thought of. It went on to beat the top human player in the world and started a mad development scrabble for AI everywhere. The successor to it was Alpha Go Zero, which took the giant leap of learning by itself without any human inputs. It beat Alpha Go by a resounding 100 to 0 games.
Since then, China had committed a national policy to AI. Gamers have also moved on to more complicated games to challenge AI. Earlier this year, a Starcraft AI managed to outsmart humans. Just yesterday, it was announced that an AI had beaten the best humans in Poker! https://www.theverge.com/2019/7/11/20690078/ai-poker-pluribus-facebook-cmu-texas-hold-em-six-player-no-limit
The other thing that I completed this week pertains to something I did as a teenager many years ago. I started to read science fiction books as the topic interested me then. The name Isaac Asimov constantly pops up as the finest author of such books that tries to predict the future. I began to read as many of his books I can get my hands on. During the 1980s, the only way to do so was to find them in 2nd hand bookshops or in public libraries. After trying to finish the “Foundation” series, my interest waned as the concepts became too abstract to me. I had moved on to other genres by then.
With my current interest in AI, his book “I, Robot” came into my mind again. It was a collection of short stories he compiled when he was in his thirties, way back in 1950. I was curious to understand how a futurist could be describing robots more than 70 years ago. This was way before they had basic functioning computers and this man was already projecting 100 years into the 21st century and letting his imagination run wild. I wanted to connect with him again after so many years, to hear about his words of wisdom on AI.
In this book, he elaborated on the development of robots through the years via the Robopsychologist called Dr Susan Calvin over 9 short stories. Robots had positronic brains that had the 3 robotic rules hard-coded into it. (1) Robots cannot harm humans, (2) Robots must obey humans except when it conflicts with the First Law, and (3) Robots must protect its own existence as long as it does not conflict with the First or Second Law. I was always very impressed that the Laws were so elegantly well thought out and to be able to encompass everything in a nutshell.
In the final 2 chapters, the robots finally caught up with humans. Eventually, humans relied on the machines to run everything. Humanoids became world co-ordinators and they become indistinguishable from humans. The fear of machines taking over had diminished over time. The timeline of the book was between 1980 to 2060, which is where we are today in 2019, the halfway mark.
As I mentioned earlier, the AI as we know today is what we call narrow intelligence. It does well in a narrow frame of reference and does not cross into multiple fields of human domains. It does a particular task extremely well and the more data you feed it, the better it becomes.
It is not that difficult to program AI solutions given the powerful software that is easily available nowadays. During my 1-year Business Analytics evening classes, we had projects used software like SAS Enterprise Miner that were crunching hundreds and thousands of data points in seconds, using various statistical formulas to optimize end results based on the steps we had directed the program to execute. Little did I know then that this was part of machine learning and it was part of my AI learning journey.
The current futurists predict that General AI will only be possible in 30+ years time, where narrow AIs combine together in a meaningful way to become what Isaac Asimov was talking about in his book. Androids and Humanoids will only be possible then.
AI has been the buzz word for the last 3 years and the advancements already achieved has been incredible. We are only at the start of an amazing journey that will continue to surprise all of us for years to come. AI Ethics and fine-tuning of unwanted consequences are some of the things we will still need to address. But what we will take for granted in the near future will seem to be impossible now. The future is so bright, I’ve got to wear shades!
Leave a Reply