Artificial Intelligence Assignment Conclusion
Introduction
This blog post is following my first blog about A.I. After 14 weeks I am writing about my views concerning AI, referencing my first post, i.e. how my views have changed or stayed the same. The lectures about artificial intelligence have thaught me a lot and I would like to share some of the nuggets I got to know.
First off, the relevant article for this conclusion1 is one that is supposed to change my views or at least nudge them into a different direction, than the articles I referenced in the first blog post. The topic is partially about how wrong we are, when it comes to making predictions about A.I. to which I tend to disagree.
My post-A.I.-module-1-course-views
If people like Elon Musk and Ray Kurzweil make bold predictions about the future of A.Is. and how we need to be wary of their development is not something to be taken lightly in my opinion. We have to reconcile this with the fact that Elon Musk owns OpenAI - a company devoted to progressing the development of A.I. and democratizing such advancements by making those accessible to the public. Musk thus has very detailed day-to-day insights into the wide A.I. topic. If he issues warnings, it is definitely warranted.
Even though I now have insight into some Machine Learning techniques and how much manual labor is still involved, i.e. to clean the data and employ the correct ML method, I still belive that an AGI is not too far away. My educated guess would be that AGI is reached in a few decades if not years, but no order of magnitude greater than that. Once AGI is in place, ASI won’t be far away either by having AGI develop a smarter intelligence. I think this effect of exponential progression is not to be underestimated.
Keep in mind that ML is very hot right now and money is being thrown at the buzzword alone. While this certainly attracts bullshitters, it will likely benefit some serious A.I. undertakings. This will move us in a faster pace towards AGI, on a year-to-year basis.
Interesting Lessons
Thanks to our artificial intelligence course at ZHAW I got to know a wider range of A.I. applications than I used to know. I was aware about Machine Learning and its subcategorization in supervised, unsupervised, neural networks, deep learning, etc. However I was less informed about the simpler types of A.I.
Two things I first heard of during the lectures and of which I greatly enjoyed the concepts are Constraint satisfaction problem and Prolog.
Constraint satisfaction problem is a way to solve a problem by setting enough constraints, to make the solution evident to the computer.
Three constraints of type “unique constraint” would be sufficient to solve a Sudoku practically immediately. You just need to tell the CSP library to make the vertical, horizontal and 3x3 fields unique.
Prolog works similarly but allows much richer statements, since the programmer can define its own keywords and then create references and inferences, allowing the problem to be solved “automagically” by the computer, following the established logic in the Prolog language.
Conclusion
I am hopeful and positive that the A.I. future will be a good one. Another industrial revolution, once again removing more rather low-skilled labor jobs but this time it will also hit white collar jobs. However this, as always, will enable humanity to go one step beyond menial tasks once again and find more meaningful work.
We are still a lot of years away, until we will ever face a true AGI. But I strongly belive that we will get there sooner than later. And instead of ignoring or belitteling this topic, we should all pay close attention, embrace this revolution, prepare for it, deal with the risks beforehand, make legislature aware and prepare for possible effects on a rather big chunk of the labor force facing automation, e.g. by seriously discussing universal basic income. It won’t be easy, but if we’re serious about it, I’m sure we’ll be moving towards the most prosperous phase of humanity yet.
-
http://rodneybrooks.com/the-seven-deadly-sins-of-predicting-the-future-of-ai/ ↩