I am a huge proponent of open source technologies and having worked predominantly in various web technologies in the course of my career, I mostly work exclusively with open source stacks.
In the recent past I have been very keen to try and wrap my head around Artificial Intelligence and more specifically Machine Learning, although I am still barely even scratching the surface of the nitty-gritty’s of the field, especially the mathematical underpinnings, this is still a huge deal!
Google open sourced it’s second generation machine learning system TensorFlow!
It is a production ready library which has support for running numerical computations using data flow graphs on multiple CPUs, GPUs or mobile devices! Data Flow graphs as explained here represent mathematical computations described using directed graphs where each node corresponds to a certain numerical computation whereas the edges connecting the nodes represent the data communicated between them.
The documentation looks really impressive, with lots of examples. I am really keen on trying something out with this in the near future.
Considering that this is open source, I strongly believe this will give rise to a lot of innovative & creative applications and from the looks of it bring down the entry barrier drastically of solving real-life problems where this can be applied and made use of to make intelligent systems!
The fact that it also provides portability to work on mobile devices sounds like a game changer that should definitely drive a lot of adoption within the machine learning community & we should see a lot of interesting applications leveraging mobile devices which are packed with loads of sensors collecting all kinds of data from our day to day life!
There has been tremendous advancement in the domain of Artificial Intelligence (AI) and Machine Learning (ML) in the recent years. Although people often like to take a shot at the state of AI on the grounds of the fact that several of the early claims on AI that it will soon become almost as real as humans haven’t really materialized, there still has been an incredible amount of growth in this domain.
Today AI is a bigger part of our daily lives than we realize. For instance every time I search for something on google, google remembers what I searched for and makes informed, calculated decisions in the future based on my search history. Often when searching for the same search pattern, it is likely that the search results I receive might be marginally different than yours because of google’s AI / ML algorithms, who try to understand my search queries in the context of my search history, and decide what I might be more interested in, usually based on things that I search most often about, things I like more based on search results that I clicked on, etc.
I have often been rather vary of the implications of such a system and find myself going to incognito window to search when I wish to search for things that I don’t want to be associated with, a prime example of this would be when I am trying to validate some viral news.
Now given that I am generally always online on google, and given the understanding of how AI’s might work and evolve, I was thinking about a side effect that might be possible to tap into by modern spyware. A spyware could simply just sit on my system and make google searches in the background with specific keywords. It could make 1000s, 10s of 1000s of searches in the background on my behalf and hence gradually, so to speak, train google’s AI / ML algorithms into believing that I am really interested in something that I really am not.
Google’s case is just one of many such services who make use of AI & ML, most recommendation systems make use of AI & ML to learn about my choices and recommend things that I might like, advertising platforms also make use of a similar approach to target users with ads of things they might be interested in and so on.
This could very well be a reality today, it is quite possible a similar approach is already being used by several Spyware to manipulate such artificially intelligent systems by feeding them wrong information and hence indirectly influencing their decision making. The disturbing point to note about such an approach is the fact that it is virtually impossible to undo the damage that has already been done.