Deep Learning: Teaching Machines To Act Human
Recently there were increased news articles about AI: Artificial Intelligence. Some very smart people were concerned about the progress made in the field of advanced machine learning. Among them were serial Entrepreneur Elon Musk, the famous researcher Steven Hawkins and legendary philanthropist Bill Gates. All of them signed an open letter expressing their concern about the future of AI. Cause for the signage was a video showing Google owned company Boston Dynamics recording a trial run of their human robot ‘Atlas’ running through the woods among other recent advances in advanced machine learning.
What is advanced machine learning?
The field of machine learning in computer science has been there for a while. Starting during the second world war, the first attempts to teach computer to learn and being human were made. The recent movie around pioneer Alan Turing shows the origins of this scientific research field. Until today the Turing Test is still applied to evaluate if a computer is categorized as intelligent.
During the 80ies and early 90ies further attempts were made to teach computers to behave human. Early solutions weren’t practical caused by the limited processing power during that time. A lot of time passed since then.
So what exactly is machine learning? It’s basically to teach a computer to make sense of data. To teach him to recognize patterns in input values and gain insights from the process. Simple machine learning can be a regression analysis or simple classification of data depending on a single value pair into different categories. Advanced machine learning, of which deep learning is a part of, applies multiple analysis layers in analyzing big data sets. The first layer of an algorithm look only at certain parts of the data and then deliver the output value to an analysis layer further up the hierarchy conducting more abstract calculations with the input from the lower layer and itself delivering values to an even more abstract layer of algorithms. This structure allows the modeling of the human brain, imitating the network of neurons in the brain with many (trillions) of synapses.
Tapping into the huge potential
Today many layers are applied to solve difficult data analysis problems, therefore the name deep learning. With this methodology it is possible to teach a computer to analyze pictures, handwriting, speech, maps or even videos. In the future all applications that seem to be ‘magical’ will be the result of some kind of deep learning. The application are many: Categorization of images, indexation of unlabelled data, analysis of maps, using big data of many sources to refine and improve prediction models and so forth.
Facebook, Google, IBM and many start-ups today already apply deep learning technologies to gain an edge solve difficult problems. Until today there is no computer who can itself program something that can program. But that day will come, its just a matter of time.
Is it dangerous? Maybe. But it can also do much good if applied correctly.
If you’re interested in deep learning, here are some very interesting companies applying this cutting edge technology:
Have you heard about deep learning before? What do you think: Is it the future? Are you afraid of AI? I’m interested what you think so please leave a comment!