Artificial Intelligence has been the holy grail of Computer Science for over a hundred years and we are finally starting to scratch the first layer of this incredibly complex system. Currently, all the major players in the Technology business are investing heavily in the R&D of AI systems, but it would seem we are still very far away from the development of a true AI.
To truly get a good grasp on where the industry stood in its quest for intelligent machines, we sat down with John Giannandrea, the former Head of Machine Learning and currently the SVP Search at Google, for a one-on-one. From the conversation, it became clear that we have had the latest developments in automation all wrong, and here is the real picture.
John was quick to clarify that there are three distinct levels of Machine Intelligence; Machine Learning, Machine Intelligence and Artificial Intelligence. Machine Learning is what we have just started to get right and it’s a system where an algorithm can be written to train a machine to behave in a certain way, given certain kinds of inputs.
Machine Learning, a higher version would be where the machine is able to take what it has learnt and adapt it to a new concept and a true AI would be the kind which is able to teach itself new concepts and evolve, just like humans. We have just started to be able to get really good at generating Machine Learning algorithms, but John said we are still very far from having a system that can take what it has learnt, and adapt it to a new situation.
At the very core of any machine resembling the simplest levels of intelligence, is “training.” Every machine has to first be “trained” to process information a certain way. For example, if you show a machine a photo of a Dog, it should be able to correctly label it as a dog. To be able to get that result, Google runs thousands upon thousands of training material through a neural network. A neural network is essentially multiple layers of digital filters that mimic the human brain.
Each layer has “ports” of sorts and they connect with corresponding ports just like the neurons in our brains, depending on the stimulus they carry. So on the input side, they will feed the neural network hundreds of thousands of images of dogs (and only dogs) and check that the output is “dog” for all images. Every instance there is an error, it is sent backwards into the neural network so it can “learn” from the mistake and adjust the recognition pattern. Google has managed to get some really great results from this and the proof lies in the Photos app, which is able to segregate photos based on their content.
You can type “cat” in the search bar in the Photos App and it will show you all the photos in your library with cats in them. That is Machine learning, and it is fairly limited as John pointed out that while you will get all the photos of cats, the “machine” would not be able to segregate them based on breed.
While it may seem “really intelligent” for a piece of software to be able to separate your photos into albums based on their content, or suggest when you should leave for work based on traffic conditions (and the time by when you need to clock into work), Machine Learning at this stage, is extremely limited.
As pointed out by John, it may be able to distinguish cats from dogs, but it cannot identify breeds of cats yet. Machine Learning works only in a very limited scope of variables and the minute even a single variable changes, it will fail to execute perfectly. For example, if you were to dress up a cat as a dog, would the Photos app consider it a dog or a cat?
Google has been using Machine Learning to develop its voice recognition software as well, being able to identify and separate the voice of the speaker from ambient noise. It can detect various languages as well, however, what it cannot do is detect intonations, emotional patterns evident in speech or even something as “simple” as sarcasm. It can only operate in a very limited set of parameters and in order to expand those, it takes a significant number of man-hours and thousands upon thousands of training sessions to get the machine to work right.
Google’s Machine Learning API are, as per John, in their nascent stages, but are developing at a rather rapid pace. Google is using Machine Learning to augment their Search (auto complete), YouTube (suggested videos), Inbox and Allo just to name a few. Inbox has a feature where it generates automatic responses for emails based on its contents and as per John, 10 per cent of mails being sent out using Inbox are using auto-responses.
Allo takes this one step further where the machine learns the way you communicate and then makes suggestions for responses based on what it has learnt. The pinnacle of this technology, however, is the Google Assistant which is able to detect language and even separate commanding voice from ambient noise. Google Now uses Machine Learning to generate relevant information for you, based on your usage patterns.
It is no secret that Google is collecting a lot of user data, and one way it uses this data to it train their Machine Learning APIs. When asked just how secure this was, John said that all data that is used for training, is aggregated into one large pool and is hence anonymised. None of that can really be traced back to where it came from. However, once the API is trained and implemented into a service, then it is able to read the information you have agreed to share with Google and make suggestions based on that.
The information sharing here is twofold, one to train the API itself, wherein your data is anonymised and then once the service is ready, it makes suggestions to you based on your activity. This is how Google is able to give us traffic information on Maps. It collects data from thousands on users who are commuting and displays it on the app, but you cannot identify which pixel on that red line corresponds to your car.
While Google uses the ML algorithms across various of its products, it has also made various APIs available to many businesses and developers. What is interesting, however, is the medical potential the system holds. For example, if a voice assistant is able to identify extreme stress or depression in the voice of the speaker, it may be able to help by either automatically connecting the user with a loved one or suggesting various counsellors in the area.
The next step, which would be Machine Intelligence, is where the phone itself is able to offer suggestions for things even before you think of doing them. For example, if you’ve just managed to land a new job, the machine intelligence in your phone should be able to suggest that you buy a new wardrobe. If you are planning on hosting a party, it could generate a suggested guest list based on the people you’ve been interacting with, factoring in how you truly “feel” about them.
The best part about Google efforts is that they have made their Machine Learning resources available for free under the name of Tensor Flow and anyone can start using the tool to train machines for specific tasks.
Google truly is trying to make significant efforts into providing us a convenience that can have far reaching consequences in our daily lives. With the hectic lifestyles that have become commonplace, having a digital assistant who can keep track of your daily affairs is a rather helpful tool.
We take hundreds of photos every month and it is nice to see them get separated and organised into various categories by themselves. The most exciting thing is that we are just starting to scratch the surface of the convenience this new technological breakthrough can bring to our lives and better products are not very far into the future.
0 comments: