Techie-Talk: Are Robots Really Taking Over? Dean Wronowski talks AI and ML

Eleanor Barlow

In our last instalment, we touched upon ML and AI. These days there are so many scaremongering stories about robots taking over. How do you see AI and ML progressing?

 
Purple and White Artificial Intelligence with the Exterior of a Woman.

Dean Wronowski: ‘Whether you hate it or love it, you have to go with it. Because, if you don’t go with it, you will be left behind without an understanding. Even though people say that they will be out of work because of all this new technology, you must remember that this new technology will also create jobs. That’s why you should just go with the flow. Try and get into it. Learn about it. Get involved in it, so that it becomes less of a scary concept. 

Whether you like it or not, it will be involved in your life. If it isn’t already. It’s like using Facebook, an action most of us make every day. There are thousands and thousands of machine learning algorithms running all-day-long on there. You probably don’t realise it, but they are continuous and they are useful.’

Can you give us another example of machine learning working in the background? 

Dean: ‘Algorithms using machine learning, to generate more data using the data they already have, is popular. People upload pictures in their homes on Facebook, for instance. Pictures with their children or their dogs, sitting in the kitchen or living room. Very ordinary posts. Now say, over the last 5-10 years, you have taken a few photos in these rooms and posted them on your timeline. Well, Facebook has a massive database of all these pictures. Facebook (or rather the machine learning behind it) can actually work out that all these pictures are very close/similar. So what it can do is regenerate your whole room based on those pictures. 

Let’s say we took ten pictures of the room we are sitting in now. And these images were posted over however many years you have had an account. Facebook can use machine learning and basically regenerate this whole room in 3D.’

But what’s so interesting about Kevin’s kitchen or Susan’s conservatory?

Dean: ‘Memories. If you have moved to a new house, or just want to return to a place in time, you can generate the whole room in 3D. You can do this by simply putting on some ocular augmented 3D glasses. Once you put them on you are in a 3D space. It will feel like you are back in the place you want to be. You can walk around the room. Even feel like you are walking through other rooms. Every so often, if Facebook discovers that a certain static picture was taken it will update the 3D space to mirror the information in that image. You can look back on memories in the past as if they are right there.

But it gets more interesting. Let’s say one person here in our little town of Bude has their room regenerated. And they have a cousin in America who also has their room regenerated. You can knock on the door of each other’s augmented reality space and go into each other’s rooms and see 3D objects.

This is happening now, but it’s just the start of it because Facebook can now generate your whole living room in 3D. You can click on an object, hold on to it as if it’s in your hands. You can even take that 3D model back to your house and see what it looks like.

Let’s say your friend has some curtains in their house that you really love, but not sure how it will look in yours. You can tap on them, pick them up and take them to your augmented reality home to see how they look before buying.

But it doesn’t stop there. Because you have a 3D blueprint of the whole model/object, and because we have 3D printers now, you can actually print what you like, because you have the spec of that model. You can basically print any object taken from any augmented space in the world. Obviously, the downside is that the more advanced this gets the more advanced hackers get, and things like gun-power and generating 3D weapons could become more common.’

Good or bad, this is a truly remarkable advancement in technology. What new technologies are your favouring at the moment?

Dean: ‘For me, it just has to be machine learning. You can just do so much with it!’

Can you explain what routes you could go down with it?

Dean: ‘There are two routes. There is supervised machine learning and unsupervised machine learning.

An example of supervised machine learning- say I want to find a picture of you amidst all the data online, I can teach the machine to only look for you. I can have a database with about fifty-thousand pictures of you that have been sorted and given labels, taken at different angles and in different lightings. The more images and the more data I give the machine, the better it is at finding you. Because you have your cloud machines and cloud processing, it’s just really fast. And if you have a real-time camera, I could analyse the pictures in real-time and the machine will be able to work out which section of the photograph is you.

But, of course, if I can do that then so can other people and businesses. It is getting more and more available. So you need to know about it and be prepared for it.

That is an example of supervised learning, because I’m basically telling the machine what to look for. Unsupervised machine learning is, essentially, when the machine works out different things itself. So, let’s say you have a picture of a flower, and you want to predict what type of flower that is, machine learning, if it is unsupervised, will actually try to work out the features in that picture and from those features try to name/categorise the flower. It will look at qualities such as the size, length and colour of the stem, the size, colour, and shape of the leaf, angle of the petal, shape of the flower head, and so on. The more features you give it to look for, the more it picks up.

But because it can pick out all these features by itself it also has debates with itself; ‘Does this feature fit? Is this feature the same?’. And the thought of a debating robot is one that can put people on edge.

A while back, I remember reading about how Facebook had a similar issue when they built two robots [Alice and Bob] on unsupervised machine learning and they ended up learning off of one another. Whether this was Facebook’s early days and they were still learning about it or not, I can’t remember. But, because the two robots were able to learn from one another through deep learning, they (it) developed a language that only they could understand. Humans could not understand it, so it all had to be turned off!’

[To read more about the ‘Facebook AI Robot Shutdown Incident’ click here.]

Thank you, Dean, for another truly insightful interview.

To find out more ways that we are telling stories with data, browse our blog or get in touch with our team today.

 (This interview has been lightly edited, for the purpose of clarity.)

Leave a Reply

Close Menu