part0064

I was considering going into some detail in relation to the technological breakthroughs that have occurred and are occurring that are enabling the tasks outlined in the above tables to be undertaken automatically. However, as interesting as these methodologies may be (at least to me), they are being described in the technical papers and articles that are now emerging in AI and latest technology-related publications. Many of the breakthroughs in the tables can, in fact, be summarised in the following way. The world is a complex place; the second law of thermodynamics applies – which states that the total entropy of an isolated system can never decrease over time. In other words, there is a tendency for entropy, i.e. randomness, to increase over time (and this gives an official explanation for why my desk looks like such an untidy mess so much of the time). All this randomness introduces a great deal of variation in just about all situations that we might wish to automate. For example, if you buy a robot vacuum cleaner that purports to be able to automatically clean your floor, it’s of little use if it can’t deal with variations that occur such as, for example, chairs being in various random locations. For the device to be really useful it needs to be able to deal with such variation – the robot vacuumer is clearly of little value if you have to put more effort into ordering the environment of your rooms than you would have to expend by simply vacuuming manually as usual (unless of course you bought it to impress your friends). All this is, however, being overcome by the big breakthrough which is now occurring – specifically the development and useful application of deep learning. This technology is providing a facility for interpreting environments rather as humans would – by using vision to recognise features in a given scene and then employing the resulting information to direct robotic operations. For example, it could be used to direct the robotic vacuumer to avoid obstacles such as tables and chairs that may have been moved to new locations since the last vacuuming.

In fact, the vacuuming robots now emerging do employ sensors such as cameras, gyroscopes, and laser rangefinders to detect and avoid objects in their environments and to create floor plans that make vacuuming more efficient. Having said that, reading the users’ reviews, it seems there are many ongoing performance issues that need to be resolved, as well as practical factors that limit their current usefulness. These include their tendency to be rather small (therefore taking a long time to vacuum a floor and needing frequent dust container emptying) as well as being relatively expensive. Therefore, so far, the economic case for them has not been proved, meaning that currently they are more like expensive toys than practical household devices. Of course, as the technology matures, we can expect this to change.

But possible future applications go way beyond robotic vacuuming. Up to now robotic systems have been confined to very structured environments, but the power of deep learning can allow them to function in all sorts of real-world complex situations that had previously been considered too challenging. Consider, for example, outdoor robotic applications – such as in farming. Outdoor situations are quite challenging for automatic machines, due to factors such as dramatic changes in light and alterations in the apparent appearance of various objects, as well as variations due to the seasons and changes in the weather. But all of this does not present such a challenge when employing deep learning – techniques such as convolutional neural networks can recognise features of interest in images even when there is a remarkable amount of variation present – they are surprisingly robust. We have got to a situation now where CNNs can recognise things of interest in scenes about as well as most humans (e.g. detecting the presence of a weed such as dock in grass when there is less than 5% of dock present). This is a great enabling technology and we are now on the cusp of a revolution in the potential for undertaking difficult and/or dangerous tasks automatically. You could say that deep learning is the game changer that will enable robots to successfully perform some of the kinds of tasks that people years ago thought they could, and may have seen them do in the movies, but who became disillusioned at the time when they realised that robots were not actually capable of them. I previously mentioned a tiger in a cage that was a bit of a disappointment, but the combination of AI with machine vision and robotics will represent some tiger - one that could break out of the cage and achieve a hell of a lot for humanity - if we allow it to do so.

So, AI breakthroughs are proving to be the key to technological revolution in the near future and the possibility of robotic solutions to many of the difficult, dangerous, or dirty jobs that men still have to undertake in the modern age is an encouraging and exciting prospect. But what of the more distant future? Of course, this is much harder to predict, but it is intriguing, since the possibilities that may open up might well blow your mind. For example, will there ever be a time when ‘Beam me up Scotty’ will refer to scientific fact rather than science fiction fantasy?

Sector	Invention
Electronic communication	The internet
Telecommunication	Smartphones
Transportation	Electric cars
Aerospace	Orbiting space station
Artificial intelligence (AI) and machine vision	Automatic face recognition (with performance similar to human capabilities)
Engineering	Simple robots able to undertake 1 specific task (e.g. vacuuming)
Engineering	‘Tele-operators’ enabling surgery to be undertaken remotely
Artificial Intelligence (AI) and Machine vision	Automatic scene interpretation, as needed for undertaking real-world tasks automatically (e.g. selective removal of weeds from grass)

Sector	Invention
Artificial Intelligence (AI) and machine vision	AI, sufficient to enable robots to sense and interpret their environments to allow them to act appropriately
Engineering	Robots, able to undertake more than one task, and agile and safe enough to interact effectively with humans
Transportation	Self-driving electric cars that can re-charge automatically
Aerospace	Human travel to Mars and beyond
Artificial intelligence (AI) and machine vision	Human-computer interaction capabilities enabling, for example, complete automation of a personalised shopping experience
Engineering, with artificial intelligence (AI) and machine vision	Robotic systems, capable of undertaking advanced diagnostics and selected surgical tasks in semi-automatic/automatic modes
Engineering, with artificial intelligence (AI) and machine vision	Automation of laborious tasks in various sectors (e.g. mining, farming, manufacturing, and so on)