As all our forum members who read the introductions thread know, I am currently studying Artificial Intelligence (AI) at university... what some of you might not know is what it is. (If you are not one of those people you may find this article a waist of time to read, or still of interest... If the former please wait until I publish something more interesting.)
What is machine intelligence?
So when I tell people I study AI the number one response I have gotten is "is that aliens?" and the answer is NO. For this reason I've decided first to define it.
Artificial intelligence is the field devoted to making "computers" (a computer can be any digital system) learn. It seems intuitive however a lot of people seem to not quite make the link in their mind. How some people think artificial means "aliens" I am not sure, as artificial by definition means, man made... not of nature.. Aliens are not man made!
What kind of things do you do?
Well here I should make a distinction between Artificial Intelligence and Machine Intelligence as they are often both bundled together and I may well end up posting about both
- Artificial Intelligence: This is the study of replicating the cognitive process via digital means to create a system that learns - Mainly when I post about this it will be in relation to Neural Networks which are a collection of programmed "brain cells" (neurons).
- Machine Intelligence: This is the study of creating intelligent robotic systems, but not necessarily using artificial means (a lot of the time artificial means are used though). An example of an intelligent machine that does not use artificial means of intelligence is a biologically intelligent robot. In the University of Reading, we use rat brains to control robots. All the inputs and outputs given to / from these brains are digital information directly from or to the robot, however the learning is done via biological means (the brain grows and the neurons develop over time, learning tasks that it repeats during infancy).
Currently my studies are in Artificial Intelligence in the form of Neural Networks and I will post more about these soon, but the premise of a neural network is that a collection of programmed "neuron" cells are used to learn and solve tasks. There are many ways to achieve this and I will only post about a few.
What can artificial intelligence be used for?
Systems that implement AI in its many forms have endless numbers of usages. For example Neural Networks are used in weather forecasting, handwriting recognition, voice pattern recognition and data analysis. For example CERN uses neural networks to categories collision events and only saves those which the network suggests are "interesting". This still produces terabytes of data however is far more manageable than if they where to store all collision data generated in the detectors.
Other types of AI can be used in production lines, ranging from basic repeated task code to control robotic construction arms to quality control testing (here if the output is 1 the item is good, otherwise it has failed). AI has even made it's way into day to day life, with services such as Siri of Google's voice search.
What are the big questions in the field of Artificial Intelligence?
Perhaps more interesting sometimes than the material itself is the "big" questions in the field of AI. These are very philosophical and ethical questions that researchers in the field must be aware of and sometimes try to solve. I will go through a few of them...
Are machines really intelligent?
This is a question that a lot of people ask, is there such a thing as Artificial Intelligence? The answer to this question not only depends on your definition of intelligence, but how you actually interpreted the question. For example, some would say anything that possesses the ability to learn is intelligent. As such neural networks are intelligent. Others do not agree! Alan Turing attempted to answer this question himself back in the early days of computer development. Turing noted that intelligence comes in different forms, and to program a computer to play chess at a high level was a trivial task... it does not show intelligence, simply computation. For Turing the test of intelligence was "Can machines think?" or rather as he later phrased it "can machines do what we do?" - As mental and physical activities are not entirely separate issues Turing developed what he referred to as the "imagination game" (You can look this up yourself.) This was later developed into what is now known as the Turing test and has been considered the true test of intelligence for many years. Turing predicted that computers would beat this test by the year 2000. The latest estimates put the date at 2029.
Due to the definition given by Alan Turing of what artificial intelligence should be, smaller levels of intelligence are generally ignored. Being able to predict the weather with a 99% accuracy using a neural network is not considered to be a great achievement, nor an act of an intelligent system, even though it is. The human centered approach to intelligence still very much drives forward research in artificial intelligence and the Turing test is still the standard test any "intelligent" system should pass.
Should human brain-cells be allowed to be used to create intelligent systems?
This is an issue that is very much under discussion at the moment. As humans believe our brain to be unique and as human nature is bias toward our species, unlike when other animal brain cells are used in experiments, human brain-cells cause a rather large discussion. Should we be allowed to grow brains from human brain-cells if they are simply going to be used to control a robot? This is a hard question to answer affirmatively without upsetting religious and ethical groups, for the sanctity of life is important and by growing a human brain, unless there is a sure knowledge that it will not be self aware, this usage of it could be considered to be cruel. On the other hand, as we do not know if the brain is sentient and we assume that it is not, by growing a human brain and using it to control systems, this gives an unprecedented opportunity to study the development and workings of the brain. Not only does this research further the advancement of artificially intelligent systems, but also the research can be used for medical purposes. If we know what causes issues with the brain, we may get a better understanding of how to solve them.
If a machine is considered intelligent should it be granted the same rights as a human?
This question is often ignored, as computers are seen very much as the "work horse" of the 21st century, however if an artificially intelligence system was developed, that could pass the Turing test, would that be enough to consider it as having equal status to humans? The answer to this question is hard to predict as the stage of intelligence here has yet to be achieved, however, once it is, what will happen to the system generated. The likely hood is, by looking through human history, an artificial intelligence will be a slave to humanity until the general public can be convinced that it is equal.
Should the military use artificial intelligence in drones?
This is a question I have very strong views on. The military are currently developing AI systems to pilot and control drones. These automated systems can perform a variety of tasks from providing visual surveillance of an area to firing rockets in more modern versions. To me this is not acceptable. Relating this to the former question on artificial intelligence rights, military use of AI will give systems a bad reputation, as any instances of "miss-fires" due to mistakes made by artificial systems will cast negative views in the public eye. However these systems should not be put in this position of control and responsibility in the first place, if people want to kill each other that is fine, but please leave our computers out of it, they should never be made to take a side.
This is the end of the philosophical debates, I shall not be commenting on them further - If I do they will have their own post and will have gained relevance that makes them deserve talking about! Keep an eye out for the next article on neural networks!