AI, Machine Intelligence, Thinking Machines and Smart Machines

A robot with rudimentary social skills
(Kismet robot@MIT  Museum)

Today we are experiencing a global media tsunami around the topic of AI and machines that can think, supersede the human brain, replace our jobs and even being a threat to the humankind (ref great minds like Stephen Hawking and Elon Musk).

Is it true, are we on the brink of developing and releasing machines that can think for themselves? Machines that are truly intelligent and that has a mind of itself?

If you ask the vast majority of people working in the tech industry, I guess you will have a vague and uncertain answer, but if you ask the top notch research people that are working on AI and developing cognitive systems, you will have a load and clear no, not in any foreseeing future, if ever! (ref “What to Think About Machines That Think” edited by John Brockman):

The media just dont get it, they are mixing all terms and definitions and focuses on imaginary robots that supposedly can out-think the human. What we do have experienced, for many decades already, are programmable machines that replaces boring, tedious and simple jobs by ticket machines, road tax systems, ATMs etc and lately more advanced and user-friendly internet banking systems replacing thousands of bank tellers and soon even traders (already happening), accountants and real estate agents.

So, in short, what happens is that the machines are getting more resources and capabilities to process more data, i.e. big data. The finance industry is a good example where we will see more and more usage of big data being processed by powerful machines with the help of machine learning. The finance analysts are spending most of their time investigating data for potential investments which are then fed into potential trading opportunities. The analysts are thus processing as much data as possible before an investment decision is taken.

A computer will be able to process much more data much faster and reach a recommended trading that in itself easily can be automated. Similar use-case you will find in life science and diagnostics. Computers with machine learning algorithms will be able to process much more data (medical journals) much faster and reach a recommended diagnosis almost instantly.

But is this thinking machines? Is it even Artificial Intelligence?

Machine learning is basically about training a prediction model enabling it to refine its prediction by providing training data. This training is an exhausting process as you need to train the model to understand all possible correlations of the data points. In the case of the financial analyst, we need to train our stock trading model all the correlations between the financial data points that the analyst would claim are relevant for the potential trading. This is an exhausting process and thus this is the main reason why we still have human financial analysts and traders, but that will not continue for long as gradually machines will be trained by machine learning to accommodate this exhausting process. But still, all these “thinking” machines are just trained extensions of our human minds.

These machines and their software are not about to leap beyond us intellectually, much less turn us into their slaves. They are simply just doing the instructed tasks much faster. 
I realise that you could argue that we are already dependant (slaves) of our handheld smart devices like iPhone etc, making us into trans-human and thus maybe there is a short step from these “smart” devices to a dependency of the “thinking” machines that act as financial analysts, traders, real estate traders etc., but is this scary? No, but very beneficial and cost efficient.

Do we face singularity - the moment when the machines surpasses our intelligence? No, not any time soon, the fact remains, these machines are going to continue being dependent on the human programmers for any foreseeable future - however that also includes the potential of programmers (human) making errors and introduce bugs that could have some serious consequences.

Einstein is quoted as saying “Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe”

What if the machines do the programming? Are we able to teach the machines programming? Not really, but we may use machine learning to assist the machines to do some extensions of its own algorithm and as there is research going on in deep learning with neural networks scaled twelve layers deep, there might be some interesting development in the near future. Deep learning has already revolutionised object- and speech recognition and achieved significant progress in melanoma recognition in dermoscopy images, but is this silicon-brain a thinking machine?

Despite these amazing progress, what about the remaining capabilities of our carbo-brain? Our carbo-brain have billions of neutrons in cortical hierarchies ten layers deep and it has the capabilities to, not only acquiring knowledge, but has a mind, thought, experiences and senses.

“The gap between our best computers and the brain of a child is the gap between a drop of water and the Pacific Ocean” (Carlo Rovelli, Theoretical Physicist)

And just one last point, these “Intelligent Machine Systems" are often referred to as Cognitive which is also a bit provocative given that the adjective is defined as “of or relating to the mental processes of perception, memory, judgement, and reasoning” - and I don’t think any of these software systems fits the bill - at the best, they are Kismet robots with rudimentary skills.