Is Artificial Intelligence intelligent?
The other day my 3-year old son woke up crying at night and saying: Why mosquitoes are not coming?
Why on earth in the middle of the night in the middle of the winter did he want mosquitoes to come? – I have no idea. The way human mind works is still an enigma.
Artificial Intelligence is a hot topic nowadays. Everybody seems to be concerned. When I do a Google search for something, it always shows me banners about AI. Education, manufacturing, healthcare, financial services, etc, etc, etc. Google claims to use AI’s based RankBrain engine anyways.
Is it real or is it a marketing hype? I wonder why nobody markets an AI beer yet. Imagine, your beer can tells you: I hear you are coming. I am very cold and I see in your eyes that you want me !!!
The first time I dealt with this topic was some 25 years ago when I was asked to develop expert system software to analyze scientific articles in journals and make recommendations on further research. It was a great idea because the Internet was just making its first steps and a scientist needed to read all those journals in a library to be aware about current research.
In late 90s I co-ran a Canadian start up creating a knowledge management platform. We used a neural network module based on Kohonen maps theory. While it gave some fancy promises, it was still very far from perfect.
A prominent company doing a similar semantic search at that time was Autonomy. It was acquired by HP later. Promises were great but the outcome was a disaster. The saga still goes on: “The curious case of HP, Autonomy and the scandal that won't quit”
Certainly, much time has passed since then. We used mainframes at that time. A standard laptop is more powerful now. Microsoft released lots of Windows versions. New programming languages have appeared. Cloud computing is around. However, is there more intelligence?
Modern systems can process data gazillion times faster than humans. They can store more data going into exabytes. Modern networks can transmit data at 100GB speeds. But that is not intelligence. It is still old 1+1=2 calculations.
Intelligence means creating something new. Fuzzy logic is omnipresent in our brain. We work with incomplete information all the time. Yet, this did not prevent Leonardo Da Vinci from predicting a flying machine in XV century and Einstein from coming up with a theory of relativity.
80% of information comes to our brain through eyes. Neurons process light and the visual cortex of the brain creates images. Let’s check for example how far AI in image processing advanced in recent years.
First versions of video analytics became popular about 15 years ago. Motion detection was leading edge technology and it was great improvement at that time. Was there really intelligence? Not really. It was just comparing certain pixels in a frame sequence to determine a change in the scene.
Best companies in the CCTV industry can now track arbitrary objects in a set of camera streams or historic recordings. This means a human still needs to instruct the system which objects to search for and then the computer will work. These companies claim they use neural network algorithms but I don’t think such a system will work without an operator.
NVidia develops a new generation of products that will use AI algorithms in edge devices. They released their new Volta architecture to make big data calculations orders of magnitude faster. https://www.nvidia.com/en-us/data-center/volta-gpu-architecture/
That is all great and hopefully new CCTV cameras and IoT devices will become smarter. However, there is a long way to go before we allow an autonomous device going on its own in the outer world and be our trusted representative. Anyways, if a human makes a mistake, there is a court system to take corrective measures. How would you hold a robot responsible?
I think it will probably take another 50 years before we could see some true artificial intelligence in action. Maybe we will know how the brain works by then too. And then my son will be able to call his AI friend WatsonSon for a chat over a pint of beer …
As an Emotional Intelligence practitioner, I am intrigued by claims that Affective Systems themselves demonstrate "Artificial Emotion". In truth, they are just highly performing Cognitive processing systems that are tune behavioural responses in eg chatbots and robots. In the language of Brain vs Mind we stray into Neurology vs Psychiatry, a spurious and imho unhelpful bifurcation of form vs function that occurred just post-Freud. I would refer colleagues to the ever-helpful writings of Oliver Sachs (eg "The Man who Mistook his Wife for Hat") as an antidote to the narrowly reductionist discussions of AI currently dominating the hypo-sphere. To be honest, only when we can seriously discuss "Artificial Sentience" or "Agency", will I begin to believe that we approach Artificial Humanity. In other words where genuinely moral discretion based on a full set of emotional, as well as cognitive, perceptive faculties can be demonstrated. The dilemma facing us is that we may find that Singularity (that point where machine and human decision-making capabilities are merged) sooner than that. put another way: we may have surrendered our moral and ethical responsibilities to many machines way before they can truly handle them. As a final consideration, we should ask: what kind of species creates sentient, morally aware creatures purely for the purposes of enslaving them? We urgently need an authoritative [UN?] body to codify mankind's shared constraints (implementable ethical guidelines), and the sanctions for developers and manufacturers that infringe them.
Muriel Touaty
Karl R K., Denis Vlieghe
About visual processing (in the nervous system): there is more inputs to the LGN (the first relay from the retinas en route to the visual cortex) coming from the visual cortex itself (and from fairly high areas, in the hierarchy that is), as well as from other parts of the nervous system ... then from the retinas! And remember: we dream (and it is largely visual, unless you are blind from birth; but the blind can draw objects they conceive of (dream of?)); in fact one could say we hallucinate the world (i.e. we create it) all the time, and we rectify/make adjustments from sensor inputs (like the eyes). Seems to me that "top-down" AI processes need more attention (again).