Making a Machine Intelligent — What Can AI Do?
Humans are not only survivors who have evolved but are also very curious animals. Technological advancements did not begin just with the creation of computers or computers becoming available for common people around the world. What is technology? The dictionary definition is “the application of scientific knowledge for practical purposes, especially in industry”. When do you suppose this began? Back in the ancient times, when man already started finding ways to satisfy his needs and to protect himself.
“The Paleolithic period began with the advent of hominid tool use. Hominids, such as Homo erectus had used simple wood and stone tools for millennia, but as time progressed, tools became far more refined and complex. Perhaps as early as 1.8 million years ago, but certainly by 500,000 years ago, humans began to use fire for heat and cooking.” — https://en.wikipedia.org/wiki/Human_history
The Way Advancements Happen:
Finding ways to create and use tools to cook, to hunt, to protect, all these are examples of humans effectively making use of their skills, and how far we have come — this is proof of the unquenchable human need for more advancements. This is why we regard ourselves “the most intelligent species”. But the smartest are usually rare, the common human mind, it could be curious or not, but does it offer any change to move forward? Again, we don’t require that the majority be highly smart. Scientists, researchers, and engineers do the findings and those with political and financial power get to make decisions on how extensive the reach of these innovations would be in their times.
Far-fetched Dreams That Have Become True:
Throughout history, people in multiple different fields have dreamt of what the majority would call “impossible”, and some dreams have been just that: dreams, and some dreams on the other hand have turned out to be real. The British physicist Robert Hooke is said to be the earliest researcher in acoustic string phones (1664–1685). And Alexander Graham Bell is said to have invented the first telephone in 1876. (It took two centuries to build a telephone from when the idea began making sense?)
Next, what about rockets? “One of the first recorded rocket launchers is the “wasp nest” fire arrow launcher produced by the Ming dynasty in 1380.” For centuries then, rockets were extensively researched and used in wars in different parts of the world. In 1920, Professor Goddard is said to have published a book on sending a solid-fuel rocket to the Moon. His ideas were both praised and criticized heavily. Yori Gagrin was the first human to travel into outer space (he completed his first orbit around Earth in April 1961). The history of rockets should tell us that 1) technology is mostly controlled by those with political and financial power even if it’s the scientists who really make innovations happen, 2) technology could very well be knowingly misused with evil intentions, and 3) humans always want more, possibly because they see moving forward as a sign of being “powerful”.
Can Machines Think?
We have several examples like this in the history of technology, how it takes centuries of research for an idea to turn into powerful, revolutionary reality. What about artificial intelligence? Can Machines Think? Alan Turing, in 1950, explained his thoughts saying that he considers questions like “what exactly do thinking and intelligence mean in the context of a machine” absurd, saying that maybe after 100 years from then, “thinking machines” and “learning machines” would be discussed on a much more advanced level. He also said, “I believe that in about fifty years’ time it will be possible, to programme computers, with a storage capacity of about 109, to make them play the imitation game so well that an average interrogator will not have
more than 70 per cent chance of making the right identification after five minutes of questioning.”
Artificial Intelligence has been a subject of great interest for at least several decades years now, and the media has made fiction in which AI can either be as intelligent as human beings in all ways or can even outsmart human beings easily, encouraging a vast majority of people to expect too much out of machines as well. That brings us to the question,
What Can Artificial Intelligence Do?
In the context of human beings (the supposedly most intelligent species we know as of now) and our understanding of “intelligence”, we have so much to offer, skills in several different areas, a strong ability to understand relationships, context, and patterns, understanding social cues and acting courteously, and even misusing intelligence. In the human context, “intelligence” could mean a lot many things, a combinations of some of these aspects or the total sum of it all.
In the context of Artificial Intelligence, it is broadly classified into two categories: artificial narrow intelligence and artificial general intelligence. What separates AI from normal computers is that an AI can think and act on its own without being programmed. “Narrow AI”, as the name implies, is AI that can display human intelligence only in certain areas and has limitations. For example, consider a machine playing chess and beating a chess wizard at his own game. Other examples of Narrow AI would include chatbots, predictive analytics (making predictions in the field of medicine and stock market, for example), facial recognition (for authentication, for example), audio books (text-to-speech translation), and such.
What About Virtual Assistants and Conversational AI?
Some might want to believe that Amazon’s Alexa, Apple’s Siri, and Google Assistant are all examples of artificial general intelligence. But are they really? While these are all excellent examples of NLP (Natural Language Processing), artificial general intelligence is defined or understood as the ability to learn and perform any intellectual task that a human being can. And sadly, we are nowhere close to that.
While growing companies and researchers are actively working on making machines more intelligent, most scientists would agree that AI is not as intelligent as we would wish it be, at least not yet. Cathy Pearl says that conversational AI is still stupid. Some products can boast of learning on the go (take Sophia for example) and “context switching” — which is extremely important in conversation and understanding (very important areas of intelligence). Still, when it comes to performing tasks, they are very, very limited.
I spoke to my Google Assistant now. Initially, she told me a joke, then I asked her, “How do I know if my understanding of artificial intelligence is right or wrong?”. Her initial response was, “Sorry, I don’t know if you were talking to me, so I’ll disregard what you said.” See, the machine can’t even identify when it is being spoken to and when not. When I repeated the question, opening a fresh, new conversation, this time, the response was, evidently, “this is what I can find”, showing me a list of Google search results. I don’t need Google Assistant to do this; I can do the Googling myself, right?
Even a kid is capable of having a more humane, sensible conversation if you and I would initiate talks. Why? Because a human brain can easily and very quickly pass information (information collected from various areas at once) to help us act the way we do. Even in areas where we lack knowledge, we don’t just shut down; we think and act. While computers are highly equipped with knowledge and now have the power of “big data” (which was not there earlier), they still can’t comprehend exactly, think using several different areas of intelligence at a time (narrow AI is narrow!), and act accordingly. As of now, machine learning is the height of “artificial intelligence” we have. There are different success stories and failures in the field of machine learning as well, but that is that.
Inventor and writer Ray Kurweil strongly believes that we will soon be having neuroimaging and computational power so effective that we would have machines mimicking human brain’s ways of thinking.
The Theory of Embodied Cognition:
According to the theory of embodied cognition, an embodiment of an animal is a must-have to cognitively think like the being. This is one area where cognitive science becomes really interesting. A tiger of leopard does not go to school or is not taught numbers or physics, yet it can effectively hunt its prey. You could broadly say that it learns all this as part of evolution and from observation, but according to the theory of embodiment cognition, the wild cat is able to naturally know from where to pounce, at what rate to puce, how to cover the distance, how to hide and not scare the prey away, and when exactly to pounce. All this is computation, comprehension, and excellent use of other cognitive tasks, which according to this theory, the animal possesses because it’s embodied the way it is. And in IQ tests conducted on animals, why don’t all non-human animals rank alike? This theory’s answer would be that each animal’s cognitive abilities goes hand in hand with its bodily features as well.
Simulation models were used to try and make machines behave like small animals, like insects would, and even after years, that has not been possible. There are all basic cognitive qualities we are talking about; we have not gone anywhere near philosophy, sentience, and wisdom yet. So for now, the truth remains that nothing promising seems to happen anytime soon in the field of artificial general intelligence.
