After the Second World War, technology boomed a lot, allowing humans to visit the moon, talk to each other from far distances and, most importantly, automate their work using machines. Alan Turing was one of the mathematicians who worked on AI, focusing on machines thinking like humans. As time passed and AI advanced through the training of datasets, many people now believe that it will replace humans in the upcoming years. But can it? Can it work like an actual human brain and become Vision from the movie Avengers: Age Of Ultron? This blog will ponder deeply on the question.
Origins of AI
The first AI started in the 1950s with numerous scientists, mathematicians, and philosophers who worked on making computers solve complex problems and “think” like humans. Alan Turing, a British mathematician, published a paper titled Computing Machinery and Intelligence, where he suggested that machines should solve problems just like a human by using information available along with logical reasoning. At that time, computers only executed commands and did not store them, thus they were not supposed to remember the things they did.
Five years later, Allen Newell, Cliff Shaw, and Herbert Simon initialized it with their program Logic Theorist which imitated a human’s problem-solving skills. This led to more research about AI and produced similar AI programs, especially Weizenbaum’s ELIZA, which was the first virtual assistant that passed the Turing Test.
From 1957 to 1974, computers became cheaper and more advanced in terms of storage and speed, leading AI to advance too by using machine learning algorithms (these are algorithms used in identifying images and finding relations between them).
AI vs. Human Brain
To make AGI (Artificial General Intelligence), we look upon the only thing we have, i.e. the human brain. To achieve this, the AGI reads a provided dataset and uses machine learning as a statistical method. In most movies, AI thinks like humans (e.g., Tony Stark’s Jarvis). But the thing is, AI still has a long way to reach there.
Common sense plays a vital role in illustrating the obstacle stated above. For instance, if you are driving down a street and you see a child playing who goes behind a parked car, you will automatically be aware that he will suddenly emerge on the road. This is because you use your experience through EQ, making it difficult for an AI to notice the problem as it is an unpredictable event, and it only uses intelligence, a.k.a knowledge.
Challenges and Limitations
As mentioned in the previous heading, AI lacks common sense because it requires experience instead of knowledge. The notable examples are:
- The existence of time and its ability to impose an order on actions in the environment.
- A chain of events, i.e. causes can predictably lead to effects.
- Actions that a person (or AGI) takes can influence the future, which may impact the person.
This is a huge drawback for an AI because common sense eliminates ambiguity in everyday life. For example, AI cannot understand the following joke:
“I know a man with a wooden leg named Smith.”
“What’s the name of his other leg?”
Furthermore, the absence of common sense also leads to the absence of reasoning, depriving AI of understanding cause and effect over time. From the human point of view, we cannot have an idea of how the AI works. And when a problem arises, we cannot identify and correct it.
Societal Considerations
“The best and most beautiful things in the world cannot be seen or even touched. They must be felt with the heart.”
— Helen Keller
Apart from our five senses (seeing, hearing, touch, smell, and taste), emotions also play a vital role in our lives. It helps us understand feelings which we do not get through mathematical equations, even if it is about sensing heartbeats. Although we do receive the answers for “How”, “Why” becomes partially answered. Let’s take an example of Sherlock Holmes: he is a very intelligent man who can produce breathtaking results through strict observations, but he cannot understand how John Watson is feeling in different situations because he cannot understand human nature as these types of information are not received through books.
In the case of AI, this can result in highly intelligent machine beings having an average IQ quite larger than humans, and solving multiple complex problems, but they will always be sociopaths, or worse, without emotions as depicted in numerous science fiction movies because an artificial neural network can never replace the human brain. However, they can try to mimic our response in different situations explained in the next heading.
Attempts
Mimicking humans is the closest way possible for an AI to act like them. Sentiment Analysis plays a key role by calculating the ratio of positive to negative words in a sentence, conversation, etc. and then outputting the overall sentiment of a person. However, the drawback is this cannot function properly if you were being sarcastic the whole time. But there were some attempts to completely replicate the human brain.
EPFL’s Blue Brain Project is a Swiss brain research Initiative led by Henry Markram. This project successfully built the first copy of the mouse brain, opening the gates to understanding the multi-level structure and function of the brain.
Another attempt was Spaun (“Semantic Pointer Architecture Unified Network”) an architecture made by Chris Eliasmith from the University of Waterloo Centre for Theoretical Neuroscience. It consists of 2.5 million simulated neurons organized into subsystems that resemble specific brain regions. It was made, in Nengo, and can count numbers, and even write them if given a robotic arm.
Future
Considering the blockades preventing AI from simulating the human brain, developers should work more on a more biologically plausible structure rather than just training with datasets. To counter this, a more holistic approach is needed showing the relationship between things like vision, language, and robots instead of reading words from the data.
It is also important to note that AI may not be allowed in advancing further, because it can use personal data to train itself, something the next web generation Web 3.0 will take note of.
Conclusion
While today’s AI, especially ChatGPT, has impressive capabilities and produced outstanding results, it can never achieve AGI until it follows a child’s thinking process and starts displaying the same common-sense knowledge, we use to experience the world around us.
Written by Umair Shakeel
