“AI is taking over the world.” “AI is stealing our jobs.” These sayings are more common in this decade. They have somehow caused overrating fear and objection toward AI.
In reality, the myth that AI can defeat humans and take over the world is quite surreal and won’t happen anytime soon. ‘Why?’ You might question. The answer is in the book Artificial intelligence vs. Children who can’t read textbooks (AI vs.教科書が読めない子どもたち) written by Arai Noriko, a Japanese researcher in mathematical logic and artificial intelligence.
This book will correct the misunderstanding and explain why AI is still not and won’t be smarter than humans anytime soon. It also addresses how far are AI abilities now, which areas of jobs will be stolen, and what skills we need to survive.
Sertis already summarised the main idea of the book here. It takes only less than 10 minutes to correct your lifetime misunderstanding. If you want to discuss anything about the book, please leave a comment down below. Let’s get started!
The word ‘AI’ (Artificial Intelligence) commonly used nowadays actually has 2 meanings.
One is a literal meaning which is ‘Artificial Intelligence’. It means the machine that can imitate or further develop human intelligence and reproduce itself without our help. This thing hasn’t happened yet. Even the most advanced AI ability is still far behind human intelligence.
The second meaning is AI as AI technologies. They are technologies that we are developing in the present, such as Machine Learning, Image Recognition, and Natural Language Processing which, at certain levels, are expected to build the ideal artificial intelligence with the abilities described above.
To conclude, the word ‘AI’ that we use now indicates technologies in the second meaning, but we misunderstand it to be artificial intelligence in the first meaning. We assume that AI is now smarter than us and create the myth of AI taking over the world, which in reality, it doesn’t seem close.
Arai Noriko has been trying to correct this understanding and enable people to see what AI is truly capable and incapable of. She developed an AI robot named ‘Torobo’ and started a project to get it into Tokyo University, the best university in Japan.
With 10-year effort, Torobo finally got a 75% score on the history exam and was the highest 20% of all students on the mathematical score.
To do the history exam, Torobo used the method of cross-checking all the choices by using multiple keywords. It would search those keywords on the internet, select the website that the keywords appear the most, and compare the choices with the information on the website to see if it was right. For the mathematics exam, Torobo used Deep Learning and Machine Learning to turn questions into equations, then solve them with Computer algebra.
However, 2 subjects have made little progress in all those 10 years, Japanese and English. Torobo gets only less than 50% scores in these 2 subjects. The language exams consist of grammar, vocabulary, conversation, and reading comprehension parts. We can use data and rules to train Torobo to process the grammar and vocabulary parts. However, for the conversation and reading comprehension parts, which need common-sense knowledge and understanding, they are impossible for Torobo.
In summary, The results described above show that the limitations of AI are incapabilities of understanding and common-sense reasoning.
Learning about the limitations of AI, you all may question that if AI can’t understand us and doesn’t have common sense, how do machines like Google and Siri that can answer our questions work?
The answer is that they select keywords from our question and analyse the statistics to find the best result. Try asking Siri to find ‘Italian restaurants with good reviews’, and you would get plenty of recommendations. Then, try ‘Italian restaurants with not good reviews’, you would likely get the same results as the first search. It is because the machine works by guessing keywords and statistics. Few people use the phrase ‘not good reviews’, so the results are not accurate.
What about Google Translate? It uses language models and numerous statistical data to learn how to select appropriate words. The machine tries hard to imitate our natural language. They don’t understand it. In Google Translate, users can suggest a better translation to the system, and it also learns from that.
What about AI that writes articles or composes songs? It is us who teach them with a large number of articles and songs. Then, it randomly chooses what to write by mimicking our originals.
So do Machine Learning, Deep Learning and Image Recognition. These technologies all use our training data as a source of knowledge. They memorise what they see or learn. If we ask them to recognise what they haven’t seen? The answer is a big no.
You may now realise that AI can’t do more than calculating and memorising. Every ability that AI has nowadays is generated from these 3 ‘languages’: Logics (If A=B B=C), Probabilities (The chance to throw a 1 with one die is 1/6), Statistics (Prediction from the past data).
Anything that can’t be calculated is beyond AI ability. AI can’t understand the simple sentence as “I Love you”, because the concept of ‘love’ can’t be turned into math or an image either.
Human brains are complex. We use understanding and common-sense reasoning in every minute of our daily life. To be as smart as us, AI needs to imitate our brains. To mimic our brain, AI needs to calculate common-sense reasoning and understanding. It is impossible because there is no absolute equation or formula that can explain our thinking. The development of AI and mathematics haven’t overcome the wall of common sense. That’s why AI won’t take over the world, at least anytime soon.
It’s time for the bad news. Even though AI can’t compete with us in terms of understanding and common-sense reasoning, when it comes to calculating, AI totally wins.
Torobo’s math score can get it into 70% of universities in Japan, including top private universities.
Only 20% of students get better scores than Torobo which, in other words, means that 80% of students are at the same level as Torobo.
Researches show that calculating and memorising jobs, jobs that mainly rely on statistics, and routine jobs, such as credit analysis and data entering, might be replaced by AI to reduce costs and errors.
Some people may argue that if AI steals your jobs, then create new jobs that AI can’t do. Sadly, the truth is what AI can’t do is what most of us can’t do either.
Skills that AI doesn’t have are reading comprehension, agile thinking, and common-sense reasoning. Okay, we all might have common sense, but most of us don’t have reading comprehension and agile thinking.
Reading test skill results, a test of how much people understand simple messages as in newspapers or textbooks, show that only 30–50% of Japanese high school students have good reading comprehension skills. Most students memorise what they learn without understanding. Their ability can be considered equal to AI, so they can’t do what AI can’t either.
The important question is how to survive when AI steals our jobs? It’s not that hard. Just look back and improve basic skills that AI doesn’t have such as reading comprehension that we once overlooked.
Our common-sense reasoning and comprehension are what make us unique. We also have creativity and agility. These qualities make us different from AI that can only do and process what humans programmed it. It can’t understand things or invent new things by itself.
As a result, in the era that AI might take over 50% of our jobs. We need a plan to survive. Start by exercising our comprehension skills to try to understand the conditions and pain points of society. Come up with new jobs that creatively solve the problems. Develop new products and services that meet people’s needs. Communicate the value of understanding and humanity through your jobs. And last but not least, stay agile.
Written by: Sertis
Originally published at https://www.sertiscorp.com/
