Original article was published by FutureImpact on Artificial Intelligence on Medium
[AI] Pieces of the puzzle
In the previous blog on general artificial intelligence we spoke about reinforcement learning, natural language processing and common sense reasoning. These are all pieces of the puzzle of intelligence. But how big is this puzzle? And when is the puzzle complete?
The philosopher Ludwig Wittgenstein once argued that words find meaning in their usage. As a start he tried to define the word “Game”. Does it need players, does it need rules? Every way you define it, there are always cases that don’t cover the definition. In the real world meanings are much like family resemblances he argued. The faces and characters of family members look-a-like, but they are not the same.
This is also very true for the meaning of intelligence. Originally intelligence was more defined to its capacity for logic, reasoning and understanding. Especially with research in artificial intelligence we found out that for humans seemingly simple tasks also require intelligence.
When you play a game you not only need to understand the rules, but also develop creative strategies, learn from mistakes and read the opponents intention. But there is also a physical level. We need to process what we see and let our body perform actions. This doesn’t count just for games, but for almost all things we do like cooking, cleaning or cycling.
This makes intelligence much broader than just reasoning. We can add the following capacities:
- Physical actions
- Emotional knowledge
- Spatial knowledge
- Critical thinking
A soccer player like Christiano Ronaldo has for example a high “Football Intelligence” which means he is creative and plans ahead in the game. Accordingly he is able to react to what he sees and translate this into actions on the field. On the field he needs to react to opponents, collaborate with his team mates, read other players movements, solve problems and so on. As he is critical of himself, he continuous improving himself. When you think of this all, soccer becomes very complex.
In this broad sense we are not just talking about intelligence as a human trait, but as something that animals or computers can have. A monkey is thus not intelligent, because it can reason or solve simple math problems, but by climbing in trees, getting the right food and protecting and taking care of other group members. Computers are not intelligent, because they can calculate, but because they can predict the weather.
Together all capacities form the whole puzzle. To become intelligent it’s not enough to posses one piece of the puzzle. Just performing one task excellent isn’t considered intelligent. Whether it’s a human performing every time the same thing at the assembly line or a computer doing a calculation very fast. Even when only one part of the puzzle is missing this is already considered strange. Much like a scientist without learning abilities or a professional soccer player without communication skills.
There are also exceptions that missing a piece can make other parts of the puzzle clearer. Stephen Hawking missed large parts of the physical piece, but maybe this makes his reasoning and logic stronger. Or a dancer may not have the highest IQ, but excels in creativity or physical performance.
Current AI focuses only on one or a few pieces of the puzzle at the same time. Like Stephen Hawking or the dancer it can achieve excellent results in these area’s. When performing reinforcement learning algorithms, the AI program is playing against itself. It’s using learning, problem solving and critical thinking. It’s however not able to communicate or act on what it sees.
You can communicate with another AI like a chatbot, or use computer vision to recognize objects. But these are other parts of the puzzle. For every part of the puzzle of intelligence we can nowadays find a related field in artificial intelligence:
Behind every field there are thousands of scientists writing millions of papers and lines of code. Of course the situation can become complicated as most field are interrelated, cover multiple topics or are subdivided.
Take for example machine learning:
So the 16 pieces puzzle we gave is an oversimplification. Real world puzzles on intelligence contain many more smaller pieces. Still, based on this simplification we can now map the several challenges and methods in a more sophisticated way. AI researcher Francesco Corea created a map where he created six AI paradigms on the X-axis. These are the various types of methods which AI scientists use to solve their intelligence challenges. The Y-axis shows the various types of intelligence:
Because scientists still need to solve fundamental problems there is a narrow focus on these fields. The scientists are driving through a tunnel in search of the light. To solve one piece of the puzzle.
When technology of certain pieces becomes a commodity we are able to combine them. For example robots can have computer vision which can recognize objects and people’s emotions. It also has a common sense reasoning build in and can talk with text-to-speech like a chatbot. Because of machine learning it can digest what the robot sees and hears and take action accordingly.
Like the internet, the foundation will be laid be scientists, but the combinations and applications will be done by startups and guided by you and me. The people who use artificial intelligence, just like our usage gives meaning to our language.