Intelligence is what you use when you don’t know what to do. – Jean Piaget

Part 1. Gaining processing power
Part 2. Becoming intelligent
Part 3. Interacting with humans
Part 4. Building useful programs
Part 5. Future of human programmers

Computers blindly follow our instructions. They are much faster than humans, but still computers are stupid things dependent on our algorithms and knowledge how to solve problems.
Even huge processing power is not enough to start programming. Non-trivial solutions require understanding of ideas, problem solving, learning from experience and much more – everything what we can define as intelligence. Can computers become smarter than human programmers?

What intelligence is required for building programs?

What is human intelligence?

We have pre-wired capabilities: large and effective brains, genetic development of specialized functional areas (vision, speech, reasoning, etc.). However, we gain our intelligence mostly through learning and experience.
Jean Piaget defined four stages for human intellectual development:
1. Sensorimotor stage: from birth to age 2 years (children experience the world through movement and senses and learn object permanence)
2. Preoperational stage: from ages 2 to 7 (acquisition of motor skills)
3. Concrete operational stage: from ages 7 to 11 (children begin to think logically about concrete events)
4. Formal operational stage: after age 11 (development of abstract reasoning).

Nobody directly puts intellect in our heads. We construct our own intellect through complex interactions with the world and other people all our life long.

Highlights:

• Memory is the basis for all intellectual activities. Memory mostly encodes the relationships between things than the details of the things themselves. It maybe exists only to make good predictions about future
• We are not only using senses and react on external stimulus (keeping car on the road), but constantly constructing an internal model of the outside world and predicting how things behave (our and other cars dynamic).
• Our emotions are important as they are brain states that quickly assign value to outcomes and provide a simple plan of action. Therefore, emotion can be viewed as a type of computation, a rapid, automatic summary that initiates appropriate actions.
• Other species are hardwired to solve particular problems, while our ability to abstract allows us to solve an open-ended series of problems.

• Interaction between senses and memory allows us to construct a qualia (properties of sensory experiences). We could create a scene in our mind and make connections with past scenes.
• With ability to narrate and long-term memory we connect conceptual systems and develop semantics (meaning) and true language. Now we can become conscious of being conscious.

Building intelligent computers.

Computers could become intelligent in several distinct ways:

1. Simulation – scan and copy brain structure and emulate with computer how brain functions.
2. Singularity – in a system complex enough, consciousness and intelligence could simply pop into existence without our participation. Star systems emerged from simple particles, biological life emerged from simple chemical elements, and new intelligence could emerge from huge number of computers and their networks.
3. Symbiosis – extension of human capabilities with machine intellect (transhuman and posthuman) using nano-, bio- and information technology advances.
4. Artificial Intelligence built by humans.

AI is the most realistic way. History of AI had few winters, one of each still continues. Despite failed grand expectations, AI brought many practical results and applications:

• Postal Services use AI to help deliver the millions of packages that pass through its transportation network every day in the most efficient way possible.
• Telecoms operators use AI to establish the quickest connections for phone calls through their networks or to retrieve web pages speedily from the Internet.
• Manufacturers and retailers use AI for optimization of their supply chains.
• Financial organization use AI to organize operations, invest in stocks, prevent fraud, and manage properties. Algorithms carry 70 % of foreign currency trades.
• A medical clinic can use AI systems to organize bed schedules, make a staff rotation, and provide medical information.
• Neural networks are used in homeland security, speech and text recognition, medical diagnosis, data mining, and e-mail spam filtering.
• Robots have become common in many industries. They are often given jobs that are considered dangerous and exhausting to humans. General Motors uses around 16,000 robots for tasks such as painting, welding, and assembly.

There are 2 main school of thoughts for building AI: Classical Theory of Mind and Connectionism.

• The classicist believes that mind is a symbolic processor, where strings are produced in sequence according to the instructions of a (symbolic) program. The mind operates by performing purely formal operations on symbols.
• The connectionist views mental processing as the dynamic, bottom up and graded evolution of activity in a neural net. It models mind as the emergent processes of interconnected networks of simple units, usually neural networks.

Conventional AI (based on classic theory)

• Expert systems (knowledge based): apply reasoning capabilities to reach a conclusion. Expert system contains some of the subject-specific knowledge, and contains the knowledge and analytical skills of one or more human experts.
• Case based reasoning: stores a set of problems and answers in an organized data structure called cases. A case based reasoning system upon being presented with a problem finds a case in its knowledge base that is most closely related to the new problem and presents its solutions as an output with suitable modifications. It is used in Recommendation and Decision Support Systems, Help Desk, medicine.
• Bayesian networks: used for modelling knowledge in bioinformatics, medicine, engineering, document classification, image processing, data fusion, and decision support systems.
• Behavior based AI: a modular method of building AI systems by hand.

Computational AI (based on connectionism)

• Neural networks: trainable systems with very strong pattern recognition capabilities. Neural net contains simple units, each unit’s activation depending on the connection strengths and activity of its neighbors, according to the activation function.
• Fuzzy systems: techniques for reasoning under uncertainty, have been widely used in modern industrial and consumer product control systems; capable of working with concepts such as ‘hot’, ‘cold’, ‘warm’ and ‘boiling’.
• Evolutionary computation: applies biologically inspired concepts such as populations, mutation and survival of the fittest to generate increasingly better solutions to the problem. These methods include evolutionary algorithms (e.g., genetic algorithms) and swarm intelligence (e.g., ant algorithms).

Learning
Computers can quickly replicate and copy their intelligence as oppose to human individual learning. However, unique intelligence of each human is a strength as it adds diversity, different strategies and wide range of solutions if applied properly (Wisdom of Crowds).

The Grand Challenge: Can a computer understand?

The Computational Theory of Mind claims that the brain is a kind of computer and that mental processes are computations. If it is true we can build AI, which can match human mind (strong AI), have intelligence and can understand meaning. Otherwise, we will struggle to build better and smarter algorithms (weak AI), which will never achieve true intelligence in the human sense.

There are three serious objections to possibility of building strong AI.

• Gödel Incompleteness Theorem
• Searle’s Chinese Room Argument
• Wittgenstein’s Beetle and The Box

1. Gödel incompleteness theorem
Any effectively generated theory capable of expressing elementary arithmetic cannot be both consistent and complete. (For more explanation see here and here.)
Any mathematical or logical theory should have effective ways to validate proofs. However, each of them has some initial propositions (axioms), which assumed to be true and cannot be proved within the system. To prove them you can go outside the system and come up with new rules and axioms, but by doing so you’ll only create a larger system with its own unprovable statements. The implication is that all logical systems of any complexity are incomplete, i.e. have true statements without prove.
Modern computer is a logical system. Therefore, computer cannot have true human intelligence as it will be always limited by fixed axioms and initial rules programmed by humans, while humans can discover new axioms and rules. In other words, a computer can play chess, regulate air traffic, trade stocks based on initial programmed rules, but it cannot define new rules which are not derived from initial set.
Another interesting outcome is that we will never be able to understand ourselves, since our mind is closed system, which can only be sure of what it knows about itself by relying on what it knows about itself.

2. Chinese room argument
In his argument Searle argued against a position that computer can have strong AI, think and understand.

Suppose a human is locked in the room. He takes Chinese symbols as input, follows thick rule book on English and returns Chinese symbols as output. But the person doesn’t understand even a word of Chinese! If you replace human in Chinese room with a computer following these instructions, it can easily pass the Turing test and convince a human Chinese speaker that the program is itself a human Chinese speaker.
But computer is the mindless operator of symbols with zero understanding of language and meaning (as your dishwasher doesn’t understand what it is washing). Therefore, we can simulate mind in computer, but we cannot create computer mind that understands meaning. Computers are missing intentionality (relationship between mental acts and the external world). A computer points to other data while human intentionality points outside the brain (e.g. to real flowers).

There are interesting counter-arguments like the System Reply, which debates that in Searle’s thought experiment a human is a processor and don’t understand Chinese, but the whole system can understand. However, the Chinese room argument still holds position against strong AI from 1980.

3. Wittgenstein‘s Beetle and The Box (Private Language Argument)
Wittgenstein shows that the idea of a language understandable by only a single individual is incoherent. Imagine, he says, that everyone has a small box in which they keep a beetle. However, no one is allowed to look in anyone else box, only in their own. Over time, people talk about what is in their boxes and the word “beetle” comes to stand for what is in everyone box.
Wittgenstein is trying to point out that the beetle is an analogy to an individual mind. No one can know exactly what it is like to be another person or experience things from another perspective (look in someone else box), but it is generally assumed that the mental workings of other people’s mind are very similar to our own (everyone has a beetle which is more or less similar to everyone else).
Therefore, computers should be part of our life, social interactions and cultural context to really understand us and construct similar meaning.

Summary

Deep Thought

Human intellectual capabilities are amazing and unique and human programmer will be superior comparing to AI for a long time. However, our modern civilization cannot progress without computers – we continue building better computers and require more and more from them. It is inevitable that computers will become intelligent whether it happens with our help or not. There are serious objections to possibility of building human-like “strong AI”; however, “weak AI” even today stepped in the territory previously believed to be human only. A computer needs programs and it is matter of time before they learn to program themselves and who knows what happened after this. But if computers continue to serve humans, even intelligent computer should effectively interact and understand us to write useful programs. This is a topic for the next post.

Useful Resources:
Standford Encyclopedia of Philosophy
Business By Numbers, The Economist
Great Ideas of Psychology, Teaching Company, Professor Daniel N. Robinson
Computing machinery and intelligence, Turing, A.M. (1950).
Mind, Brains, and Programs, by John R. Searle
10 Unsolved Mysteries Of The Brain, Discovery Magazine
Evolution in Your Brain, Discovery Magazine

Posted by Andriy Solovey | Permalink | Trackback