artintell

[artintell]

Chinese room argument

The argument was brought about by John Searle in 1980 in his paper, “Minds, Brains, and Programs”, published in Behavioral and Brain Science.It gives an idea about learning and understanding of machines and Artificial Intelligence. This argument tries to explain the difficulties with repect to a machine gaining complete and pure understanding and developing consciousness. A machine can only mimic based on the programming and data provided to it.

The actual argument

In this scenario, a person (A) is placed inside a room. The room has different placards with chinese symbols, the meaning of which is not understood by (A). However, there are specific rules which define a specific response when (A) is given a symbol. (A) cannot verbally communicate with the people outside the room. To the outside of the room, people write Chinese symbols on a placard, insert it inside the room via a slit. By seeing the palcard and following the rules, (A) gives the response via the slit. To the people outside, it might seem that (A) understands chinese, while in reality, he is only following rules without actually understanding the meaning of the symbols.

Relevance

It implies that (A) is just following instructions. Similarly, when a machine is given a specific amount of data, it only responds according to the data present and does not thimk or has consciousness. It only is following instructions which was fed to it by a human being.

Criticism

Critics to the Chinese room experiment believe that humans too, while responding or learning a language, base their decisions on information that they have had previously in terms of experience. Humans tend to use the information of experience and give appropriate responses.

Another line of criticism falls on understanding the system in the Chinese room argument. While it can be argued that the person didn't understand Chinese, but he, along with the placards, the rules and the entire room did understand Chinese. The system as a whole was able to comprehend what was asked from it and was able to reply back. This is the same as a computer, with its cognitive features and sensors, along with its processing unit and data, understanding and responding to a particular problem and providing appropriate response.

[artintell]

The Turing Test

Turing Test was introduced by Turing in his 1950 paper, "Computing Machinery and Intelligence," which considered the question, "Can Machine think?"

It is basically a test to see if a computer has become sophisticated enough to mimic human response.

So in the Turing test, there are 3 subjects:

1. An interrogator (human)

2. Responder 1 (computer)

3. Responder 2 (human)

The interrogator asks a series of questions and the responders will have to give an answer in a stipulated amount of time. At the end of the test, it is the job of the interrogator to decide which responder is human and which responder is the computer.

In this test, the actual questions or the answers don't matter that much. It really doesn't matter how many correct answers were given by each responder. What matters is if the responses of the computer were similar to that of the human. And also the fact that was the computer able to fool the interrogator into believing that it was actually a human.

The Turing Test today

In an updated version of the Turing test, there are more than one human interrogators interacting with both the responders. If the computer is able to trick 30% or more of the interrogators, after 5 minutes of conversation with each of them, into believing that it is a human, then it passes the Turing test.

In this regard, the Loebner Prize was instituted by Hugh Loebnor, an American inventor and activist, in 1991 which is an annual Turing test competition. He added additional rules. The rules required the human and the computer program to have 25-minute conversations with each of four judges. The winner is the computer whose program receives the most votes and the highest ranking from the judges.

In 2018, Google Duplex was introduced at the annual Google I/O Annual Developer Conference. The machine scheduled a hair salon appointment and interacted with a hair salon assistant via the phone as part of the conversation. Though some critics view the outcome differently, some believe Google Duplex passed the Turing test.

Limitations of the Turing Test

1. Requirement of a very controlled environment to be performed.

2. Turing test does not assess all types of intelligence.

3. Computers not having communication skills cannot be tested.

4. The test only give a comparative result and not a definitive one.

[artintell]

Are we moving towards General AI?

AI is still in its initial phases of development. Whatever Ai has developed, no matter how interactive and new, is still weak ai. But there are works going on to move towards general AI.

But that would require deep learning.

What is deep learning?

As per a paper titled "Deep Learning" by Yann LeCun, Yoshua Bengio & Geoffrey Hinton, "Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction."

Impact of deep learning in generative AI?

According to the theory of multiple intelligences proposed by Gardner, at least eight different types of intelligence exist: logical-mathematical (reasoning, number smart), visual-spatial (picture smart), verbal-linguistic (word smart), musical-rhythmic and harmonic (sound smart), bodily-kinesthetic (body smart), naturalistic (nature smart), intrapersonal (self-smart), and interpersonal (social smart).

While neural networks help in developing basic understanding of decision making, it still is heavily data driven. To generate true AI, there is a need to develop original understanding. It would also include making original mistakes and then learning from them. This learning will have much more value as compared to data driven learning.

Also what needs to be understood is that there is a difference on how a human brain and a data driven center processes a certain information and stores that information. While in human beings, the data comes from experiments of life, in AI the data is pre fed. Also the information, in case of humans is leaked in some sort, in case of AI it isn't the case.

A simple example can be taken with the fact how a child learns to understand alphabets. He/She may first see the alphabet, understand its structure and sound, use it with a demonstration and then try to store all this data. There is a very good chance that the kid would make mistakes in the next go. But with continuous reinforcement, the understanding of the letter develops.

On the other hand, a computer won't have the same approach. It would store the data of its structure, sound and demonstration in binary codes and would never forget it. This is the very reason why original understanding doesn't develop. The fact that new experiments in the form of mistakes are not performed is what makes a computer distant from general intelligence.

So we can say that prospects and hopes are there, we still have a lot of work to do.

[artintell]

Artificial Intelligence

What is intelligence?

Throughout ages, different thinkers and philosophers have tried to define intelligence. Aristotle, in his treatise On the Soul (De Anima), claims that human beings are the “most intelligent of animals” because they have the most precise sense of touch. However he does not explain the importance of the sense of touch that makes humans so intelligent. 

Thorndike defined intelligence as the power of good responses from the point of view of truth or facts.

While there are different parameters and definitions of intelligence, fo the sake of simplicity we can say that intelligence is the general sense of understanding, ability of logic, reasoning, problem solving and to adapt to new experiences based on old ones.

What is Artificial Intelligence?

John McCarthy offers the following definition for Artificial Intelligence (AI) " It is the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable."

Stuart Russell and Peter Norvig then proceeded to publish "Artificial Intelligence: A Modern Approach". In it, they delve into four potential goals or definitions of AI, which differentiates computer systems on the basis of rationality and thinking vs. acting:

Human approach:

Systems that think like humans

Systems that act like humans

Ideal approach:

Systems that think rationally

Systems that act rationally

Weak AI and Strong AI

Weak AI- Artificial intelligence with limited capability is referred to as weak AI. The employment of sophisticated algorithms to execute certain problem-solving or reasoning tasks that do not span the entire spectrum of human cognitive capacities is referred to as weak AI. Voice-based personal assistants such as Siri and Alexa, for example, might be called poor AI programs since they work within a narrow pre-defined range of tasks.

Strong AI- Strong AI refers to computers or programs that have their brains and can think and complete complicated tasks without human intervention. If you are having difficulty in understanding strong ai then just remember the terminator movies. Although I would like to keep it realistic.

APPLICATIONS OF AI

Better customer experience in product selling

Digital personal assistant

Cybersecurity

Self driving cars

Code debugging

Progressive learning algorithms