<
>
Download

Referat
Ingenieurwissenschaft

Gymnasium, Stuttgart

A, Mueller, 2016

Rafael K. ©
5.25

0.24 Mb
sternsternsternsternstern
ID# 64113







Artificial Intelligence

An Introduction and First Developments


Table of contents


  • Definition


  • General Information


    • Introduction and issues

    • The laws of robotics

    • AI specialization and subfields

    • Problems and goals of AI

    • Artificial psychology

    • Current achievements


  • Potential risks of Artificial Intelligence

    • Narrow AI vs. general AI (AGI)

    • Tay

    • AI safety research


As the field of AI is almost endless, this GFS is made to give a small insight to the most important topics.


Definition

Artificial intelligence (AI) is defined as the intelligence exhibited by machines or software. In addition it is also the name of the academic field of study, which teaches how to build computers and computer software that are capable of simulated intelligent behavior. The first known use of “Artificial intelligence” was in 1956.


General Information

Introduction and issues:

The worldwide widespread research of AI is highly technical and specialized. It is deeply divided into subfields that often fail to communicate with each other. The reasons for this are various cultural differences as well as also the branch at which the individual researcher is active. Completely new subfields have grown up around particular institutions and the work, which has been derived by experimentsshowed up a new possibility nobody could ever imagine before.

Some subfields focus on the solution of miscellaneous specific issues set to the artificial intelligence to solve in a certain time or way. Others focus on one of several possible approaches or on the use of a particular tool or towards the achievement of particular applications.

Several technical issues result into even more subdivisions of AI research.


John McCarthy, the inventor of the term “Artificial Intelligence” states:

AI is the science and engineering of making intelligent machines."


The laws of robotics:

The possibilities of robots endowed with artificial intelligence are almost unlimited. That’s why Isaac Asimov invented the three laws of robotics in 1942 and completed them 41 years later in 1983 with a fourth, respectively zeroth law, to precede the others.

These are as follows:

  1. A robot may not harm humanity, or, by inaction, allow humanity to come to harm.

  2. A robot may not injure a human being or, through inaction, allow a human being to come to harm.

  3. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

  4. A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.


AI Specialization:

As previously mentioned there is research in different subdivisions for new innovation in AI. Every researcher has specialized himself on a certain kind of problem.

Artificial intelligence includes the following general areas of specialization:


  • game playing: programming computers to play games against human opponents (for example: chess, go, )

  • expert systems: programming computers to make decisions in real-life situations (for example, some expert systems support doctors diagnose diseases based on symptoms and give possible advices) or in general they perform a task that would otherwise be performed by a human expert (financial forecasts, scheduling routes for delivery vehicles).

  • natural language: programming computers to understand natural human languages (for example: Siri made by Apple, Cortona made by Microsoft, diverse car systems)

  • neural networks: Systems that simulate intelligence by attempting to reproduce the types of physical connections that occur in animal brains. These kind of AI is often used for voice recognition systems, image recognition systems, industrial robotics, medical imaging, data mining and aerospace applications.

  • robotics: programming computers to see, hear and react. Commonly used for high-precision jobs such as welding and riveting. Furthermore they are used in life-threatening situations -- for example in cleaning toxic wastes or defusing bombs.


Problems and goals of AI:

Developing of AI inevitably leads to a lot of different issues and also absolute failures. There’s still a long way to go for the human intelligence to achieve the completion of a fully working artificial intelligence.

Researchers have set themselves high goals to fulfill.


Generally a valid goal in AI development is to …

  • invent a AI system that perceives its environment and takes actions that maximize its chances of success

  • make AI teachable and able to reproduce the new gained knowledge as well as able to develop a plan to solve a given problem (math, labyrinth, )

  • fluently communicate to a system with spoken language

  • form an AI with perception and the ability to move and manipulate objects

  • invent fully working logical thinking. Combine and transfer achieved knowledge to use for new issues

without any malfunction.


Mark Zuckerberg, the founder of Facebook stated that he’s working on a new project for his personal life which maybe could revolutionize the life of everybody.


General intelligence” is still among the field's long-term goals (beside Social intelligence and Creativity). Developers want to build a machine capable of performing "general intelligent action".

Actually, the most popular approach is based on statistical methods and computational intelligence. Minor but also common is the use of traditional symbolic AI. Statistical methods help robots to maximize their chance of success in winning against humans, also in very complex board games, for example Go or chess, like the generally known probability theory. Forecasts state that intelligent robots will publish financial reports, sports commentaries, clickbait and myriad other articles formerly the preserve of trained journalists.


However, fast progress in development over the last few years doesn’t mean that robots provided with AI are inevitably take control over the world.


Computers might beat us at board games, but that doesn’t mean they’ll take over the world” Adam Roberts


There is a various pool of methods AI is based on, including versions of search and mathematical optimization, logic, methods based on probability and economics, and many others.


Artificial psychology:

Artificial psychology is theoretical competition discipline invented by Dan Curtis in 1963.

It states that Artificial Intelligence has to fulfill two conditions.


Condition I

  • 1. AI makes all of its decisions autonomously

  • 2. AI is capable of making decisions based on information that is either new, abstract, incomplete or a combination of those three.

  • 3. AI is capable of reprogramming itself based on the new gained data

  • 4. AI is capable of resolving its own programming conflicts, even in the presence of incomplete data. This means that the intelligence autonomously makes value-based decisions, referring to values that the intelligence has created for itself and don’t conflict with the 4 laws of Isaac Asimov.


All four criteria are met in situations that are not part of the original operating program


Artificial psychology was founded on the claim that our legacy, human intelligence—can be described so precisely that a machine can be build and thought to resemble it.


This brings up philosophical discussions about the nature of the mind and the ethics of creating artificial beings “equipped” with a human-like intelligence, issues which have been scrutinized by myth, fiction and philosophy since hundreds of years.

Artificial intelligence has been the matter of enormous optimismbut has also suffered vast setbacks. Contemporary AI techniques have become an essential part of the technology industry, ensuring the burden for many of the most challenging problems in computer science.



Current achievements:

Nowadays machines, respectively robots, have achieved to

  • beat Lee Sedol who is formerly known as the best Go player of the word (AlphaGo)

  • identify objects and persons in images to help blind people (facebook.com, )

  • lip-read. This technology could help people who have recently lost their hearing or voice and, on a more basic level, it could improve our interactions with gadgets that are usually controlled by hand.

  • walk and recognize obstacles to sidestep them

  • chat on social media platforms and messengers a previously selected content automatically in the future (e.g. weather, news about a certain topic,.)

  • drive cars itself

However, almost all of these achievements still need improvement and aren’t ready for a public release. The development of safety features might be still in beta level (e.g. a robot isn’t able to identify all objects for blind people, only some) but also the necessary laws haven’t been made by politics yet (“who can be judged for an accident happening with a Self-driving car, the car owner, because he didn’t react fast enough or the developer because he made a mistake in code writing?”).


Solving equations statistic:

The error rate of one AI by year. Red line - the error rate of a trained human”

Potential risks of Artificial Intelligence

Narrow AI vs. general AI (AGI):

A growing crowd of experts within and outside the division of AI has raised concerns that future developments may represent amajor technological risk for flora and fauna, humans and finally to earth; a scenario depicted already in several movies. (cp. Terminator I-IV)

To specify those possible dangerous circumstances it has to be differenced between “narrow AI” and “general AI”.

The first category is considered harmless. Those machines or robots endowed with narrow AI use algorithms specifically made for tackling an almost perfect specified problem under certain non changing conditions; such just reproducing procedures cannot adapt to new or broader challenges without almost fully replacing the previous set algorithms.

While narrow AI may outperform humans at whatever its specific task is, like playing board games or solving given problems, AGI would outperform humans at nearly every cognitive task. Probably their social intelligence and creativity will exceed the humans we actually could never imagine.

For example, it seems possible that AGI could be weaponized or used as a tool for social control, or someone might create an extremely powerful artificial intelligence agent with a small typing error in program code making it able to hack itself and break all the rules set for it. In addition it seems possible that the progress of development along these directions could be surprisingly rapid, leaving society underprepared for the transition.


Tay:

At the moment AGI is in early alpha stage. Microsoft has begun tests with Twitter chatbots in 2014 in china (XiaoIce). XiaoIce is followed by 40 million people and known for “delighting with its stories and conversations,” according to Microsoft.

On March 23, 2016 Microsoft published another Twitter chatbot called “Tay” in the USA. Their goal was investigating how artificial intelligence programs can engage with Web users in casual conversations.

Unfortunately, Tay wasn’t that far developed. Private Twitter users had a malicious influence on “her”. They confronted Tay with massive racism and Holocaust questions she couldn’t answer properly. Some of Tay’s responses were inappropriate (sexist and racist) and indicative of the types of interactions some people are having with it.

The project was shut down after 16 hours.


AI safety research:

Referring to scientists full AGI will be invented within the next seventy years.

AI’s potential benefitsand their potential for harmwill increase equally rapid.

To decrease the potential risk of general AI, planning and more detailed research to prevent unexpected catastrophic consequences is unconditionally necessary.


Last but not least it has to be mentioned that AI safety research is definitely not negligible. There’s nothing more important than inventing general valid circumstances for AI/AGI and it’s researcher to keep AI beneficial. Researches in common fields (from economics and law) to the more technical topics (as verification, validity, security and control) are clearly important for everyone’s future.


In the long term, an important question is what will happen if the quest for strong AI succeeds and an AI system becomes better than humans at all cognitive tasks.”

Source:


Bibliography

ünstliche_Intelligenz

Literary sources:

Künstliche Intelligenz (Pearson Studium - IT) – Stuart Russell

N. J. Nilsson: Artificial Intelligence: A New Synthesis. Morgan Kaufmann, 1998.


Handbuch der Künstlichen Intelligenz – G. Görz



| | | | |
Tausche dein Hausarbeiten