Artificial intelligence - a short introduction§
Amélie Cordier
Département Informatique, IUT Lyon 1
Disclaimer§
This is only an introduction to AI (Artificial Intelligence).
These slides contain only a few things... listen and take notes! You'll find a lot of links, browse them and be curious!
Is AI really everywhere?§
Examples of AI in our everyday life§
Exercise: where else?§
Can you give other examples of AI in our world?
Artificial intelligence, a definition§
Exercise: Oh, but wait. What do you think?§
- In your opinion, how old is AI?
- What is your definition of AI?
Note
Attention à bien distinguer la date à laquelle on a commencé à s'intéresser au concept de l'IA et la date a laquelle le terme IA a été proposé pour la première fois.
AI, a definition§
Warning
There are many definitions of AI. Let's try to understand why...
"Artificial Intelligence (AI), studies the way we can design and build agents that act intelligently".
→ This is a rather vague definition. Let's get into details...
An agent?§
"Artificial Intelligence (AI), studies the way we can design and build agents that act intelligently".
An agent acts in an environment... The agent does things. It has inputs and performs actions.
Exercise§
- Can you give some examples of agents?
- Is a dog more intelligent than a jellyfish? Why?
- For each agent you identified, would you say that it is intelligent?
- What are the criteria you used to decide if the agents are intelligent or not?
Note
Exemples d'agents : les bactéries, les fourmis, les chats, les chiens, les mobiles, les thermostats, les robots, les télévisions, les humains, les organisations.
"Act intelligently": what does this mean?§
When we observe an agent, we can ask the following questions:
- Are the actions appropriate to circumstances and goals?
- Is the agent able to adapt when environment or goals change?
- Is the agent able to learn from its experiences?
When we try to answer these questions, it helps us decide if the agent seems to act intelligently.
Warning
The difference between "the agent seems to act intelligently" and "the agent acts intelligently" is subject to a big scientific and philosophical debate. Indeed, it raises the issue of consciousness and of defining exactly what intelligence is. We'll come back to that later...
Observable behavior§
Can we assert that it is only the observable behavior of the agent that defines its intelligence?
In 1950, Alan Turing proposed a test (a kind of imitation game).
Let's have a look at the Turing Test
Warning
Note that Turing was more interested in explaining how to assess if an artifact behaves intelligently or not than in explaining how to build this artifact!
Exercise: limitations of the Turing test?§
Questions:
- What is required to set up the Turing test?
- Do you see any limitation to the Turing test?
Can we reach convincing AI?§
For now, convincing AI only exists in fiction and movies, right?
- 1883: Pinocchio
- 1968: 2001: A space Odyssey (HAL)
- 1969: Colossus: the Forbin project
- 1981: Blade Runner
- 1982: Tron
- 1983: WarGames
Can we reach convincing AI?§
And also:
- 1984: Terminator
- 1985: DARYL
- 1987: Robocop
- 2001: AI
- 2004: I, Robot
- 2013: Her
Can we reach convincing AI?§
Where are we now? Is this artificial intelligence?
- OK Google
- Siri
What about conversational agents?
And what is that? (I'm sure you know...)
- Akinator the genius
Biological inspiration in AI§
Where does our intelligence come from?§
- Genetic predispositions
- Culture, language, tools, concepts, wisdom
- Teaching and learning
- Experience, knowledge and skills
Understanding where our intelligence comes from is a major source of inspiration (and doubts) when studying AI.
Collective intelligence§
Earlier, we defined agents as individuals or organizations...
Collective intelligence is amazing to study!
- Look at ants
- And termites
- Computer based modelisation of ants behavior
Artificial intelligence vs. fake intelligence?§
Fake intelligence is not possible!
If an agent behaves intelligently, then it is intelligent.
Artificial intelligence is just intelligence created artificially.
Several approaches of AI§
Different viewpoints§
People working on AI have various objectives, often complementary:
- Building agents that seem to behave intelligently, regardless to the process they use;
- Building agents that work as much as the human mind as possible.
In addition, they can be interested in:
- Rational intelligence (logical reasoning)
- Human like intelligence (more than logical reasoning)
Different goals§
Scientific goal: understand the principles that make intelligent behavior possible.
Engineering goal: design of useful intelligent artifacts.
Methodologies§
For building AI, we can:
- Observe natural and/or artificial agents
- Make hypothesis on the pre-requisites to build intelligent agents
- Design, build and experiment intelligent systems.
Exercise§
- Given the viewpoints introduced above, we can identify four ways of describing AI. Can you describe them?
- Consider the following analogy: "Can computers really think?" and "Can airplanes really fly?". Can you comment on that?
Note about the importance of perception§
One important limitation when designing intelligent agents is related to perception.
The agent has to make appropriate decisions with regard to its available perceptions.
It is impossible (even for us) to have a direct access to the state of the world. We all have a finite memory space, a finite time, and some perceptual limitations.
This raises the question of building representations of the world.
A very very very brief history of AI§
AI Timeline§
On AAAI.org, a timeline of the history of AI
Graphical timelines
- http://www.livescience.com/47544-history-of-a-i-artificial-intelligence-infographic.html
- http://www.scaruffi.com/mind/ai.html
An extensive timeline
Wrap up§
- The origins of AI is link to Turing's article: « Computing Machinery and Intelligence » (Mind, October 1950).
- The term Artificial Intelligence has been proposed by John McCarthy (with Marvin Minsky) at the Darthmouth College conference in 1956.
- Just to cite a few, here are some names of influential AI researchers: Alan Turing, John McCarthy, Marvin Minsky, Alan Newell, Herbert Simon, Jean Piaget, Jacques Pitrat, John McCarthy, John Searle.
AI: different schools§
Strong AI vs. Weak AI?§
Strong AI and Weak AI are two hypotheses about AI
Weak AI§
- Pragmatic approach
- Simulation (or reproduction) of intelligence
- No notion of "thinking"... only smart execution of heuristics
Weak AI is a very useful approach for practical applications... but weak AI systems are limited to narrow tasks and cannot evolve.
Strong AI§
- Intelligent behaviors
- Self-consciousness, cognitive states
- Understanding of its own reasoning
- Emotional intelligence (?)
- Curiosity
Is it possible? Is it a problem of computational power only? (not anymore...). Is it a problem of design?
Problem: how to explicit knowledge used to solve complex problems? For example, can you tell me how you do to recognize faces of your friends in pictures?
So, strong AI vs. weak AI?§
Strong AI vs. Weak AI: https://www.youtube.com/watch?v=5nwUJnlvjCc
The Chinese room experiment by John Searl: https://www.youtube.com/watch?v=TryOC83PH1g
AI vocabulary you should know§
Computational agent§
A computational agent is an agent whose decisions about its actions can be explained in terms of computation.
The decision can be broken down into primitive operation that can be implemented in a physical device.
This computation can take many forms.
It is an open question whether all intelligent agents are computational.
Symbolic AI§
Symbolic AI relies on symbolic representations for problems and knowledge.
Symbolic AI raises issues of knowledge representation, logic, search in spaces of states, etc.
Connexionism§
Small interconnected units makes it possible the emergence of behaviors and/or cognitive processes.
Most common example: neural networks.
Symbolic opposed to connexionist AI§
AI: links to other disciplines, and inspiration§
Links to other disciplines§
AI is strongly linked to many disciplines:
- Cognitive science, Philosophy
- Neurosciences and cognitive neurosciences
- Neurobiology, evolutionary biology
- Epistemology (study of knowledge)
But also psychology, economics, political science, sociology, anthropology, control engineering. What computer sciences brings is a way to get more powerful experimental tools than before.
Instead of being able to observe only the external behavior of intelligent systems, now we can experiment with executable models of intelligent behavior.
Challenges of AI§
- Reasoning, automated reasoning
- Natural language processing and understanding
- Visual, auditory perception
- Translation (context)
- Learning (supervised, unsupervised)
- Artificial emotions
- Pattern recognition
Challenges of AI (continued)§
- Reasoning under uncertainty
- Planing
- Reasoning with time and space
- Robotics, mecatronics, cybernetics
- Curiosity
- Creativity
Tools for AI§
- Agents, multi-agent systems
- Automata
- Data-mining
- Machine learning
- Ontologies, knowledge-based systems
- Logics, inferences
- Probabilistic reasoning
Types of applications§
Types of applications:
- Diagnostic and decision support systems
- Complex problem solving (resources allocation)
- Assistance (dangerous tasks)
- Task automation
Domains§
For:
- Bank and insurances (risk assessment)
- Security, defense (drones, decision support systems)
- Video games
- Teaching
- Health and safety
- Logistics
Programming languages§
- LISP
- Prolog
- Scheme
- SmallTalk
- Yafool,
- KL-one,
- OWL, RDF, description logics
Controversies in AI§
Is AI controversial?§
Over the years, AI had ups and downs... but recently, it is quite controversial...
- Eugene Goostman "passed" the Turing test (very controversial)
- Stephen Hawking said that AI could spell the end of the human race
https://www.youtube.com/watch?v=fFLVyWBDTfo
- So, AI? END or FRIEND?
Let's study AI controversies together§
Rules:
- Make teams;
- Choose a subject (see after);
- By the 5th of February: I give you more readings on the subject you chose;
- 16th of March: 10 minutes presentation on the topic you chose.
Topics§
- AI failures (drones, autonomous cars), who is responsible?
- AI and transhumanism?
- Ethical issues and privacy, tractability? Is is a problem of AI?
- Singularity
- Robotics, AI and Asimov laws
- Deep blue, Watson, what do the lack to beat us on other fields?
- Can we achieve the development of a strong AI?
- Should we let AI play with our money?
References§
Web references§
For an historical point of view on AI: http://ai.stanford.edu/~nilsson/QAI/qai.pdf
AI, an history (in French): http://fr.wikipedia.org/wiki/Histoire_de_l%27intelligence_artificielle https://www.u-picardie.fr/~furst/docs/3-Naissance_IA.pdf
http://artint.info/html/ArtInt.html
AI progresses http://en.wikipedia.org/wiki/Progress_in_artificial_intelligence
On the importance of context in AI http://www.cs.rit.edu/~rlc/Courses/IS/ClassNotes/History.pdf
Papers and books§
Artificial intelligence : a modern approach ? Stuart Russell & Peter Norvig
Turing, A. (1950). Computing machinery and intelligence. Mind, 59: 433-460. Reprinted in [Haugeland (1997)]. http://artint.info/html/ArtInt_350.html#reason:Haugeland97a