Telecom Paris
Dep. Informatique & Réseaux J-L. Dessalles ← Home page February 2021 |
Logic and Knowledge Representation
with
Etienne Houzé
Louis Jachiet
→ other AI courses
Prolog is also an excellent opportunity to train oneself in a way of thinking that goes beyond the domain of computer science. It consists in addressing a problem in terms of constraints rather than in terms of procedure, leaving to others (to the machine in the case of Prolog) the burden of finding out a solution that will comply with the constraints.
H. Hi, who are you ?
M. My name? Martine. And you ?
H. What is the square root of 17684?
M. Don’t know. I’ve always been crap at maths.
H. What does the scent of violet remind you?
M. To the smell of the bed sheets in my grandmother’s cupboard
H. Suppose I’m facing you. Now I’m turning to look to the right. On which side are you now?
M. I’d be on your left.
Would you be able to write a program that plays M’s part? The program needs to understand natural language. It also needs to build up a scene to represent what is meant. And it has to understand the relevance of what is said. All things that are not so easy to do.
Recent improvements in statistical machine learning (data mining, deep learning) let some famous people to predict such a surge of machine intellectual capabilities that they could lead to the end of humankind (see Stephen Hawking’s or Bill Gates’ warnings). These concerns probably motivated other important figures to sign a call for better use of AI techniques.
We should not underestimate the danger of giving robots too much decision power. For instance, in high-frequency trading, virtually all decisions are made by computer programs; financial crashes may be the result, as was the case in 2010. However, current techniques are strongly limited, as they rely mainly on statistics. They can only reproduce or extrapolate from known behaviour. It is highy unlikely that the mere extension of statistical techniques may constitute a threat to humankind. Let’s consider a few examples.
Consider for instance the way a statistical machine may learn how to add to numbers. You’ll be never sure that the machine will get it right.
→ See my question to Jérôme Pésenti (from IBM Watson) during his talk at Telecom and his kind answer.
Let’s take another example. Try to ask this question:
How many phalanges are there in a Human body ?
State very briefly how you think your reasoning differs from the procedure used in the search engine. |
Yet another example. Anyone who is able to count will find an obvious way of continuing the following series:
1,2,2,3,3,3,4,4,4,4
Would you be able to write a program that can solve such riddles? Mmm..., think twice before considering the task trivial. Your program would need to implement a principle of simplicity (or minimum description length): the ‘best’ solution, here probably 5,5,5,5,5, should keep the description of the series as short as possible. This issue will be studied as a topic in this course.Some Web sites such as this one are specialized in performing induction on numerical series. Verify that the ‘correct’ solution is proposed.
Indicate an alternative solution proposed by the site. Explain why you think that solution is not so good. |
Importantly, solutions to such induction problems are found using computations that may have no statistical basis whatsoever (one-shot learning). However, these computations are sensitive to structure. In the previous example, anyone can perceive repetitions and incrementations (and the repetition of incrementation). This structural analysis is absent from statistical method (though deep learning can sometimes be used to extract frequent features from large sets of examples when available).
This course will deal with symbolic aspects of AI. This means that we will use structures, structural matching and structure-sensitive procedures. This will allow us to deal with problems (such as aspects of Natural Language Processing and problem solving) that are intractable using brute statistics. It will allow us to produce systematic behaviour and to deal with exceptions and anomalies. And we will see how background knowledge can dramatically improve machine learning, making one-shot learning possible.
♦ Symbolic models are characterized by the use of structures. They rely on discrete time (e.g. a loop index in a sequential algorithm). Procedures are sensitive to structure. The simplest one consists in matching structures.
♦ Sub-symbolic (or dynamical) models, by contrast, are much closer to physics. They involve metrics, parallel processes and rely ideally on continuous time (though they are generally simulated on discrete sequential machines). Most current sub-symbolic technics are based on statistical computations.
Mention two topics in Artificial Intelligence that belong to the symbolic approach
and two other topics that are rather sub-symbolic. You may have a look at the topics mentioned on the AFIA Web Site, on the UC-Berkeley AI page, or on the ICAI’16 web site, or in the AI100-group 2016 report (read p. 6 on). |