Chapter 1 - Introduction Flashcards
How can we define AI?
We can define AI in regards to 2 questions: Fidelity vs rationality, and reasoning vs behaviour.
What do we mean by acting humanly?
Acting humanly can be summarized as trying to pass the Turing test. In order to “act humanly” a machine would need to be able to process natural language to communicate with humans, be able to STORE KNOWLEDGE that it reads or hears or knows, be able to automatically reason in order to draw conclusions, and finally be able to learn by patterns (ML).
You could also include the “total Turing test” and provide criteria like computer vision and robotics as well, but Turing himself viewed the physical part as not necessary as far as intelligence goes.
Acting humanly is the combination of fidelity and behavior. It is only a measure of intelligence.
Acting humanly as a measure of intelligence is ONLY concerned about the output. It dont give a fuck about internal processes. All that matters is what we, who interact with the machine, observe.
Is “acting humanly” or “passing the turing test” a focused area? why/why not?
No. Passing the test doesn’t really give anything, except perhaps westworld like tendencies.
There are other fields within AI that produce optimal solutions to problems, and there is hardly difficult to behold the fact that those fields are prioritized ahead of making machines act like humans for the fuck of it.
Name an example where imitation of real life creature is actually a more difficult approach?
Consider birds and flying.
What do we mean by “thinking humanly”?
Thinking humanly refers to humans’ thought processes. How do we think?
Thinking humanly is a measure of intelligence that is constructed by combining fidelity and reasoning. In other words, it refers to how well the machine resemble the human thought processes.
There are 3 ways we use when trying to grasp how humans think:
1) Introspection. Introspection refers to trying to capture the flow of thoughts. Tankestrøm.
2) Psychological experiments. Involves closely observing people to try to understand their process of thinking.
3) Brain imaging.
What is meant by thinking rationally?
Thinking rationally is a measure of intelligence that we get by combining rationality and reasoning. This means, we view intelligence as how the machine manage to find the best choices/solutions. We are interested in logic and syllogisms.
What is meant by acting rationally?
A measure of intelligence which we get by combining rationality and behavior.
This means that we view intelligence as the output and how good this output is. We are not concerned with the insides. Could be rational, could be humanly. Idgaf.
What is meant by “the standard model of AI”?
Why has this model become the main focus area?
The standard model is a paradigm where people have focused on creating and understanding agents that do the right thing. In other words, acting rationally. It is, in fact, so well accepted that it has become known as the standard model.
The reason for the standard model’s success is because of several points:
1) Rationality is more based on science and math, at least it is more well defined, which means that it becomes easier to work with.
2) Rationality use probability to combat uncertainty, which is very helpful. We wouldn’t do this if our goal were to imitate humans.
3) The best outcome is often of great interest. Fidelity is cool, but perhaps not as useful as some machine that can massively improve our solutions.
The summarize: The standard model is a paradigm where people and scientists have focused on understanding and creating agents that do the right/correct thing.
There is great benefit in adding that “perfect rationality” is not always attainable. For instance, if the solution require too much time to compute, we should have a way to get a “good enough” solution, which should require less time. We coin this “limited rationality”.
What is the “value alignment problem”?
The value alignment problem is the problem concerned with achieving an agreement between our own values and the objective that we supply the machine with.
To illustrate: If we specify the goal as “drive from A to B safely”, the model might not know how, because driving generally is not safe. This is a value problem. We value safe, but we also value driving. There is a trade off. We must understand how the machine can operate well when difficult/complex situations like this one occurs.
Name the different fields the constitute the foundation of AI
Philosophy
Mathematics
Economics
Neuroscience
Psychology
Computer Engineering
Control theory and cybernetics
Linguistics
What foundation questions from philosophy are important?
1) Can formal rules be used to draw valid conclusions?
2) How does the mind arise from the physical brain?
3) Where does knowledge come from?
4) How does knowledge lead to action?
Forklar dualisme
Dualisme refererer til Descartes syn på mind vs matter. Dersom sinnet kan beskrives med fysiske prinsipper, så fjerner det fri vilje. Dermed, skal en ha fri vilje, må sinnet være noe mer enn bare fysikk.
Why is the question on “how knowledge leads to action” so important in AI?
It is important because AI is mainly focused on finding the correct/optimal actions. We need to understand what makes one action better than the other.
How does knowledge lead to action?
Aristotle said that actions come from knowing the result of the action based on current information.
In other words: There is a logical connection between the goals and the knowledge of the actions outcome.
We (the machine) have a goal. It also possess knowledge on what a certain action would do. Then it could simply check whether the action leads to the goal or not.
What are the mathematical questions that lays the AI foundation?
1) What are the formal rules to draw valid conclusions?
2) What can be computed?
3) How do we reason with uncertain information?