Task 1, What is AI?
What are 4 criterias that can be used to evaluate a search algorithm's performance.
Completeness: Is the algorithm guaranteed to find a solution when there is one? Optimality: Does the strategy find the optimal solution? Time complexity: How long does it take to find a solution? Space complexity: How much memory is needed to perform the search?
Explain the difference between a weak AI (narrow AI) and a strong AI (general AI)
A weak AI is designed to a do a specific task compared to a strong AI which are designed to do multiple different tasks. A strong AI is also able to learn, make decisions in uncertainty and plan for example while a weak AI is a lot more constrained to the specific task.
What type of problem is not suited for AI?
Problems which have a defined answer which can relatively easily be found using common algorithms of mathematical formulas. AI could be used for these type of problems, but aren't needed and wouldn't necessarily do better than the pre-existing solutions.
What is the difference between strong AI and weak AI?
The difference between the two is mainly their capabilities. Strong AI can emulate the human mind while weak AI can only simulate it.
What are main reasons machine learning and neural networks became more successful in the 21st century?
This is primarily attributed to advances in computing power and availability of large amounts of data.
How do search trees relate to "repeated states"?
When visiting different paths in a tree, one could revisit the same paths already visited if one does not care in choosing next path to explore.
Discuss the differences about ANI and AGI
ANI is sometimes referred to as weak AI, it is a level of AI that we have reached. It is AI that is specialized in one area. There exist chess playing AI beating the best humans in chess, IBM's supercomputer Watson that won jeopardy. ASI is sometimes referred to as strong AI or Human-Level AI. This level of AI is as smart as a human being, and is capable of performing any intellectual task that a human can. We have not yet reached this level of AI.
Discuss why or why not the Turing test would be a good way to assess general intelligence.
Alan Turing's original reasoning was that humans were the standard of intelligence, and the test is a formal way to determine how good a given machine is at appearing indistinguishable from a human purely intellectually. Some criticisms are that the Turing test does not consider perceptual skills or manual dexterity, or that it (unnecessarily) expects a machine intelligence to behave or be capable of mimicking a human [2], putting Turing's fundamental argument to question. Turing himself also mentions that the test might encourage the machine to make mistakes in order to appear more convincing, which would otherwise be undesirable.
What is the meaning of superintelligence in the subject of AI?
Superintelligence is a hypothetical agent that far surpasses the human experts in basically any task.
When using A* for pathfinding, what is an example of a heuristic function that might be used?
The Euclidean distance between a given node and the goal node. Specifically, it is a lower bound of the true distance.
Explain what artificial planning and scheduling is
Artificial planning is identifying steps to a goal and executing them in the correct order by scheduling them. This can be done by trial and error - one possible path is tested first, if it does not work, the artificial agent may then backtrack and reevaluate its plan.
Give five examples of Artificial Narrow Intelligence used in todays society.
Cars (anti-lock brakes, tuning the fuel injection systems, self-driving) Phones (Google Maps, check tomorrows weather, talk to Siri) Email spam filters (starts with intelligence, then learns even more continuously) Recommended products online, Google Translate, Landing planes, allocating resources, Bots in games, Googles search service.
What is the difference between Depth-First and Breath-First Search?
Depth-First Search (DFS) - starts by going down towards the left-most node at the deepest level. Then working its way to the right by going up to the parent, then to the next child node, then back to the parent and the next child. This is repeated until the parent doesn't have any more children, it now acts as the child, goes up to its parent and the process is repeated until it reaches the right-most node.[3] Breath-First Search (BFS) - starts the search at the root and then to its children starting from the left. If the goal is not found it repeat the processes at the left-most child and its children where it acts as the root. This is repeated until the bottom rightmost node is reached.
Explain the difference between Uniform-Cost Search and Dijkstra's algorithm.
Dijkstra's algorithm is a variation of Uniform-Cost search. They both find the shortest(cheapest) path to the solution. UCS requires less memory as it gradually explores the nodes. Both explore the cheapest path possible every step. That is, takes the cumulatively cheapest node possible from the start. UCS can handle infinite graphs and requires less memory in practice.
Explain the difference between machine learning and classical (symbolic) AI
In symbolic AI all the steps are based on human-readable representations, using logical deduction and search algorithms to solve problems [4]. Machine learning techniques, on the other hand, are designed with a set of parameters that adjust to produce desired behavior based on data or some form of selection. Many machine learning techniques can be examined using known models from mathematical statistics.
Explain what "Industrie 4.0" is. Mention three technologies that is included in "Industrie 4.0".
Industrie 4 is the current trend of automation and data exchange in manufacturing technology. It includes: Cyber physical systems- Software components with distributed units such as integrated units, smart objects, Human beings, physical environment like: Smart Cities, Smart Grids, Smart Factories, Smart Buildings, Smart Homes and Smart Cars. Internet of things - Network of connected heterogeneous objects: sensors, smart devices, embedded computers, mobile devices. Cloud computing - Cloud computing metaphor: the group of networked elements providing services need not be individually addressed or managed by users; instead, the entire provider-managed suite of hardware and software can be thought of as an amorphous cloud.
Nick Bostrom describes three ways a superintelligent AI system could function, which three ways?
It could function as an oracle, granting us answers to nearly any question posed to it (Google is a primitive type of Oracle). As a genie, which executes any high-level command it's given (use a molecular assembler to build a new and more efficient kind of car engine) As a sovereign, which is assigned a broad and open-ended pursuit and allowed to operate in the world freely, making its own decisions about how best to proceed.
What is domain independent planning?
It is a domain where the environment is not predefined for the AI, thus it need to be able to adapt to an unknow environment.
Do you think that superintelligence is something that should be addressed or something you should be aware of for?
It's not easy to know what's going to happen if we finally created something similar to the definition we have today of superintelligence. What would one like that? life shape do with all the knowledge it possesses? Will they have a counterpart to emotions? Will they be able to feel empathy and compassion for others? If not, it may cause us to will not last long on this earth. If, on the other hand, they had other motives, like that living with us would probably not cause major problems. Today, we are living people with animals whose level of intelligence is far from spring, but an interaction on earth is nevertheless no problem. If it were to be the same with a new species of superintelligence, it's hard to know today after two lectures and some articles, but I think so far the most seems exciting and desirable. Name a few examples of AI implementation in our everyday lives. Google Assistant in Android Mobile, Email Spam Filter, Google Translate and lots of things who keeps track of what I googler and buy to later recommend me things. In addition to these few, there are today massive AI implementations in the world like one does not think about or really care about everyday life.
Discuss some ethical concerns with strong AI.
One issue is that, like any powerful tool, it could be controlled and misused by governments or corporations. Another is the issue of consciousness, whether an artificial mind of comparable intellect to a human would have rights, if it would be ethical to simply abort or delete an AI process.
What is the difference between problem solving and planning?
Planning refers to determining a sequence of actions that are known to achieve a particular objective when performed. Problem solving instead is all about a systematic search through a range of possible actions in order to reach some predefined goals or solution.
Strong AI
Reasons Solves complex problems Draws conclusions Plans Learns (ML) Communicates Possesses consciousness Has self-awareness
Weak AI
Reasons and solves simple problems within a domain Uses models for problem solving Translates languages Handles spell checking Handles DB searches Lack of consciousness
What does STRIPS stand for and what is it?
STRIPS or Stanford Research Institute Problem Solver is a planning approach developed by Fikes and Nilsson in 1971. It involves the system identifying the differences between the current state and the goal state and then selects actions which reduces these differences, the method is called means-ends analysis.
What is state space search?
State space search is about mapping up all possible stages one can end up from one starting stage. This mapping is called a Space State. Then you can then, as in a direction graph, cross all possibilities until you reach the desired final stage. The classic AI- The example may be this game where you can fill a box with any of the boxes around in order to finally reach the lower stage of the picture. However, this can be applied to more complicated problems, such as how an agent is going to move stand in a room to reach the final goal in the shortest time or way.
What is the Chinese room?
The Chinese room is an experiment described by the philosopher John Searle, in which a person who doesn't speak or understand Chinese is placed in a room. In the room there are a number of card with Chinese characters on them along with instructions written in a language the person understands (English). The person is given a story in Chinese and a number of questions about it. The person is able to use the cards to answer the questions about the story in Chinese by following the English instructions. If the system us configured correctly, the answers would be good enough that the people asking the question would believe that the room (the person inside the room) understands Chinese. Searle's reasoning is that neither the person, the cards or the room knows Chinese but the system still displays enough knowledge of Chinese that the people asking the questions believe that it understands the language. Just because a system functions intelligently (displays understanding), it doesn't necessarily mean that it is intelligent (understands).
What is the Turing test?
The Turing test was designed by Alan Turing to test if a machine/system is intelligent. A test person is placed in a room with two separate terminals, one is connected to the machine and the other is connected to another person. The test person can ask question in both terminals and after minutes have to decide which of the terminals is connected to the machine. If the test person guesses incorrectly 30% of the time the machine passes the test and can thus be considered as intelligent.
Why has AI become so popular in recent years and what might explain the rise of it?
The computing power is constantly rising and since AI requires a lot computing power the use of complex AIs is getting more efficient every year. Furthermore, large sets of data are required for the training of useful Ai-agents. Recently more and more data are collected which enable more efficient AIs. Also, algorithms for AIs are constantly achieving better result and the development of AIs is funded by a raising number of stakeholders, e.g. tech giants and governments.
Explain the difference between Uniform-Cost search and A*.
UCS is uninformed while A* is informed. A* is a combination of UCS and Best-First search. A* chooses the most efficient path, by minimizing the total path cost of the next node while also minimizing the approximative distance (according to heuristic) to the solution. A* and UCS provides an optimal solution (which Best-First search alone does not). With a good heuristic, A* is much better. Defining a good heuristic might be difficult.
Compare and explain the difference of weak and strong artificial intelligence
Weak AI also known as Narrow Artificial Intelligence (ANI) specializes in one area and does it very well. We are constantly surrounded by ANI such as Alphago beating some of the best Go players around the world. Even simple programs such as calculators can compute any number faster than any human possible. Strong AI also known as Artificial General Intelligence (AGI) are referred to computer programs that are basically just as smart as human across many domains. To be considered as AGI the program must be able to have the following qualities: Reason Solve complex problem Draw conclusion Plan Learns Communicate Possesses consciousness Has self awareness
Compare the AI technology we have today - Weak AI with what is called Strong AI. Which are they biggest differences between the two types? Do you think we will ever achieve it as is called Strong AI?
Weak AI is very adapted to just the tasks that are to be performed and can not be met true of tasks outside the work area for which it is built. A Strong AI on the other hand, can handle several areas of work, draw conclusions and plan and learn a higher level than a Weak AI. (Lecture) However, if we will ever achieve Strong AI, it is a little clever to answer. In the article The AI Revolution: The Road to Superintelligence by Tim Urban, he addresses what difficulties it is is currently available. One of them, and perhaps the easiest to solve, is today's lack of hardware. According to Tim, only one computer has so far managed to get up to the same amount of calculations per second as a human cope. That computer is today (2015, when the article was written) fastest supercomputer and takes up 720 square meters and uses 24 megawatts compared with the brain's 20 watts. Thus, with the present case of Moore's law and today's performance In computers it feels hard to soon reach the power needed to reach a reasonable level Strong AI. Even if it succeeds, we must make the AI more human, smarter, more adaptive in its functioning. In order to achieve that, we may need to recreate something that mimic evolution, which feels quite far away. It feels almost more likely that man dies before that happens.
Why are search strategies needed for AI?
When an AI system is faced with a problem it needs to decide a set of actions in order to move from the initial state to the goal state where the problem is solved. It stores the sets of possible actions in a search space where one of more of those sets will lead to the goal state. The system usually wants to find the goal state as fast as possible with the least amount of resources used therefore search strategies are implemented to search through the search space for a set of actions which will lead to the goal state.
Give three examples of everyday (Artificial narrow intelligence)ANI-systems that you encounter daily.
• Talking to Siri on my iphone. • When I use google search (ranks pages and figures out what to show just you in particular). • The spam filter for my email.