week 6 - ai in business Flashcards
what is artificial intelligence?
the study + design systems that have the ability to perceive + evaluate the environment + take actions that maximises the changes of success.
what are the definitions of intelligence?
- human performance (achieving results)
- rationality (doing the right thing)
- thought processes + reasoning (internal character)
- intelligent behaviour (external character)
what are the different approaches in understanding ai?
- the pursuit of human-like intelligence (psych, observations, thought process + hypothesis).
- rationalist approach (combo of maths, stats, engineering, control theory + econ).
what is meant by the turing test?
a test which can help determine whether a computer can pass as a human to understand its intelligence.
cares about the outcome regardless of whether its reliable or accurate.
what does a computer need to do to pass the turing test (and are also the six disciplines that are ai)?
Natural language processing to communicate.
Knowledge representation to store what it knows or hears.
Automated reasoning to answer questions + draw conclusions.
Machine learning to adapt to new circumstances + detect patterns.
Computer vision + speech recognition to perceive the world.
Robotics to manipulate objects + move about.
what is the cognititve model approach?
where ai simulates the human thought process.
what are the 3 ways humans think?
- introspection (observing thoughts + emotions in action).
- psychological experiments (observing an action in person).
- brain imaging (observing a brain in action).
how is the cognitive modelling approach used?
once it has a precise theory of the mind it becomes possible to express the theory as a computer program.
what does cognitive science do in relation to the cognitive modelling approach?
it brings together computer models from ai + experimental techniques from psych to construct precise + testable theories of the human mind.
what is meant by the “law of thought” approach?
it entails the ideas of thinking rationally through logic and applying that into ai algorithms.
describes how the human mind should work laying down the foundation for the field of logic.
what does it mean by codifying right thinking?
logic provides structured patterns for reasoning, ensuring that info is arranged in a way that others can follow - creating rules that govern rational thought.
what are the limitations of logic?
it always assumed that knowledge is 100% certain (in reality it deals with uncertainty).
how can we reduce these limitations of logic in the “law of thought” approach?
the use of probability in ai can help fill this gap by allowing reasoning with uncertainty, modern ai uses probabilistic models instead of strict logical rules.
is thinking rationally the only way to create intelligent behaviour?
no, rational thinking is only 1 way to create intelligent behaviour + needs other approaches to create enough to be successful.
what is meant by the rational agent approach?
a development of a decision theory and ai planning that focuses on rational action not just rational thought.
what is meant by an agent in ai?
anything that percieves its environment + takes action to achieve a goal. this could be a self driving car or chatbot.
what is meant by a rational agent?
one that makes decisions to maximise its chances of success.
what are the key characteristics of a rational agent?
- autonomy (operates independantly without human intervention).
- perception (observes environment through sensors).
- goal oriented (makes decisions that align with a defined objective).
- adaptability (can learn + adjust its actions based on new info).
- persistence (can work over a long period of time without consistent monitoring).
what are the different ways in which we can think rationally?
- deduce that a given action is best + then act on that conclusion (not always correct).
- ways that can’t be said to involve inference (eg recoiling from a hot stove is a reflex action that is used more successfully than a slower action taken after careful consideration).
what is the conclusion made of the ways of thinking rationally?
that thinking rational is one way to act rational.
what are the advantages of the rational agent approach?
- can be mathematically defined + well tested.
- recognises there are many possible strategies for making intelligent decisions.
- easier to analyse scientifically as it is based on mathematical models of decision making.
what are the disciplines that contributed ideas, viewpoints + techniques of ai?
- philosophy
- maths
- economics
- neuroscience
- psychology
- linguistics
- computer engineering
- cybernetics + control theory
is ai moving from automation to augmentation?
the relationship between humans + machines is shifting to a more collaborative + augmentative one. instead of viewing ai as a threat that replaces human workers it should be seen as a tool that enhances human abilities.
what is meant by opacity?
ai is opaque meaning it is not always possible to expkain its outcomes.
the opposite of transparency, referring to when info is hidden or difficult to understand.
what are the reasons for ai being opaque?
- complexity use of the algorithmic code
- the use of multiple algorithmic components which are not all know or under the control of the designers.
- intrinsic characteristics of learning algorithms that redefine parameters + relationships with each additional data point which makes it hard to explain, algorithms have to be tweaked to make their results understandable.
how does algorithmic opacity affect value creation?
if ai is not explainable it is impossible to determine the value it generates for orgs.
what is meant by explainability paradox?
it refers to the challenges in ai where making a model becomes more + accurate makes it less interpret-able so looks at the needs for human understandable explanations of their decisions + behaviours.
how does this create dilemmas for designers and users of ai?
they need to balance the need for accuracy + performance with the need for transparency + accountability.
it can be difficult to understand + interpret, even for experts, which can raise concerns about their reliability, fairness
what is a benefit for prioritising complexity + accuracy of ai models?
highly complex ai models such as deep neural networks can achieve impressive levels of accuracy in tasks such as image recognition or natural language processing.
can you discuss an example where privacy issues arose from prioritising complexity + accuracy in ai?
NHS + Deepmind partnership
- wanted to help detect kidney failure in patients however:
- data was shared without proper consent from patients which included sensitive info.
- the nhs did not inform the patients on how their data would be used.
- deepmind initially claimed the data would not be shared with google however in 2018, google took it over. this caused fears about the commercial use of data.
how can an org assess the value creation of ai?
hard to determine ai’s true value for an org due to its opacity + rigidity.
ai algorithms need to be deeply analysed to see how it functions, without this it can lead to misinterpretations of its effectiveness.
doesn’t adapt to real world unpredictable conditions + forces the environment to fit into pre-defined logic.
what is meant by generative ai?
ai models that generate new content based on learned patterns from existing data.
they don’t just recognise patterns, they produce creative outputs.
what are some examples of generative ai?
- text generation (chatgpt).
- image generation
- music generation
- video + animation
- code generation
how does gen ai work?
its based on deep learning + neural networks.
they are trained on huge datasets + use statistical probabilities to generate new, coherent content.
what do the key architectures of gen ai include?
- generative adversarial networks: two ai models compete against each other, one generates data and one evaluates it.
- transformers: these use massive datasets to generate human-like texts + responses.
what are the challenges + ethical concerns of gen ai?
- bias + ethical concerns: can generate biased or harmful content based on flawed training data.
- copyright: who owns it?
- misinfo + deepfakes: it can create realistic fake videos + news, leading to misinfo.
lack of explainability: gen models are black boxes, making it difficult to understand how they make decisions.
what is an example of the concerns + challenges of gen ai?
ai-generated deepfake videos can be used for both entertainment + fraud, raising concerns about digital trust.
