MIS 200: Artificial Intelligence

what is Chat GPT?

- a generative AI - a chatbot that can engage in a prolonged conversation with you - a system that interactively searches through and synthesizes massive amounts of information (sorry Google) - a text generator that will respond to your commands to for instance write an essay

why is AI taking off now?

- after years of work with neural networks, the knowledge was developed to create very complex neural networks. this is known as deep learning. - millions of video gamers drove a need for very powerful graphics cards (GPUs). These GPUs turned out to be incredibly good at performing the calculations needed for neural networks

what AI is not?

- an all-knowing, self-aware being that is smarter than people and will enslave humanity (see the terminator, westworld, black mirror, the matrix, etc.) - but we don't know what the future will bring and we have a responsibility to regulate artificial intelligence and its uses - the point where AI is smarter than humanity is called "the singularity" somehow it is always 30 years away, but in the past few years the development of AI has accelerated, so it has likely gotten closer now.

why do AI systems so quickly become abusive and racist?

- because their knowledge is based on how humans behave on-line and everything humanity publishes. look at social media and you will see that a lot of it is abusive, misinformed and racist. some people argue that chatbots are a mirror of humanity. - AI companies try to keep guardrails on AI, preventing it from discussing certain topics.

what are the goods and bads of ChatGPT?

- both spectacular and very mundane - it is spectacular because you can have a conversation with it as if it is a person. it will remember what you told it. - until now all chatbots were very simple. they had almost no memory and could only work sentence by sentence. chatGPT is way more advanced.

what are the limitations of ChatGPT?

- but it makes mistakes and regularly goes off the rails: it has serious limitations - without meaning disrespect to our robotic overlords, ChatGPT is right now little more than a cute interface for everything that is out on the world wide web.

Chat GPT, Bing, Bard, Beyond?

- chat GPT data up to Sept. 2021 - microsoft's bing (on microsoft edge browser): up to data. provides sources as footnotes. - bard. unknown data recency - microsoft copilot will be added to office products - companies and some products are getting their own customized versions. large language models like these need to be trained first on all their documents, then can be used for specific purposes.

what is the grand vision of artificial intelligence?

- computer hardware and software systems that are not as "smart" as humans (or smarter) - so far, this vision has eluded computer programmers and scientists

mini-case: artificial intelligence beats radiologists in reading mammograms - problem?

- high level of information inaccuracy - opportunities from new technology

mini-case: artificial intelligence beats radiologists in reading mammograms - what does it demonstrate?

- how organizational performance can benefit from using technology such as artificial intelligence to facilitate acquisition and application of knowledge - illustrates the ability of AI pattern recognition systems to achieve higher level of accuracy than human radiologists - radiologists are not replaced, but their work is augmented

where do you find AI?

- in siri, in self-driving systems, in the management systems of left, in weather prediction, in diagnostic systems for cars and people, in your washing machine. - in generative systems. generative AI creates music, pictures, virtual worlds, marketing campaigns... - .... and text, for instance in ChatGPT, Bing, Bard

what are the limitations of advanced AI systems?

- it can with great certainty give you information that is completely wrong (this caused a $100 Billion drop in Google's valuation). - it can go very wrong. this is know as hallucinating. examples are a Tesla trying to steer into oncoming traffic and ChatGPT declaring its love for a journalist (telling him his marriage is bad and he should dump his wife)

what will happen to work?

- it will change rapidly. - whatever is repetitive has been getting automated for decades, now more complex tasks will aslo be automated in the long run, but in the short run work will only be augmented: - for instance, a system will make recommendations for a marketing campaign, you as an expert finetune the recommendations. - a professor asks ChatGPT how to teach a topic, the system makes recommendations, the professor polishes them. - in a next round it is likely that the systems will autonomously perform tasks for you, using Robotic Process Automation - Short term: lots of opportunities. - Long term: probably nobody knows. Maybe ask Chat GPT?

mini-case: artificial intelligence beats radiologists in reading mammograms - solutions?

- monitor accuracy and costs - evaluate study results - collect mammograms for system training and testing - train pattern recognition system - compare AI system and human radiologist findings - AI pattern recognition system

what are the two types of AI?

- narrow AI & general AI - all the systems we have so far are examples of narrow AI

what is a narrower, more realistic vision of artificial intelligence?

- systems that take data inputs, process them, and produce outputs (like all software programs) and that can perform specific complex tasks that would be difficult or impossible for humans to perform

a brief history of AI?

- whenerver we automate something that used to bdone only by people it seems a bit magical and we call the system intelligent - this started with systems automatically allocating workers and ordering parts for manufacturing - next came diagnostic systems called expert systems - the phase after this was the development of neural networks.

general AI?

can do any kind of task well. shows real intelligence in all areas. displays the intelligence of countless many very bright people.

what is artificial intelligence?

perceiving, synthesizing, and inferring information-- demonstrated by machines as opposed to intelligence displayed by non-human animals and humans

narrow AI?

performs a specific task

what is an example of neural network example?

the nn is trained. it is given millions of pictures. for every picture it is told it there is a cat in it, or no cat in it. over time it learns to recognize pictures with a cat, even if it has never seen the picture before. it analyzes pictures very differently from how we do.