This blog is the first of a series by guest contributor Bruce McCormack, Vice President at EUROGI, focusing on topic artificial intelligence (AI). The series will provide some clarity on the widely used term AI and explore the opportunities and threats it may hold. This first post explains what is AI and what frameworks we can apply to use it.
The first most straightforward way is to begin is provide some definitions. Of course (!) the best way to start looking for definitions is to ask an AI chatbot "what is AI?". The answer from Microsoft’s latest version of the AI enabled Bing search engine is “the ability of a computer or a machine to perform tasks that are usually done by humans who have intelligence.”
After reviewing a number of definitions provided in legislation and by academics I come to my own simple definition, namely, “the ability of a machine to mimic the ways in which the human brain works.”
The immediate questions which arise from these definitions are … what is human intelligence, and how does the human brain work. Very many PhDs and many, many academic papers have grappled with these separate but related issues … the conclusion being … ‘well, ahumm, err .. we know a lot about these matters, but we still really do not know at a fundamental level just about anything about them’. Just to make the point - we can not explain at all how process at the quantum level, the fundamental basis for just about everything in the universe, including ourselves, gives rise to intelligence or how our brains work.
So, what you may ask is the way forward? A useful approach is to review what AIs can do and to see to what extent its abilities accord with what we intuitively know human brains can do. A simple framework based on inputs, outputs and the necessary processing in the middle provides a useful starting point.
These may be a whole variety of types of information, which are used for AI model training and testing, much of it coming from the internet, which represents a substantial proportion of the whole of human knowledge. The types of inputs from the internet include:
- Text: such as all of Shakespeare’s works, First World War soldiers’ diaries from the trenches, Einstein’s papers on relativity, the Bible and Quran, and much more;
- Audio: including Beatles music, Martin Luther King’s ‘I have a dream’ speech, the sounds of dolphins communicating under water, etc;
- Imagery: such as Salvador Dali paintings, satellite imagery, family ‘happy snaps’, as well as ‘ordinary’ peoples’ life experiences (in receny years via social media in particular);
- Video: YouTube (maybe 2 billion videos), Tik Tok videos, the first humans walking on the moon, David Attenborough nature documentaries, Hitler giving speeches, dogs being trained by their masters, to name a very few;
- Computer code: in a whole variety of languages;
- Mathematical equations and statistical tests
Viewed from a broad perspective the outputs may be viewed in different ways, one way being as set out below:
Methods of Processing
- Content: such as original articles, movies, music, computer code, paintings, videos, graphs, speech, maths calculations;
- Predictions: such as share prices at some future date, where armed conflicts are likely to arise, how proteins will fold in different conditions, and very many more, the weather tomorrow in my town;
- Recommendations: on any number of issues, such as what should I wear today taking account of the likely weather conditions, what Netflix movie should I watch, what is the best vegan meals for me to eat, etc.;
- Classifications: of customers into market segments, emails into spam and non-spam, and many other types of classifications;
- Explanations: across an enormous variety of subjects; why do some people become criminals, why has a particular area of a country performed poorly from an economic perspective, how does the human brain work (!), etc..
Across the spectrum of AI there are various ways in which they process the data which is inputted. One way of viewing the processes used is as set out below:
When I ‘step back’ to consider what types of information AI can access, how they can process it, and the types of outputs which they can provide I am, to use a colloquial word …. ‘gobsmacked .. totally gobsmacked’. They are so very human-like. To temper this feeling of wonder and awe I need to remind myself that currently not all AIs can access all of the types of information, process it in all the ways identified, or give all the types of outputs listed. Humans, using the grey spongy material in our heads (comprising 60% fat with the other 40% made up of water, protein, carbohydrates and salt) can, or at least some humans can, do all of these things.
- Optimisation: where some quantify is either minimised (e.g. time in a logistics supply process), or maximised, such as getting the best set of arguments which a lawyer could make in court when trying to win a case;
- Learning: (also referred to as Machine Learning), where some elements of the model are trained through multiple iterations through the input information so as to be better able to ‘understand’ patterns or other features of the information. Various types of learning exist, including for example trial-and-error learning which for example is used to build new molecules or play video games;
- Reasoning: where based on identified patterns, rules, inferences, facts or beliefs extrapolation are made to new situations or problems. Various forms of reasoning can be used in AIs; deductive (a conclusion is reached by the application of logic); inductive (where a general principle is derived from data); or abductive (drawing conclusions based on limited or ambiguous data);
- Generation: where the AI is creating new data or content that does not, or has not ever existed before, such as new short stories, poems, music, solutions to maths problems, or art works.
Score, 1 to humans, maybe 0.6 to machines processing 1s and 0s, which in their Central Processing Units are made mainly from silicon, copper, plastic, gold, ceramics, and some other elements.
But … researchers and companies around the world are striving both themselves, and also with the aid of AIs, to write the algorithms needed to do this. Creating what is termed 'Artificial General Intelligence (AGI)' is the aim of these endeavours, where AIs could do any intellectual task that humans can do, much faster and more accurately, and would have self-awareness. I find this both exciting, and very, very frightening.
My next blog will focus on AGI, the beneficial opportunities which it would create, and the threats to humans as a species, and to the planet.
About the author
Bruce McCormack, Vice President of Irish Organisation for Geographic Information (IRLOGI)
Bruce is a professional town and regional planner based in Ireland who is also involved in geospatial matters, being the Vice President of EUROGI (European Umbrella Organisation for Geographic Information), the European Umbrella Organisation for Geographic Information, and Vice President of IRLOGI, the Irish Organisation for Geographic Information. He leads a EUROGI group of AI experts which is preparing a paper on ‘AI and Geospatial.’
Posted: 06/07/2023 11:16:41