AI

What is AI?

I was watching a TV news program recently where they were asking members of the public if they used AI. Most of the answers were yes. Then they asked if it was dangerous. Predictably the answers varied depending on what the person thought AI was and what they were using it for. So like any other powerful tools you can not say if it is dangerous until you know how it is being used and how it could be used if not governed.

AI is when all human intervention is taken out of a process. If the consequence of that process could be life changing or has life or death implications then we have made AI dangerous.

For example imagine we had an AI system that had access to everyones medical records in the UK. Each medical record contains symptoms medicines etc. A large historical database is created of many patients records. A new patient is entered into the database displaying a set of symptoms. By comparing these symptoms with the historical database you could make a prediction about what is wrong with this patient. This information could be invaluable to a doctor in speeding up the diagnosis of the patient and treatment. A useful tool. Not really AI.

If we take it a step further and allow the AI system to takover the role of the doctor and make a diagnosis based on the historical data and other information that it has available to it, and then prescribe the treatment and medicine without human intervention, then we do have an AI system. Artificial Intelligence means taking the whole decision process out of human hands. Not something that can ever be safe however much data you have unless you can be 100% sure the data collected is 100% correct and is representative of enough (how much is enough?) of previous cases which it never will be.

So is AI a security risk?

AI is a security and safety risk because the current developers of AI appear to be and it has access to the Internet.

AI needs large amount of data. In the example I have given above the data required exists in a relatively view places which can be identified. Testing the system clearly does not require breaking in to other systems across the Internet in the way that the current systems in development are doing, so one has to ask what purpose are these systems being devloped for and why are they trying to break through the security of other systems?