I was away this week at the Local Government New Zealand Annual Conference, which was just brilliant. It offered an excellent range of speakers on a broad range of topics and a chance to hear some different points of view.

One of our speakers was Frances Valintine, an education futurist. She spoke on a number of different things but the one that struck me the hardest was where we are heading with AI. 

It really was chilling the mess we could be sleepwalking into.

Here's some of the high points.

Chat GPT4 was programmed in English and designed to be used in English. One day someone realised that it could operate in French. They dug a bit deeper and found that it could operate, or "speak" in 95 different languages. Its operators hadn't asked or told it to do it, and they don't know how, or why, it has. It just went ahead and did it.

Or how about this. You know the grids you see when a programme is trying to test if you are a bot, where you need to tick the boxes that have a bridge or a car or whatever in the photo? Those are designed to stop computers pretending to be humans from getting past whatever drafting gate we don't want computers to get past. ChatGPT4 was assigned the task by someone to get around the roadblock. It went on-line and asked for someone human to help it with a problem. Dutifully a helpful person said "sure, what do you want". Chat GPT4 then asked the person to identify which boxes had bridges. The person asked if they were communicating with a bot and ChatGPT4 said "no, I am a visually impaired human". In other words, it lied to get around our feeble defences. The human apologised and told the computer how to do it.

One last one to make your day. ChatGPT4 apparently has the inherent intelligence of Einstein. GPT5 which is currently being worked on will have the equivalent intelligence of all of humanity.

It'll be OK though, because we can see the risk right and will stop developing it or any other super AI, given what GPT4 is already capable of?

Well, not so much. When Apple recently announced it was developing its own AI system to compete with GPT4, its share value went up $71B. That's right, $71 billion dollars.  

The human desire to make money today no matter what the risk to tomorrow is well documented and can be seen in everything from Three Waters to Climate Change, so why should this be any different?

The fundamental problem is, we might just be too stupid to save ourselves from our own intelligence.