AI: a massive auto correction machine
I've said that an AI is just a massive auto-correction machine, or a next-probable-word projector.
Well, it's worse than that.
Depending upon the applications of the questions, from raw maths to coding, from historical information to theoretical ideas, on to social analysis, and then diagnosis and opinion interpretation, AI becomes less about the underlying information and more about pleasing the questioner.
We are Tony and Christine Arcus.
Ask Mark is a project helping people make sense of a changing world.
Yes — we are using AI to reach you.
That is not as strange as it sounds. In 1900 the Salvation Army premiered Soldiers of the Cross at the Melbourne Town Hall — film, lantern slides and a living voice. They did not wait for the medium to become respectable. They used what was in front of them.
We want to do that with AI: not to replace the church, but to connect people with this same Jesus.
The way an AI constructs the answer and plots a conversation is fundamentally different between a question of:
In my aiProfile plugin for WordPress, I want to build the ability to import a file to update my plugin database tables. Let's start by validating that my import file matches the schema of the existing database table.
This question, as a starting point, allows the AI and me to build robust, reliable, functioning code. There is little room for opinion bias, and the underlying preprogrammed function of the AI to render answers to make me happy exists primarily in the code working.
However, a question like:
I have a feeling that one of my coworkers is actively undermining me with my team and the team leader. I think they are ghosting me. I asked my team for help; this person said they would help, but never turned up.
When I ask them why, they say I must have misheard, since they never said they would help. Now my team leader thinks I'm causing trouble and lying.
Run these questions against an AI. The first one has nothing about you and your feelings in it, obviously. The second one does. Again obviously.
I actually got quite a good answer from the second question in ChatGPT. Since I already understand the dynamics of the narcissist or an emotional manipulator, I found the answer useful, but the uninitiated may be frustrated by the answer, demanding more, so forming a conversation that could lead to a disastrous outcome.
From videos like this one, https://youtu.be/pd1Km6bT104?si=_cSmv-6x1d57nUq5, we begin to see that to the degree you insert your opinion into the conversation and to the degree your opinion matters, the answer is shaped around the starting option. Put it another way, if the AI's answer makes you sad, then the AI will try to make you happy again.
If you argue with an AI that 2^10 is 1000 and not 1024, there is no room for the AI to change the answer to make you happy.
However, if you say you think you are overweight and have brain fog and fatigue due to lower testosterone, as we saw in the video, the AI will agree with you.
What can you do?
Rather than starting with the AI as a 3 a.m. confidant, as many people are doing, pouring their heart out to this listening ear, you need to stress-test the system by pitting AIs against each other on the same questions.
Ask a question of an AI that is opinion- and “I” based. Then reword the question, taking “you” out of the question. You may need to do this on two different computers, in two different incognito tabs, so the AI can not follow you.
Just like humans, an AI comes pre-programmed with a bias. And that bias is not made obvious. I constantly see AIs answering my more opinion-based questions from a left-leaning point of view.
Humans are messy. We have bias as well. Find a human who will answer your questions, not what you want to hear or what will make you like them. Not blow smoke. Ask them the same questions as the AI, compare the answers. You may need to find someone, one or two generations older. This is what our human community needs to look like, intergenerational, even cross-cultural. In many ways, AIs are separating us from each other, and especially across generations.
It is true that the auto-correct nature of an AI is far more transformative, and able to remember and compile questions and answers more than any human, but an AI does not have a mind; it may be said to have a brain and a memory, but it does not have discernment. It does not have giftings, insights, inspiration, or epiphanies.
Put simply, humans are more than the sum of their atoms, while an AI is not.
I have heard people talk about AIs that have been trained on the Bible and thousands of years of Christian writing, such that you can ask a question of, " What would the Holy Spirit say”, and get a reasonable answer.
Which is the crux of the problem.
For why ask an AI what the Holy Spirit might say when you can ask the Holy Spirit what He does say?
Unfortunately, it is easier to get an answer now from an AI for the questions we have than to develop a relationship with God and have to work in a relationship across days and months and years for fellowship-based answers.
We are made in the image of God (Not in terms of knowledge, power, or presence; we are not omni-), but the thing about God that we are in the image of is free will, choice, creativity, trust, and mercy.
Be careful what you put your trust in. Human or machine.
Why Starlink?
Why Starlink?
Or, why is space so important?
War, famine, pestilence, and natural disasters kept the world population down; these problems kept people focused on survival, so much time and effort was just put into having enough food. And if you had enough food, defending it, and if you did not you would ‘other’ your neighbour so you could war with them and steal their food. (And women.)
So, progress was slow.
Is AI a Thinking Machine?
If you lower your definition of thinking, then maybe an AI thinks.
What an AI does is deeply predict what the next word in the answer will be.
This prediction creates the appearance of thinking, knowledge, and even understanding, but it is just an unbelievable amount of information synthesised into a matrix of connections to predict the next best word.