Understanding AGI

In the very first post of this series, we tried to define AGI and we carefully used the word “hypothetical” in connection with “intelligence”.

Intelligence is defined as a concept that refers to the ability to learn, understand, and apply knowledge and skills. It encompasses various cognitive processes such as reasoning, problem-solving, memory, perception, and learning. Intelligence can be broadly categorized into different types.

1. Cognitive Intelligence : This involves logical reasoning, problem-solving, and the ability to process information efficiently. It is often measured by IQ tests.
2. Emotional Intelligence : This refers to the ability to recognize, understand, and manage one’s own emotions, as well as respond appropriately to the emotions of others. It includes skills like empathy, self-awareness, and social skills.
3. Practical Intelligence : This is the ability to apply knowledge and skills in real-world situations. It involves adaptability, common sense, and the ability to navigate everyday challenges.
4. Creative Intelligence : This involves the ability to think innovatively, generate new ideas, and find unique solutions to problems. It is often associated with artistic and inventive abilities.
5. Social Intelligence : This refers to the ability to understand and navigate social situations effectively. It includes skills like communication, cooperation, and understanding social norms.

Intelligence is not a single, fixed trait but a combination of various abilities that can be developed and improved over time. Different theories and models of intelligence have been proposed, each highlighting different aspects of what it means to be intelligent.

The conclusion here is that intelligence is not one thing but the combination of serveral things that come together.

Now having “reasoning” LLMs (LRMs), does it mean we have mastered the step to cognitive Intelligence?

I don’t think so. Even though we label the resoning process with “thinking”, doesn’t mean the model is able of thinking. LLMs are next word guessing machines and the “resoning” process is just an internal extension of your prompt to enhance the results by adding more “reasonable” information into the context window the LLM can use to produce the final answer. The same can be achieved with a prompting technique called Chain of Thought (COT) which is the manual way of doing it. Reasoning just improves the time to answer by automating this process. So this is not Cognitive Intelligence.

Talking about Emotional Intelligence… let me ask you a question: Do you think a machine will have emotions? Ever?

I doubt it, because human beings and machines are two fundamentally different concepts - and being emotional means not being rational and there is nothing more rational than a machine interpreting 0 and 1s. This means in turn a machine will judge black and white, no grayzones here. You are guilty of theft or not. There is no grayzone of understanding you stole the bread, because you are starving.

This might be a good or a bad thing. I assume it depends on perception. But our laws imply “in dubio pro reo” which means if there is any doubt about the crime we need to judge in favour of the accused. A wise concept of our law makers to protect the innocent.

But Practical Intelligence is achieved, right?

Letting your LLM use tools all over is so practical, isn’t it? It can do git commits for you, check the weather or your travel itenary. NICE! Indeed.

But does it show intelligence? Or is it rather a Rube-Goldberg-Machine where the rules to “use this tool when this is asked for” is dynamically processed by the LLM.

When Edmund Cartwright invented the mechanical loom, I can imagine people wondered how it works and being an intelligent machine to solve complex mechanical tasks only humans were capable before. But have we seen any intelligent proof of the beautiful mechanical machines that made our lives so much easier in the last 200 years? No, we don’t.

And while the LLM might be loom of the computer age, sparking a comparable revolution (and unemployment) as the industrial revolution.

It is just that.

Nothing else.

Creative Intelligence. That’s achieved!

Or is it? You all see these beautiful pictures generated by AI and recently videos and characters that are flawless and almost indistinguishable from real videos and pictures. This must be creativity, right?

Let me ask you a question: Is a photo copier a machine that shows creative intelligence?

It can produce art like images as the great masters did, in almost real time. It’s just, it only is a copy of a piece that someone with real creative intelligence once created.

GenAI is no different. It just copies from hundreds and thousands of pieces and creates a unique composition each time you ask for it. It is not a unique, well thought re-imaging of your prompt, but rather a mathematical way of arraging elements and colors, while the mood and impression while viewing it comes from you. It is your implication that make it feel like art, but it isn’t.

Social Intelligence?

In machines? I know a lot of people use LLMs lately as advisor. Medical Advisor, Mental Advisor, Relationship Advisor, … choose or add one you have used before.

I encourage you to not do it. At all!

As with creative intelligence discussed above, AI today is just a better photo copier. No Intelligence. It can replicate what it has been trained with. All the Self Help Books, all the publications, written by people, who not only understood their profession, but also are a social human being, like you.

Don’t fall for the trap that a natural feeling conversation is in your best intend. It is your secret desire to be heard. But your thoughts, your wishes and fears are heared by a machine that has not even the fundamental understanding of being social. Build a network of human peers, talk, learn and listen to their social experiences and think about what it means for you.

What does it mean?

Your emotional intelligence is key in tricking yourself into thinking that parts of intelligence are already achieved in todays AI. The truth however is, we are far, far away from it. AI companies telling you it is just around the corner to justify their humongous valuations and insane investments into datacenters and energy consumption.

The only real intelligence on this planet, for now, is you. And by the way you are magnificantly energy efficient.

So maybe we need another way of doing things when we really want to achieve AGI…

_