Eerke Boiten, Professor of Cyber Security at De Montfort University Leicester, explains his belief that current AI should not be used for serious applications.
The right tool for the right job. It’s not intelligent, it is just trained. It all boils down to stochastic.
And then there is the ecological aspect…
Or sometimes the moral aspect, if it is used to manage someone’s “fate” in application processing. And it might be trained to be racist or misogynist if you use the wrong training data.
The moral aspect is resolved if you approach building human systems correctly too.
There is a person or an organization making a decision. They may use an “AI”, they may use Tarot cards, they may use the applicant’s f*ckability from photos. But they are somehow responsible for that decision and it is judged by some technical, non-subjective criteria afterwards.
That’s how these things are done properly. If a human system is not designed correctly, then it really doesn’t matter which particular technology or social situation will expose that.
But I might have too high expectations of humanity.
Accountability of a human decision maker is the way to go. Agreed.
I see the danger when the accountant’s job asks for high throughput which enforces fast decision making and the tool (llm) offers fast and easy decisions. What is the accountant going to do, if (s)he just sees cases instead of people and fates?
If consequence for a mistake follows regardless, then it doesn’t matter.
Or if you mean the person checking others - one can make a few levels of it. One can have checkers interested in different outcomes, like in criminal justice (… it’s supposed to be).
An LLM cannot think like you and I. it’s not able to solve entirely new problems. And it doesn’t have a concept of the world - it paints hands without knowing what a hand does.
It is a system which learns the rules of something by means of reinforcement learning to tune the coefficients of its heap of linear equations. It is better than a human in its area. I guess it can be good for tedious, repetitive tasks. Nevertheless it is just a huge coefficient matrix.
But it can only reproduce what is in the training data - you need lots of already solved examples in the training data. It doesn’t work for entirely new problems.
(that’s also the reason, why LLMs don’t give good answers to questions about specialized niche topics. When there are just one or two studies, there just isn’t enough training data for the LLM.)
Right? I see comments all the time about it just being glorified pattern recognition. Well…thats what humans do as well. We recognize patterns and then predict the most likely outcome.
How? You’re focusing on one thing a human does and using it to point to how human like LLMs are, while ignoring everything else humans do. You’re missing the forest for the trees.
I didn’t say that at all. What I said was LLMs solve problems just like a human does. Pattern recognition. Then I asked you to provide an example of one thing a human does that doesnt boil down to pattern recognition. The words we speak and type are patterns. The decisions we make are based on patterns we learned in the past. Thats really all I meant by it.
LLMs don’t solve problems. That’s the point being made here. Many other algorithms do indeed solve issues, but those are very niche, as the alogos were explicitly designed for those situations.
While yes, humans excel at pattern recognition, sometimes to the point of it being a problem, there are many things we do that have nothing to do with patterns beyond the fact that they are tangentially involved. Emotions for instance don’t inherently follow patterns. They can, but they aren’t directly tied. Exploration also doesn’t come from pattern recognition.
If you need examples of why people flat out say LLMs aren’t solving problems, look at the recent “how many r’s in strawberry” which has admittedly been “fixed” in many models.
At the end of the day LLMs take in historical data and use it to predict what comes next. Just like humans do. But I guess we can disagree and leave it at that.
The right tool for the right job. It’s not intelligent, it is just trained. It all boils down to stochastic.
And then there is the ecological aspect…
Or sometimes the moral aspect, if it is used to manage someone’s “fate” in application processing. And it might be trained to be racist or misogynist if you use the wrong training data.
Yeah. Considering the obscene resources needed for ChatGPT and the others, I don’t think the niche use cases where they shine makes it worth it.
The moral aspect is resolved if you approach building human systems correctly too.
There is a person or an organization making a decision. They may use an “AI”, they may use Tarot cards, they may use the applicant’s f*ckability from photos. But they are somehow responsible for that decision and it is judged by some technical, non-subjective criteria afterwards.
That’s how these things are done properly. If a human system is not designed correctly, then it really doesn’t matter which particular technology or social situation will expose that.
But I might have too high expectations of humanity.
Accountability of a human decision maker is the way to go. Agreed.
I see the danger when the accountant’s job asks for high throughput which enforces fast decision making and the tool (llm) offers fast and easy decisions. What is the accountant going to do, if (s)he just sees cases instead of people and fates?
If consequence for a mistake follows regardless, then it doesn’t matter.
Or if you mean the person checking others - one can make a few levels of it. One can have checkers interested in different outcomes, like in criminal justice (… it’s supposed to be).
The same could be said about every human being…
An LLM cannot think like you and I. it’s not able to solve entirely new problems. And it doesn’t have a concept of the world - it paints hands without knowing what a hand does.
It is a system which learns the rules of something by means of reinforcement learning to tune the coefficients of its heap of linear equations. It is better than a human in its area. I guess it can be good for tedious, repetitive tasks. Nevertheless it is just a huge coefficient matrix.
But it can only reproduce what is in the training data - you need lots of already solved examples in the training data. It doesn’t work for entirely new problems.
(that’s also the reason, why LLMs don’t give good answers to questions about specialized niche topics. When there are just one or two studies, there just isn’t enough training data for the LLM.)
This was already disproven a year ago.
They replaced the training data with an evaluator. (which rates the LLMs output for training?) Interesting, thanks.
Edit: this reminds me of the self evolving (virtual) robot problem, a robot which is rated by an external moderator and improves over time. I.e.: https://www.sciencedirect.com/science/article/pii/S0925231221003982
Right? I see comments all the time about it just being glorified pattern recognition. Well…thats what humans do as well. We recognize patterns and then predict the most likely outcome.
That is one part of many that a human brain does. This is like trying to say the color red is a rainbow, because the rainbow has red in it.
Can you expand on that?
How? You’re focusing on one thing a human does and using it to point to how human like LLMs are, while ignoring everything else humans do. You’re missing the forest for the trees.
I didn’t say that at all. What I said was LLMs solve problems just like a human does. Pattern recognition. Then I asked you to provide an example of one thing a human does that doesnt boil down to pattern recognition. The words we speak and type are patterns. The decisions we make are based on patterns we learned in the past. Thats really all I meant by it.
LLMs don’t solve problems. That’s the point being made here. Many other algorithms do indeed solve issues, but those are very niche, as the alogos were explicitly designed for those situations.
While yes, humans excel at pattern recognition, sometimes to the point of it being a problem, there are many things we do that have nothing to do with patterns beyond the fact that they are tangentially involved. Emotions for instance don’t inherently follow patterns. They can, but they aren’t directly tied. Exploration also doesn’t come from pattern recognition.
If you need examples of why people flat out say LLMs aren’t solving problems, look at the recent “how many r’s in strawberry” which has admittedly been “fixed” in many models.
At the end of the day LLMs take in historical data and use it to predict what comes next. Just like humans do. But I guess we can disagree and leave it at that.