• Replies 60 replies
  • Subscribers 105 subscribers
  • Views 9752 views
  • Users 0 members are here
Options
Related
Hints and tips
  • To quote someone's message, first select/highlight the relevant text, and then use the Quote button that appears
  • To "mention" someone, type @ followed by the start of the forum name, and choose from the list; that person will get a notification of being mentioned

Artificial intelligence discussion

Lotus

Hi, I've just completed a short course on artificial intelligence. These are some things I learned:

Artificial intelligence is often thought of as scary robots, e.g. Terminator, but AI is used in computer systems to do things like suggest films or books you might like, based on your previous choices. It uses a range of technologies, from simple rule based systems such as " if this is true (or if this is false), then do this" to a complex 'neural net' that is a simplistic model of how the brain works, and can be trained (machine learning)

As well as being used on social media or shopping sites, AI can identify things such as tumour cells in a medical scan or identify galaxies in photographs or from space. Large language models can generate rich interesting text in any language and translate, restructure or re-phrase text. A related field is image generation, where brand new images are generated by the AI. 

In Soeul, AI mini robots were used in a pilot scheme to show senior citizens how to use technology such as smart phones, to enable them to better participate in society and help with preventing loneliness.

Youve probably heard of self driving cars. One accident that happened was a self driving car hitting a truck that pulled out in front of it at a junction in the city. This was because the AI had been trained to drive on a highway, not city streets, and the image of a truck it had been given was a view of the back of a truck. When it saw the side of the truck as it pulled out, it thought it was a road sign and so attempted to drive under it. The AI had no perception that the space under the truck was lower than the vehicle it was driving.

There are issues around ethics, inclusion and sustainability. For example, if AI is trained using data that is biased, the AI will acquire that bias. Although it can be used for facial recognition, it can sometimes struggle to determine gender and the darker the skin tone, the less accurate are the results. Training AI uses massive amounts of resources.

What are your thoughts?

And would you be comfortable having an AI robot in your home to help with tasks?

Parents

  • Something I've noticed during the last few months is that many researchers have acknowledged that AGI may be still out of reach.

    While powerful, the Chatbots / LLMs which are based of the transformer architecture, seem unlikely to achieve AGI (Artificial General Intelligence), at least by simply scaling the models (was the past strategy.)

    For two main reasons, at least that's what distilled in my mind:

    1. Lack of Generalisation

    2. Lack of Continual Learning

    They still will have most of the issues and gains you write about, but this is an aspect which is discussed in many areas as well. Said in their language, the singularity (when AIs reach AGI, and maybe beyond, to super-intelligence) isn't going to be reached with the current architectures.

    That is my read from the ones I follow most commonly: Ilya Sutskever, Andrew Karpathy, Yann LeCun etc.

    Others have somewhat different view.

    Just thought this tangent could be of interest for some members.

Reply

  • Something I've noticed during the last few months is that many researchers have acknowledged that AGI may be still out of reach.

    While powerful, the Chatbots / LLMs which are based of the transformer architecture, seem unlikely to achieve AGI (Artificial General Intelligence), at least by simply scaling the models (was the past strategy.)

    For two main reasons, at least that's what distilled in my mind:

    1. Lack of Generalisation

    2. Lack of Continual Learning

    They still will have most of the issues and gains you write about, but this is an aspect which is discussed in many areas as well. Said in their language, the singularity (when AIs reach AGI, and maybe beyond, to super-intelligence) isn't going to be reached with the current architectures.

    That is my read from the ones I follow most commonly: Ilya Sutskever, Andrew Karpathy, Yann LeCun etc.

    Others have somewhat different view.

    Just thought this tangent could be of interest for some members.

Children
No Data