-- Matt Mahoney, [email protected]

On Sat, Nov 15, 2025, 10:13 PM John Rose via AGI <[email protected]>
wrote:

> On Saturday, November 15, 2025, at 7:39 PM, Matt Mahoney wrote:
>
> Here are my 3 definitions of consciousness:
> 1. The mental state of awareness, able to form memories that depend on
> input (to distinguish from remembering dreams).
> 2. The difference between a human and a philosophical zombie.
> 3. The property of deserving to be protected from suffering.
>
>
> I posit that conscious compressors can compress better than less conscious
> compressors. Proving that and the particular circumstances is possible I
> believe. IOW consciousness can be used to enhance compression in at least
> some circumstances, in various ways. For example in some classes of
> multiagent systems compressors. Not sure about a lossless text compressor
> but when you think about it the original text may be coming from conscious
> human agents. But the text compressor might be large depending on the
> conscious integration. Speculation though... could be all bunk.
>

If you mean type 1 consciousness measured as the learning rate in bits per
second, then it's true that more "conscious" compressors will compress
faster and better. Humans learn at 10 bits per second short term and 1 bit
per second long term. The Hutter prize top compressors learn at about
10,000 bits per second on a laptop. Simple compressors like zip learn at
100M bps on small files but forget beyond a 32 KB window, which makes the
rate again 10 Kbps on enwik9.

Lossless compression requires deterministic computation so that the exact
sequence of predictions is repeated during decompression. Neurons are
noisy, which is fine for making approximate predictions but not for
compression, even if the predictions are more accurate.


------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T7ff992c51cca9e36-M914661aa38a124e11a98b9a3
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to