Nobel prize 2024 - magic just got less wriggle room

this clustering is not evident: finding one viable protein does not make it any easier to find further proteins; viable proteins are not generally "nearby" one another.

Detecting clusters in data is exactly what AI is all about otherwise those neural networks wouldn’t work.

[Laird]

Clustering is exactly what AI is all about otherwise it wouldn’t work.

I don't think that that follows, at least in a relevant way in this context. It seems to me to be a bold but unsubstantiated assertion.

The fact that AI can predict the structure of a protein from a given sequence of amino acids doesn't entail any clustering (in the sense of "proximity" in a semi-random walk) of the amino acid sequences that do produce proteins, let alone those that produce life-sustaining proteins.

[Laird]

I see that you edited your post before I posted, but it isn't a substantive enough edit to change my response.

I don't think that that follows, at least in a relevant way in this context. It seems to me to be a bold but unsubstantiated assertion.

The fact that AI can predict the structure of a protein from a given sequence of amino acids doesn't entail any clustering (in the sense of "proximity" in a semi-random walk) of the amino acid sequences that do produce proteins, let alone those that produce life-sustaining proteins.

I agree. It only shows that the clusters exist, which was completely unknown before. It can now be hypothesized that they are in close proximity to each other during the 'semi-random walk.' A simulation to prove or disprove this hypothesis could be interesting.
.

[Laird]

I agree. It only shows that the clusters exist, which was completely unknown before. It can now be hypothesized that they are in close proximity to each other during the 'semi-random walk.' A simulation to prove or disprove this hypothesis could be interesting..

This is again so confused that it's actually disappointing.

Clustering as I introduced the term is the close proximity of proteins to one another in the context of a semi-random walk. If clustering exists, then by definition proteins are in close proximity to each other during the semi-random walk, and "hypothesising" such a thing is redundant.

To say as you do then that this AI result "shows that the clusters exist" is to say that this AI result entails that proteins are in close proximity to one another in the context of a semi-random walk, and yet you began your response by agreeing with me that it does not.

Seriously, @sbu, at this point you should just accept that you misfired in the opening post in this thread. The Nobel prize was in any case worth a thread, so there's a saving grace.

This is again so confused that it's actually disappointing.

Clustering as I introduced the term is the close proximity of proteins to one another in the context of a semi-random walk. If clustering exists, then by definition proteins are in close proximity to each other during the semi-random walk, and "hypothesising" such a thing is redundant.

To say as you do then that this AI result "shows that the clusters exist" is to say that this AI result entails that proteins are in close proximity to one another in the context of a semi-random walk, and yet you began your response by agreeing with me that it does not.

Seriously, @sbu, at this point you should just accept that you misfired in the opening post in this thread. The Nobel prize was in any case worth a thread, so there's a saving grace.

No, that is not what a cluster means in data analysis. A cluster is, by definition, the partition of a dataset X = {x1, x2, …, xn} into k distinct clusters C1, …, Ck where each x belongs to one cluster. It doesn’t make any sense to make claims like ‘oh, this is not what I mean by cluster. Clusters in data analysis is well defined.

AlphaFold's success suggests that there may be fewer degrees of freedom in protein folding space than previously thought, which directly contradicts the narrative being promoted by the Discovery Institute.

[Laird]

that is not what a cluster means in data analysis

Then you took a term that I had introduced and explicitly defined and continued to use it but with a different definition, without clarifying that that's what you were doing.

Would you agree that that is not a helpful thing to do in a conversation?

AlphaFold's success suggests that there may be fewer degrees of freedom in protein folding space than previously thought, which

...even if true (dubious), has absolutely nothing to do with @nbtruthman's argument.

directly contradicts the narrative being promoted by the Discovery Institute.

Not that I can see.

[Laird]

In any case, my definition of "clustering" is compatible with the one you cite: those proteins which are in close proximity to one another (in the context of a semi-random walk) consist in a "partition" of the data as you describe it.

What this AI result does not seem to entail is that these (particular) partitions exist. It seems that you don't agree with that after all. Why don't you explain how you think this is entailed then?

In any case, my definition of "clustering" is compatible with the one you cite: those proteins which are in close proximity to one another (in the context of a semi-random walk) consist in a "partition" of the data as you describe it.

What this AI result does not seem to entail is that these (particular) partitions exist. It seems that you don't agree with that after all. Why don't you explain how you think this is entailed then?

I thought I used the term “cluster” before you did but I may be wrong.

Anyway it’s now proven that “clusters” exists in protein space. Forget about proximity - If data has clusters it simply means that instead of being represented by n data points the same data can be represented by k data points with k< n =>protein space is (overwhelmingly) smaller than the narrative by DI.

[Laird]

Forget about proximity

OK, but then we also forget about relevance to the argument, because the argument relies on proximity.

If data has clusters it simply means that instead of being represented by n data points the same data can be represented by k data points with k< n

That doesn't sound right, unless n - k data points (protein foldings) are identical with some other data point, which I don't think you'd suggest is the case here. [Edit: OK, but being more charitable, you probably don't intend that the representations are identical. I think I see what you mean.]

protein space is (overwhelmingly) smaller than the narrative by DI

You seem to be putting up a straw man here. Maybe you can cite something relevant from the DI and explain how this AI result affects it.