Comment by a_bonobo | Hacker News Reader

Comment by a_bonobo | Hacker News Reader

Parent Full thread

This is the code used in the paper:

https://gitlab.com/machine-biology-group-public/pancleave

>This package implements a scikit-learn-based random forest classifier to predict the location of proteoylytic cleavage sites in amino acid sequences. The panCleave model is trained and tested on all human protease substrates in the MEROPS Peptidase Database as of June 2020. This pan-protease approach is designed to facilitate protease-agnostic cleavage site recognition and proteome-scale searches. When presented with an 8-residue input, panCleave returns a binary classification indicating that the sequence is predicted to be a cleavage site or non-cleavage site. Additionally, panCleave returns the estimated probability of class membership. Through probability reporting, this classifier allows the user to filter by probability threshold, e.g. to bias toward predictions of high probability.

In the face of current hype around LLMs and 'fear of AI', calling a Random Forest Classifier 'AI' is a bit... far

The AI term has been coopted by Hollywood notions of synthetic human-level intelligence but it really is just referring to the academic discipline which Random Forests definitely falls under and has for decades.

We've also been using the term Machine Learning for a long time to try to avoid this. Using "AI" where "ML" would suffice is just trying to cash in on recent hype for clicks.

Still an interesting result of course.

My understanding is "ML" came into existence after the term "AI" became an untouchable and disrespectful dirty word in academia after a previous hype cycle went bust. Statisticians wouldn't be caught dead working on "AI". See also xkcd purity scale (435).

It looks like machine learning took over in the 2010s after a long convergence starting in the 2000s.

https://books.google.com/ngrams/graph?content=artificial+int...

This is from published books, so probably less affected by popular narratives. For that, we have Google Trends!

It looks like they trade places repeatedly when constrained to news searches: https://trends.google.com/trends/explore?cat=5&date=all_2008...

Not really though it may seem that way. Not all AI is ML. There are a ton of things within AI such as planning algorithms or Expert Systems. Again those are under the academic discipline of AI but not in the subfield of ML.

It's also a refinement on the term. AI can include a lot while ML is much more specific.

Right. ML is less likely to arouse emotions in either direction.

Yes I agree, but OP's link is not aimed at academics, it's aimed at the general public. And you can't expect the general public to know that distinction: they see 'AI', they think 'ChatGPT', not Random Forests

AI has been discussed in public for years and was attached to huge numbers of things well before chatgpt. Chatgpt only came out less than a year ago.

A random forrest classifier is a classic example of AI, quite popular before hardware caught up with the computational needs of neural networks.

Random forests of significant size also suffer some of the same inexplicability problems that neural networks suffer from, so it makes even more sense to make the comparison.

Every algorithm is now called AI, unfortunately. 'AI' is being used 'exponentially'.

Random forests are absolutely AI as the term has been used for a very long time in academia and in public.

I fear this kind of retconning will become more commonplace as new entrants to the space who are unfamiliar with the history of the field put words in the mouths of academics that academics themselves would vociferously reject.

Academics have long pushed back against the labeling of everything as "AI", and only the most public-facing ones with more than a buck to make on the hype treadmill ever stoop so low.

In academia, "AI" is usually reserved as a synonym of "AGI" in large part because of the fuzziness and lack of consensus around a good definition of "intelligence". The term is simply not applied to actual outputs of research because there's no way to justify calling anything "intelligence".

As the saying goes, what's the difference between ML and AI? ML is what the PhDs call it, AI is what the MBAs call it.

> In academia, "AI" is usually reserved as a synonym of "AGI"

Now this is retconning. I did my uni course called AI almost twenty years ago. The same processes had been called that for a long time.

Under current usage of the term, I think the bimetallic switch in my office thermostat qualifies as an AI.

Agree completely, particularly with the smirk I can see in the final word! As an academic, "AI" means "fund me". It's the new "big data". And "exponential" is the new "rising" (or, sometimes, "falling").

At a job where the company specialized in mathematical optimizations (think operations research, logistic planning, timetable optimization, etc.) we jokingly called the solvers with heuristics "A.I. technology, short for Advanced If technology". Some customers got the joke, some didn't, but they were generally content with the results :)

Now the question is, could this be done for stuff like marine immune genomes?

> In the face of current hype around LLMs and 'fear of AI', calling a Random Forest Classifier 'AI' is a bit... far

Just because the state of art evolved doesn't mean we have to erase history of the field. This is a ML algorithm so calling it AI is perfectly in line.

The fear you mention is built on the lack of understanding of what ML is. Showing that some AI has "dumb" yet useful implementations can help show the limits of this category of technology.

Lol the wikipedia article of RFs is literally quoted to be "Part of a series on Machine learning and data mining" yet you have MLDevSecBullshitOPs-type experts here schooling you on what consists AI or not.

https://en.wikipedia.org/wiki/Random_forest

Reply on news.ycombinator.com