Hacker News Reader

Comment by loceng | Hacker News Reader

"The tool, called mediKanren, scanned millions of biomedical abstracts hunting for relationships between existing compounds and the gene involved in the disease."

Is this AI or someone put the effort into compiling a database and added automatic search function?

View on HN

Isn't that all AI is in its current state? Just like computer chess. It knows the moves, and can calculate possibles and iterate through them way faster than a human. It's not like it is thinking for itself saying, what happens if I try this. The devs of the AI point it in a direction, and it sorts through things much more deeply and faster than humans.

All of that is great when we get useable information or even when it exhausts all options to no positive result. It's just another tool to find out answers. The term AI is used waaaay too broadly.

That was the state of AI in the 90's. AlphaGo Zero and AlphaZero have shown that it does not need to search exhaustively when it can search intelligently using reinforcement learning. The 'direction' is something the system learns and is not given by the devs.

Take out 'exhaustively', though, and everything parent comment said is correct. AI is just (big data + statistical inference) implemented in massively parallel ways using linear algebra.

"just" is always a weird modifier in these kinds of conversations. Most things can be reduced to "just" some basic operation(s). What is amazing is the complexity that emerges from those simple operations at scale.

Granted that emergent behaviors are important. The reason for injecting "just" is to counter the AI hype. Imagine if the headline read "With a nudge from linear algebra, ketamine emerges as a potential rare disease treatment". Dropping the word AI in there, in addition to boosting the articles SEO, will suggest to those with a casual interest that his must be a big deal, because "I've heard so much about how AI is solving <insert big problem here>"

Well, if you dive into data science and ML, you'll see it is much more than just a bit of linear algebra. Some topics (and as a data engineer, I am not an expert):

- Distributed computing

- Feature selection

- Budgeting / accounting of experimental design (these TPU clusters are not cheap)

- ML architecture

- Involvement of domain experts (multidisciplinary teams)

- Storage

The whole list is much longer, but just some topics to think about. For me the term "AI", although overly broad, come to mean the engineering and organisation needed to pull these kind of projects.

Reminds me of minikanren and I think this may be related

Yeah, I had the same thought. Also, these are two things that absolutely _should_ be together haha...

indeed they do share Will Byrd as their author

it indeed is based on MiniKanren.

This is logic programming. So AI in the symbolic sense. And much better suited to medical 'AI' in the current state of medical information retrieval capability than most of the statistical crap that gets all the hype nowadays.

I worked at a biomedical informatics shop that did this very thing by AI/ML 10 years ago. Essentially: automated metanalysis.

I don't think that's automated meta analysis. It's meta research, which is not the same thing. Take a look at the MediKanren papers. This is logic programming and it's not based on statistics, but first order logic.

In fact, it's a kind of expert system. But in the 80's, expert systems didn't have access to that:

https://skr3.nlm.nih.gov/SemMedDB/

Reply on news.ycombinator.com